y Bb f Vga Aaaaaaaaa

--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
A BSRIA White Paper
ARTIFICIAL INTELLIGENCE
IN BUILDINGS
THE SILENT REVOLUTION
Henry Lawson
WP 14/2021
Copyright BSRIA Limited

Provided by Accuris Licensee=LEEDS BECKETT UNIVERSITY/7100928001, User=Leyton Diaz, Carlos
No reproduction or networking permitted without license from Accuris Not for Resale, 07/09/2024 03:36:51 MDT
ARTIFICIAL INTELLIGENCE IN BUILDINGS
THE SILENT REVOLUTION
Henry Lawson
Date: June 2021
Issued by: BSRIA Limited

Old Bracknell Lane West, Bracknell,
Berkshire RG12 7AH UK
T: +44 (0)1344 465600 | F: +44 (0)1344 465626
E: bsria@bsria.co.uk | W: www.bsria.com/uk/
ISBN: 978-0-86022-786-1
Feedback on BSRIA publications is always gratefully

received. Please email any comments and suggestions
to information@bsria.co.uk
DISCLAIMER
BSRIA is the owner or the licensee of all intellectual property rights in this
publication, and in the content published in it.
All risk associated with the use of material from this publication is assumed
entirely by the user.
Full terms and conditions can be found at www.bsria.com/uk/terms/
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
2|
Provided by Accuris
© BSRIA Licensee=LEEDS BECKETT UNIVERSITY/7100928001, User=Leyton Diaz, Carlos
CONTENTS
1 INTRODUCTION......................................................................................................................... 4
2 SCOPE ....................................................................................................................................... 5
3 WHAT IS ARTIFICIAL INTELLIGENCE?................................................................................... 6
4 WEAK AND STRONG ARTIFICIAL INTELLIGENCE............................................................... 9
5 WHEN IS A BUILDING SMART OR INTELLIGENT?.............................................................. 13
6 THE KEY AIMS OF BUILDINGS.............................................................................................. 16
7 REGULATIONS AND STANDARDS SET THE SCENE FOR BUILDINGS........................... 17
8 PROFIT ..................................................................................................................................... 20
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
9 ENERGY EFFICIENCY............................................................................................................. 21
10 ON-SITE RENEWABLE ENERGY........................................................................................... 22
11 CONVENIENCE........................................................................................................................ 23
12 OPERATIONAL AND MAINTENANCE EFFICIENCY............................................................ 25
13 EASE OF CONSTRUCTION.................................................................................................... 26
14 COMFORT AND WELLBEING................................................................................................. 27
15 HEALTH AND SAFETY............................................................................................................. 28
16 SECURITY................................................................................................................................. 29
17 ENTERTAINMENT.................................................................................................................... 31
18 STATUS AND REPUTATION.................................................................................................... 32
19 OTHER SPECIALISED BUILDING REQUIREMENTS.......................................................... 33
20 AI AND PANDEMICS................................................................................................................ 34
21 THE IMPORTANCE OF A HOLISTIC APPROACH................................................................. 38
22 AI AND BUILDING SERVICES: THE CULTURAL DIVIDE..................................................... 40
23 AI WILL REQUIRE ITS OWN STANDARDS........................................................................... 41
24 CONCLUSIONS........................................................................................................................ 44

© BSRIA |3
1 INTRODUCTION
Ever since humans first started moving away from a It also looks at other key stakeholders such as owners,
hunter-gatherer existence thousands of years ago, they tenants and people who use the buildings, for example
have been steadily transforming their environment, as employees or customers. Those responsible for
whether deliberately or, in cases such as climate change, installation, operations and maintenance also have a
as an unintended consequence. key interest.
The waves of change created by humans have been The paper considers the role of AI in helping buildings
accelerating. Artificial intelligence (AI) has existed as a to provide comfortable, secure and safe environments
coherent concept only since the 1950s but is already which are also engaging and convenient for their
starting to change our lives in many ways. Some have users. It recognises that security includes not just the
even suggested that in another 60 to 70 years, AI physical building but also cybersecurity, where AI has an
could have transformed our world so much that it will enhanced role to play.
effectively end the era of human domination which
began a few millennia ago and pass control to a new It looks at AI as a factor in reducing the cost of running a
and more advanced intelligence. building, whether it is a question of energy efficiency or
the cost and reliability of service and maintenance and
This paper is less concerned with such dramatic the imperative to keep the building functioning.
propositions and more with the practical changes that
AI is already bringing to building services and what it is It recognises other goals that building managers may
likely to make possible within the next few years. adopt or have imposed on them, such as on-site
generation and distribution of renewable energy,
To date, AI in building systems is still in an early stage. which is made much easier with the help of AI. It
This is partly because AI itself is still in its infancy in also recognises other factors that may motivate
many ways and building services are traditionally fairly building managers, such as providing entertainment
conservative. or enhancing their status and prestige – an often-
underrated motivation.
The paper considers what we mean when we talk about
AI in buildings, and distinguishes between different The paper also touches briefly on the significance of
levels of AI. It recognises that AI is a process, while standards and regulation, both those aimed specifically
building designers, owners and users are generally more at AI and those which influence it.
interested in outcomes. It also examines the different
levels of intelligence that can be found in a building, The paper acknowledges that COVID-19 has not only
both now and in future. As part of this, it should help presented building owners and managers with huge
readers distinguish between the claims made about AI and unexpected challenges, but also affected the way
in buildings and the reality. society views many non-residential buildings and the
very willingness of people to use them. Some of this
It homes in on the main aims that different decision- change may well be long-term, and so consideration is
makers have at different stages of a project. These given to the relevance of AI in a post pandemic age.
include the developer who wants to ensure that the
building is completed on time, within budget and that it A paper of some 15,000 words can only provide a
is attractive enough to make a profit whether when sold very high-level view of this fascinating and important
or as a rental property. AI is relevant to each of these subject. Hopefully, however, readers will be inspired to
ends. enquire further into some of the areas in more detail.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
4|
Provided by Accuris
ARTIFICIAL INTELLIGENCE IN BUILDINGS - THE SILENT REVOLUTION
2 SCOPE
2.1 TECHNOLOGY DISCUSSED There are many existing and potential applications
Artificial intelligence is a huge subject. Even when for AI in homes. Some existing smart home products,
examined in the context of buildings and building including learning thermostats and voice-activated
technology, this paper can only scratch the surface. In devices, already deploy a form of AI. There is also a
particular, the examples given represent only a small long history of technology first deployed in the home
selection of the AI solutions that are already available or transferring to the non-residential sector, and vice
likely to become available. versa.
The BSRIA White Paper WP 9/2018 Megatrends: However, the scale of homes, and the fact that they
Smart Building Technology provides some additional are usually designed for and occupied by a very small
background information and observations that may be number of people, means that they have different
useful. attributes to a typical non-residential building.
2.3 THE IMPACT OF COVID-19

Some existing smart home The COVID-19 pandemic has already had a massive
products, including learning impact on many aspects of life, including the use of
buildings and attitudes towards them. While the long-
thermostats and voice- term impact is still unclear, it seems quite likely that
activated devices, already building services will continue to be affected in the
deploy a form of AI. medium to long term. Smarter technology, and AI in
particular, is already being presented in various forms
as a means of addressing or mitigating some of the
problems caused by COVID-19.
2.2 TYPES OF BUILDING
This paper focuses on the actual and potential Because of its centrality to building services, this is
application of artificial intelligence in non-residential worth its own section.
buildings.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA |5
3 WHAT IS ARTIFICIAL
INTELLIGENCE?
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
The term artificial intelligence (AI) was first coined
by mathematician John McCarthy in 1956. While the
concept of “thinking machines” had been around for
many years, it was mostly confined to the world of
fiction and theoretical speculation until the advent of
serious mechanical and electrical computing. It was
computing pioneer Alan Turing who, in 1950, posited
the now famous Turing test. This was put forward as
a means of answering the question “Can a machine
think?”.
McCarthy defined intelligence thus: “Intelligence is the

computational part of the ability to achieve goals in the
world. Varying kinds and degrees of intelligence occur in
people, many animals and some machines.”
McCarthy’s allusion to “varying degrees of intelligence” is thinking. This is similar to the saying “If it walks like a
seems to point to something important. We have duck and quacks like a duck, it’s a duck”. The strength
different expectations from an “intelligent cat”, an of this reasoning is that it approximates to the way we
“intelligent five-year-old child” and an “intelligent judge the intelligence of humans and other animals. We
President of the United States” because the reference do not have direct access to their thoughts; we can only
point and benchmarks are different. In this context observe their behaviour and make inferences.
the whole idea of describing a computer, a device or a
building as “intelligent” seems problematic. The Turing test is, however, a very tough challenge. No
machine has yet passed it, and in actual tests, some
3.1 THE TURING TEST humans have actually failed it – after all, we have all
The core of Turing’s answer to the question “Can met humans whom we might describe as robotic.
machines think?” is the proposition that if a machine Beyond this, some philosophers of mind have argued
can produce responses that are indistinguishable from that even passing the test would not necessarily mean
those of a human, then this has important implications. that the machine was actually thinking in the way
Arguably, we have as much reason to describe a that we understand it. However, that is a subject for a
machine as thinking as we would to say that a human fascinating white paper in its own right.
6|
Provided by Accuris
3.2 MORE THAN JUST FOLLOWING Howard Gardner's nine

modalities of intelligence
INSTRUCTIONS
It is reasonable to say that genuine AI goes beyond
simply following a set of instructions, however complex.
Instructions of the form “If X, then do Y, else….” will Musical-rhythmic
always lead to a predictable answer. and harmonic
However, if one relies on fixed rules, then it does not
take long for the number of scenarios to become so
large that it is in practice impossible to write the rules Visual-spatial
to cover all of them. It may also take too long to work
through all the options.
Intelligence implies an ability, however limited, to

Linguistic-verbal
learn from one situation and adapt the knowledge to
another situation. Humans formulate new rules based
on experience, for example that small red round things
taste nice. They then discover exceptions (such as that it
doesn’t taste nice if it is made of plastic or if it is a bitter Logical-mathematical
pill) and then exceptions to those exceptions.
3.3 INTELLIGENCE IS A MATTER OF DEGREE

IQ tests are founded on the idea that intelligence can Bodily-kinaesthetic
be, to a degree, measured and ranked. While this form
of measurement is contentious, most would accept that
some people are, broadly speaking, more intelligent
than others, while most would accept that a pig is less
intelligent than a human but more intelligent than a Interpersonal
goldfish.
3.4 INTELLIGENCE IS OFTEN TASK-FOCUSED

However, we also accept that some humans have very Intrapersonal
specific areas of intelligence. People can have excellent
verbal skills, but be poor at maths, while others can be
very creative but have poor emotional intelligence or
any combination. Naturalistic
In the 1980s, Howard Gardner postulated nine
modalities of intelligence (see opposite).
AI clearly has its roots in category 4, but is gradually Existential
expanding into other domains, though it may be some
time before we have AI systems capable of existential
thought. In particular, French authors and philosophers
can probably feel safe for the time being.
© BSRIA |7
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

Training and education encourage people to focus on 3.6 INTELLIGENCE (USUALLY) REQUIRES
quite narrow and specific areas of expertise. We expect NETWORKING
the most capable people to be able to reapply their To collect, transmit and share data requires some form
skills to a degree but would not expect a brain surgeon of physical networking. In the 19th century, development
to branch out in middle age and become a leading of the telegraph allowed information to be transmitted
financial lawyer. globally in close to real time but in very small volumes.
Since then, there has been an accelerating revolution
Most artificial intelligence to date has been applied to in communications. Most commercial buildings now
specific areas, such as temperature control, driving a have a local area network (LAN), many have wireless
vehicle, or winning a complex game. While some of the networks and the great majority have access to the
logic can be reapplied, this takes the application of new internet and to other communication networks.
rules and normally new training is required.
3.7 AI REQUIRES PROCESSING POWER
Moore’s law holds that the number of transistors that
The past few years have seen can be concentrated into a single integrated circuit
revolutions in the way that data doubles approximately every two years. While obviously
not extendible indefinitely, it has held good since about
can be collected, in the volumes
1970. Taking into account the power of exponential
in which it can be stored and growth, this represents a phenomenal increase in the
processed and in the ways that it processing power that can be accommodated in even
can be analysed. small objects.
At the same time, there has been a similar downward

3.5 AI REQUIRES DATA trend in the cost of processing power. This combination
Any intelligent action in the real world requires of vastly increased processing power and lower cost
knowledge about the state of the world, which requires makes it possible to solve complex problems involving
data picked up through direct observation or second- masses of data much more quickly, and also to
hand. distribute computing power to cheap everyday devices
from wristwatches to consumer white goods.
A smart thermostat uses data relating to temperatures
at different times, the user’s interaction with the control
3.8 THE IMPORTANCE OF THE CLOUD
and their presence or absence. The more relevant data
The combination of cheap mass processing power and
that is collected, from a wider range of sources, the
fast, reliable and secure communications has made
more appropriate the response can become.
it possible to transfer a lot of data processing and
analytics to dedicated data centres, located in the cloud.
The past few years have seen revolutions in the way
The data still has a location but does not need to be
that data can be collected (especially via the internet),
accommodated or directly managed by the organisation
in the volumes in which it can be stored and processed
that is using it.
and in the ways that it can be analysed.
This means that an organisation or individual can access
AI without needing to house or own the computing
resources needed to deliver this. This is key to making AI
a generally available phenomenon.
8| © BSRIA
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

4 WEAK AND STRONG ARTIFICIAL

INTELLIGENCE
4.1 WEAK AI 4.2 STRONG AI
Weak AI is also sometimes called narrow AI. It implies Strong AI, also sometimes called artificial general
an AI system that is capable of learning but in a limited intelligence (AGI), relates to an intelligence that can
and focused area. In the context of building services, be applied to an indefinite number of scenarios in the
one example would be a smart learning thermostat. same sort of general way that human intelligence can
This tracks the way that users adjust the temperature be applied. Any machine capable of passing the Turing
at different times and in different situations and starts test could be described as possessing strong AI, because
looking for patterns, which it can then use to make it would be giving apparently intelligent answers to a
predictions about what is most likely to meet the range of questions that it couldn’t predict.
user’s requirements in future while limiting energy
consumption.
Other examples of weak AI in building services would

include voice recognition devices – such as Amazon
Alexa – and security systems which can interpret a video
picture of someone approaching a building and identify
it as a known person.
Weak AI may well perform its specific tasks much more

quickly and with less error risk than a human could do,
thanks to a combination of faster processing power on
the one hand and following logical rules on the other.
Weak AI is also not subject to the competing emotions,
desires and interests that can distract humans.
THE LIMITATIONS OF WEAK AI

Weak AI only works for a specific task, and its
intelligence is typically largely irrelevant when applied Strong AI is sometimes equated with an ability to have
in a different scenario. For example, the AI capabilities conscious thoughts. However, there is some debate
of a smart thermostat will be of little relevance when as to whether a non-biological system can ever be
trying to train a security system to recognise certain conscious, or how that could be verified. In any case,
individuals based on their appearance. as of 2021, we still seem to be some distance away
from any AI which could be classified as strong. The
In the training phase, the AI system is likely to make consensus is that this will take at least several decades,
wrong calls. In some situations, this could be acceptable but developments of this magnitude are impossible to
but in others it could be disastrous, for example where a forecast accurately.
wrong decision is made as to whether there is a serious
threat, such as a fire. However, the boundary between weak and strong AI is
not hard and fast, any more than we can draw a simple
line between more or less intelligent animals. What we
the boundary between weak are likely to see is an intermediate stage where a system
and strong AI is not hard and fast, is given goals. For example, to save energy or improve
occupational comfort, an AI system may be able to find
any more than we can draw a new ways of addressing these which had not previously
simple line between more or less been thought of by humans and so had not been
intelligent animals. included in the original remit.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA |9
4.3 THE SINGULARITY DIVERSITY OF DATA SOURCES AND DATA SHARING

In the context of AI, the singularity refers to a scenario The process of applying AI can be assisted by collecting
in which the intelligence of machines, whether and using data from a range of different sources. For
individual or collective, has moved beyond human example, if a company is supporting a building system
intelligence to the point where their processes using AI, then if data can be collected from a range of
and decisions can no longer be fully controlled or different buildings, more can be learned in a shorter
understood by humans. The world would then, space of time, especially relating to less commonly
potentially, have moved beyond human control. occurring events.
For the purposes of this paper this should probably not This is analogous to the human process of learning.
detain us long. There is no way of forecasting when, If someone can learn about the experiences of other
if ever, such a singularity might ensue, let alone what people, whether by communicating with them or by
consequences it would have or what could or should be learning about them through various media then they
done in response to it. are likely to learn much more than if they rely solely on
their own experiences.
Nonetheless, it is worth noting that even outside the
context of AI there are many systems which are not This of course raises questions of data privacy and
fully understood or totally controllable. For example, confidentiality, whether the data is personally sensitive
this applies to the operation of complex markets. In the or of some business value.
context of building services today, it could well apply
to the process of combatting the course of an epidemic PROCESSING METHODS IN AI
within a building. The process of arriving at decisions in a machine
based on processing chips is obviously very different
4.4 ASPECTS OF AI to that which occurs in a human brain. One important
TRAINING difference is that a machine can accurately run through
In order to apply AI to a real-world situation, there a very large number of possible options far more
is normally a training period involved as the device quickly. For example, in playing a moderately complex
attempts to find the correct pattern, with adjustments game like chess, a machine can run through many
made as more evidence comes, for example where the millions of options in a way that a human could not in
user overrides the smart device’s response. Training can any sensible timeframe.
be supervised, semi-supervised or unsupervised.
For scenarios which involve a limited number of options,
In reality, just as an intelligent human never stops even if that number is very large, an AI system is likely
learning, so an AI system will, ideally, continue to make to hold an advantage over a human.
adjustments in response to new data. Of course, as
with humans, this activity of “leaning on the job” needs A fully networked building system, especially if it
to have some controls and limits placed on it to avoid manages a campus of multiple buildings, is likely to be
serious consequences of bad decisions. collecting many millions of pieces of data on a daily
basis. AI will likely have the edge in interpreting this.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
10 | © BSRIA
Humans, in contrast, often cope with their own MACHINE LEARNING

limitations by limiting the number of options to try, Machine learning deploys algorithms that are designed
which is one reason that we tend to dislike processes to improve automatically based on the results obtained.
involving very large numbers of choices. A key part of It is especially deployed in predictive analytics, where
human intelligence depends on being able to select the predictions are made and then readjusted. In the
most appropriate lines to pursue, which often involves context of building services, an obvious application
intuition, or the so-called Eureka! moment. At present, would be in predictive maintenance, where, for
AI is less well adapted to this. example, data on a machine’s state, collected over time,
can be used to predict when it is likely to need service
NEURAL NETWORKS or replacement.
In the context of AI, neural networks represent
an attempt to emulate some of the processes of a DEEP LEARNING
biological neural network such as that found in a human Deep learning is a class of machine learning algorithms
brain. While artificial neural networks are nowhere that uses multiple layers to progressively extract higher-
near as complex as a human brain, they are particularly level features from the raw input. For example, in image
useful in identifying common traits between different processing, lower layers may identify edges, while
‘experiences’. For example, in building security, a human higher layers may identify the concepts relevant to a
will normally fairly easily recognise a familiar person human such as digits, letters or faces.
even with a different hairstyle, with or without glasses
and with or without a suntan or in different lighting
conditions.
A neural network can help an AI system learn what the

common points are that make it possible to identify all
these as belonging to the same person. This is just one
practical example of a neural network.
Input Hidden Output
layer layer layer
A simple example
of a neural network
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 11
COLLABORATION WITH HUMANS or university background have tended to be successful

In building services, as in other disciplines, a lot of in the past, then the AI system might identify the
the debate to date has focused on how likely AI is to correlation as significant, and as a result favour similar
replace human operatives. However, in many ways, AI, candidates. In this instance, if the selection in the past
especially in its current weak state, complements human was influenced by human bias – such as a recruiter’s
intelligence. favouring alumni from his or her university, then
the AI system might well reproduce this in its own
Humans still hold an advantage when it comes to setting recommendations.
out base hypotheses and avenues of analysis to pursue,
while AI is much more efficient at processing huge This type of bias will usually be something to avoid but
volumes of information to confirm or refute the original is likely to be much harder to identify in some instances
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
hypotheses, and also identify relationships and patterns than in others.

that humans would find difficult to spot.
THE AI INFLATOR?
One of the most haunting questions posed by AI is
With smart technology there is already a tendency for
how this process of collaboration will evolve as AI gets
suppliers and reporters to overuse the word “smart”.
progressively stronger.
There is, similarly, a tendency to ascribe artificial
intelligence to processes which would probably not meet
THE ISSUE OF BIAS the criteria for even weak AI.
It is often claimed that one advantage of AI is that it
avoids the biases, whether conscious or unconscious, to A process which simply follows programmed instructions,
which humans are prone. It is true that AI in its current however complex, without any process for learning or
state avoids emotions, distractions or other impediments for applying rules or algorithms to unforeseen scenarios,
such as tiredness. It is, however, possible for an AI cannot really be described as AI.
system to absorb biases. This can happen where it is
the intention of the designers, but it could also be an As AI becomes increasingly talked about, we can expect
unintended side effect. many suppliers to try to use the term beyond the point
where it is justified. When investigating claims of AI, it is
If, for example, AI is used to screen job candidates, always worth asking what exactly justifies the claim of
then if candidates of a certain gender, age, ethnicity intelligence.
12 | © BSRIA
5 WHEN IS A BUILDING SMART

OR INTELLIGENT?
Any discussion of smart or intelligent buildings has This paper suggests a pyramid of smartness, rising
been hampered by the absence of generally agreed progressively from the least to the most intelligent.
definitions of these terms, especially one indicating how Only in the upper half of the pyramid can building
they might be applied to buildings. functions be meaningfully described as deploying AI,
and then mostly AI of the weakest kind. It should also
This has encouraged rather loose claims that systems be noticeable that the lower stages of the pyramid
and devices offer smartness or intelligence. can apply to objects, whereas the higher stages could
be thought of in terms of connected and organised
systems.
The pyramid of
smart and intelligent
Smartness is a Con�nuum buildings
10 Drawing from the wider world Strong

--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
9 Iden�fying new goals

8 A self-improving system Ar�ficial
intelligence
7 Taking preventa�ve ac�on
6 Recommending ac�ons to problems
5 Finding excep�ons / anomalies Weak

4 Capable of interpre�ng informa�on
3 Connected / communica�ng
2 Programmable
1 Mechanised
0 Dumb object

© BSRIA | 13
Dumb objects are purely passive. They include basic An example is a dirty filter leading an air conditioning
items forming the structure of the building. They may unit to use increased power to force air through. An
incorporate moving parts, such as valves which are important point to be noted is that the weakest AI
adjusted manually. systems may not be able to point to the cause of the
problem. However, by learning what is normal and then
Mechanised devices perform a task which is fixed or identifying the exception they can flag it for human
can be configured from a limited range of options. intervention in a situation where a human may not have
An example would be a thermostat, which can be set noticed the problem for some time.
physically for different timings and temperatures on
different days. Recommending actions in response to problems
represents a further step towards AI. If we take the
Programmable devices can be set to adjust to an open- previously mentioned example of a dirty or blocked
ended number of scenarios and functions but using a filter, the AI system may, by collecting information
language which is essentially limited by “if, then” type from multiple systems over a period of time, identify
rules. An example would be a direct digital controller a pattern where excessive energy use is followed
(DDC) which can be programmed to control different by a filter change – which is duly registered in the
devices applying appropriate rules. system. The system finds a correlation and can in time
recommend changing the filter as a solution to the
Connected / communicating devices transmit and problem.
receive data and/or instructions. An example would be a
communicating room controller. Taking preventative action. The old saw that
“prevention is better than cure” also applies to building
A device receiving information needs to be capable systems. AI may be able to identify a pattern of events
of interpreting it. This may entail some form of data which, over time, can lead to the failure of a device.
translation if the information originates from a device For example, a pump may show increasing vibrations
using a different taxonomy. An example would be an leading up to failure. By collecting and analysing data,
air conditioning unit which adjusts temperature or air a building automation system can identify a point at
throughput based on information received from sensors which the pump is likely to fail within a given period, for
or from a building automation system. example four weeks, unless remedial action is taken.
Finding exceptions and anomalies is arguably the first A self-improving system. An AI system learns by
stage at which one can start referring to real AI. When a formulating rules that fit the data it receives. Over time
device is collecting data over time, for example relating it is quite likely that exceptions will occur. In this case
to electrical power usage, then typically a pattern will the learning algorithms will formulate new or revised
start to emerge with peaks and troughs for different rules which fit the exceptions better. This process can
devices at different times on different days. continue indefinitely. For example, an AC unit may use
If a spike or trough then occurs at a time when it increased power for a number of different reasons. Over
has not previously been encountered, then this time, AI may get better at distinguishing between the
suggests that something anomalous has happened. possible causes.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
14 | © BSRIA
Identifying new goals. At this stage, we are moving Drawing from the wider world. If we think of strong AI
into the realms of strong AI which behaves more like as entailing, among other things, the ability to emulate
an intelligent human. Most humans are familiar with human behaviour, then it needs to be able to draw on
a scenario where we suddenly see a problem in a a huge amount of information. In the case of intelligent
completely different way and think of a possible solution buildings, this goes well beyond the information
that does not simply derive from previous ideas. Many collected within the building or even from a wide
scientific discoveries and inventions originate in this collection of buildings. Buildings increasingly interact
way. An AI system could identify a correlation that no with other systems including weather systems, power
human had thought of and recommend a new course of grids, traffic systems and security systems. One example
action. For example, an AI system could notice that men might be a building security system that checks arrivals
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
and women, on average, respond better at different into a building against a database of images of people
ambient temperatures and adjust areas of the building wanted by the police. As in many cases involving AI, this
accordingly. Where AI starts to get both exciting and raises questions of privacy and consent.
disturbing is with the possibility of AI identifying actions
that no human has yet thought of. To take a hypothetic
example, it may work out that playing ABBA songs at
double speed in the morning improves productivity in a
typical office environment. Perhaps an unlikely example,
but AI is likely to unearth actions that are even more
improbable.

© BSRIA | 15
6 THE KEY AIMS OF BUILDINGS

In order to understand the scope for use of AI within OPERATIONAL AND MAINTENANCE EFFICIENCY
buildings it is helpful to look at the goals of those Operational and maintenance efficiency reduces the
responsible for designing, commissioning, running cost of maintenance and also reduces the risk posed by
and occupying buildings. Every building has more than outages.
one aim, and most have multiple aims which vary in
importance at different stages in the building’s lifespan. EASE OF CONSTRUCTION
These aims are also affected by owners’ and users’ This relates to the ease and efficiency with which
changing regulations and priorities. buildings are constructed. The challenge of completing
buildings on time, to specification and on budget is
Some of these aims are likely to overlap or work in particularly important.
synergy. Others may potentially conflict. Sections 7 to
19 of this paper consider these in detail and they are
COMFORT AND WELLBEING
briefly summarised below.
Wellbeing includes both physical and psychological
aspects. Building owners and operators typically wish to
REGULATION AND STANDARDS further wellbeing as an end in itself and in furtherance
Regulations and standards (both legally enforced of other aims, such as improving employee productivity
standards and standards that emerge through the and retention, or encouraging customers to revisit.
market) set constraints within which buildings must
operate.
HEALTH AND SAFETY
Developers and building operators need to adhere to
PROFIT health and safety regulations. Beyond this, maintaining
Where a building is constructed for commercial a healthy and safe environment has always been a key
purposes the developers and contractors usually aim aim of buildings.
to make a profit. The same goes for commercial letting.
Even where the aim is not totally commercial, the
SECURITY
owners or operators usually want to avoid unnecessary
Security includes protecting people and property within
costs.
a building, preventing unauthorised access, and also,
increasingly, cybersecurity.
ENERGY EFFICIENCY
Making buildings as energy efficient as possible reduces
ENTERTAINMENT
operational cost, helps to meet environmental targets
Some buildings have entertainment as a key objective.
and can improve the owner’s and operator’s reputation.
Others incorporate entertaining elements as a means
to other ends, for example to encourage employee or
ON-SITE RENEWABLE ENERGY customer retention.
By generating energy on-site which can be used, stored
or exported to the grid, buildings can help achieve
STATUS AND REPUTATION
environmental objectives and, potentially, reduce
Buildings, including the services that operate them, can
operational costs.
be an important statement of the status and values of
the owner or occupier.
CONVENIENCE
Convenience and reliability ensure that the services
OTHER SPECIALISED BUILDING REQUIREMENTS
offered by a building are available and functioning when
Some buildings have a specific form or features that
needed and are easy to use, access and navigate.
relate to their function.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
16 | © BSRIA
7 REGULATIONS AND STANDARDS

SET THE SCENE FOR BUILDINGS
This section considers ways in which regulations and 7.1 THE REGULATORY PROBLEM
standards impact on the decisions and priorities of As technology becomes more complex and fast-
those responsible for a building. Later on in this paper, changing, this poses an increasing problem to would-be
section 23 AI will require its own standards, considers regulators. Given the still somewhat blurred and shifting
the moves to create standards that can be used to definition of AI, it is not surprising that legislators often
influence and control the development and use of AI struggle to understand the concept, how it could be
itself. regulated, and the advantages and disadvantages of so
doing.
In a rapidly advancing field, there is also a risk that any

Regulations and standards regulations or standards may soon become obsolete.
regarding buildings and their Legislators such as Members of Parliament tend to be
operations tend to be focused generalists and are not always best placed to make
regulatory decisions, or even to fully understand advice
more on outcomes rather than offered by subject matter experts.
on regulating the detailed means
by which these are achieved.
Regulations and standards regarding buildings and their

operations tend to be focused more on outcomes rather
than on regulating the detailed means by which these
are achieved.
Regulations generally include penalties for non-

compliance. If the penalties are light or the regulations
weakly enforced, then they may be widely ignored.
However, building operators will often be aware of the
potential risk to their reputation where they are shown
to be non-compliant.
Standards range from those which are incorporated into

regulations to ones which emerge through commercial
pressures. The latter include standards that are
increasingly adopted because others in the industry are
already using them. This is an example of the so-called
network effect.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 17
7.2 REGULATIONS AND STANDARDS “The smart readiness indicator should be used to
REQUIRING AI measure the capacity of buildings to use information
With AI still in its infancy, government action to and communication technologies and electronic
encourage use of AI is likely to be directed more at systems to adapt the operation of buildings to the needs
economic objectives, such as developing an advanced of the occupants and the grid and to improve the energy
digital economy rather than at specific tasks. There is efficiency and overall performance of buildings.”
little impetus for legislation specifically to promote AI in
buildings. The EPBD does not, however, refer directly to artificial
intelligence.
The SPIRE standard, launched in September 2020,
claims to be the first global standard promoting smart In February 2020, the European Commission published
buildings. Set up in collaboration between Underwriters its White Paper on Artificial Intelligence: A European
Laboratories (UL) and the US Telecommunications approach to excellence and trust[ii]. This made no explicit
Industry Association (TIA) it is claimed to be “the references to buildings.
world’s first comprehensive smart building assessment
and rating program”. 7.3 REGULATIONS AND STANDARDS TO
CONTROL AI
The SPIRE standard provides tools for self-certification These are likely to emerge in future. AI is already
and also provides verified assessments. It focuses on six implicated in the fact that most AI entails the collection
dimensions: and processing of large volumes of data. Where this
data is classifiable as personal data or as commercially
• Power and energy sensitive data then it is quite likely to come within the
• Health and wellbeing remit of data protection or data privacy regulations.
• Life and property safety As far as buildings are concerned, certain systems which
• Connectivity collect data for the purposes of wellbeing and comfort
could be affected by privacy rules. These could include
• Cybersecurity HVAC systems which make use of data collected by
• Sustainability wearable devices in order to optimise temperature or
humidity.
More information about the SPIRE standard can be
found at https://spiresmartbuildings.ul.com/ Building owners may also wish to limit the ways in which
data about their buildings is collected by suppliers of AI
The information provided to date does not include a technology, for example on grounds of confidentiality or
very clear definition of smart buildings, or of the role of business sensitivity.
AI. However, it seems clear that if standards of this type
are to gain support in future they are likely to include at However, as AI is used to make more critical decisions it
least an implicit reference.The European Union’s Energy is quite likely that regulation of its use will be extended.
Performance of Buildings Directive (EPBD), as updated in Examples of critical health and safety rules could
2018[i], includes numerous references to smart buildings concern AI used to take decisions about responses
and smart readiness. These include an indication that to a danger such as fire, or to letting individuals into
buildings.
i https://ec.europa.eu/energy/topics/energy-efficiency/energy-efficient-buildings/energy-performance-buildings-directive_en
ii https://ec.europa.eu/info/sites/info/files/commission-white-paper-artificial-intelligence-feb2020_en.pdf
18 | © BSRIA
--``,`,,``,```,

--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
7.4 THE EFFECTS OF LIABILITY LAWS Systems which continually collect data from various
One of the many interesting questions posed by AI building systems can identify when those are in breach
relates to the legal liability resulting from decisions that of regulations.
are taken by AI. Examples of this are more obvious in
systems such as self-driving vehicles where a decision AI can be deployed to predict when systems are likely
could result in death or injury, but this could also to become non-compliant and to invoke preventative
happen with critical building systems. For example, if maintenance. They can do this most obviously by
someone were to become trapped in a room because analysing data collected over time, ideally from a range
the AI system failed to recognise them and allow the of different sites and finding events that preceded
door to open for them to leave, then they would have failure.
a case against the manufacturer or operator of the
system. AI is also likely to play a role in carrying out regulatory
checks on buildings. For example, detailed data
At some time in the future one might even envisage the provided regarding energy performance of different
AI becoming liable. Until then, the designers, producers parts of a building can be analysed to show exactly
or owners of the AI will be liable. This will itself impose where and when any breach has occurred and enable
a discipline. remedial action.
7.5 USE OF AI TO ENFORCE REGULATIONS For example, if a building’s energy usage contravenes
One of the many barriers to enforcing health and standards, it should be possible to identify which units
safety regulations in buildings lies in making it easier to are responsible for the excessive usage.
establish whether they are compliant with regulations,
and to monitor when any breach that occurs. Until now,
this has tended to involve periodic inspections.

© BSRIA | 19
8 PROFIT
Where a building is constructed, or renovated, 8.2 REDUCED ENERGY BILLS
principally in order to make a profit, this objective Energy bills represent a direct and measurable cost
is achieved by the basic mathematical process of to most building owners or tenants. Where a smarter
maximising the margin between the costs of the project building management system can be shown to reduce
on the one hand and the revenues generated on the energy expenditure, and where it can identify energy
other. wastage and suggest or implement remedial action,
then that is likely to increase the resale or rental value
Where the building is intended for resale, then that of the property.
difference is comparatively easy to calculate once the
building has been sold. Where the building is intended AI potentially allows energy savings to take account of
for rental, then the calculation is more complex, based more subtle and complex factors, such as how building
on the anticipated revenue stream. users respond to changes in heating or lighting levels.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
Clients buying a building may well also attribute 8.3 IMPROVED WELLBEING
different value to various building features, including AI, Until the COVID-19 Pandemic, wellbeing, while
than those who take it on as tenants. AI is still mainly increasingly talked about, looked doomed to remain
seen as a means to an end and is likely to add value in the guest on the fringes of the party and not the star of
functions where it contributes directly to the aims of the show. Given the problems associated with placing
the purchaser or tenant. a monetary value on wellbeing, it was hard to envisage
clients being willing to pay higher prices or rentals for
8.1 THE DISJUNCTION BETWEEN WHO PAYS systems designed to promote wellbeing.
AND WHO BENEFITS
A developer will normally only invest in features or The pandemic may well change this, given the
functionality which are required by law or where the understandable reluctance of people to enter public
purchaser is likely to demand it as part of the deal and commercial buildings unless they have assurances
or pay a premium for it. In the present situation a that the risk of infection is negligible. However, this is
developer is unlikely to invest in solutions involving likely to take time to work its way through the cycle of
AI purely on spec. This is likely to change as clients development and refurbishment.
become better informed about the outcomes that AI
can achieve. They may well then come to accept certain
functions in a building which require AI.
Where a smarter building

management system can
be shown to reduce energy
expenditure, then that is likely
to increase the resale or rental
value of the property.
20 | © BSRIA
9 ENERGY EFFICIENCY
Energy efficiency has long been a central aim of building Occupants of a building will tend not to pay much
services. The oil price shock of the 1970s coincided with attention to the HVAC system as long as it is delivering
the emergence of building automation systems. From a comfortable environment. Situations can arise where,
the 1980s onwards, climate change and the role of CO2 for example, heating and cooling are working against
emissions have figured increasingly large. each other. This can occur where the two devices
are not coordinated. Each will ramp up its activity to
Buildings account for some 40% of greenhouse gas counter the other.
emissions in developed nations such as the UK – about
half of this from non-residential buildings. While the result is a ‘comfortable’ environment, energy
is being wasted. As well as identifying the conflict and
Much of the focus on improving the energy efficiency wastage, AI can adjust the conditions to meet more
of buildings has rested on improvements to the building complex needs.
structure, and also to the efficiency of devices such as
boilers and air conditioning units. For example, the people shown below may each
have different comfort requirements. If the building
AI has a role to play both in identifying malfunctioning has enough HVAC outlets, then it may be possible to
equipment and in finding logical inconsistencies in the customise the environment for each person while
way equipment is behaving. minimising the risk of the outlets wasting energy by
playing against each other.
One application of AI which has been widely cited to
the point of cliché deserves mention as it provides a
good example of how AI can reach beyond what is easily
observable by the human observer.
Using AI to Identify and Correct

Heating Anomalies
Energy use over time Energy use adjusted
Device A Heating Heating

Set to heat Cooling Cooling
Comfortable conditions Comfortable conditions

maintained
Device B
Set to cool
Abnormal energy Normal energy use

use detected maintained
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 21
10 ON-SITE RENEWABLE ENERGY

While much of the impetus to reduce the environmental energy from the grid versus that received for energy
impact of buildings has focused on improving energy exported. This may well vary over a short timeframe
efficiency, using buildings and sites to produce depending on demand. To get the best results, attention
renewable energy is also increasingly common. needs to be paid to what is expected to happen, for
example, whether demand for energy is expected to
Examples include solar panels, solar heating, wind rise, both externally and internally.
turbines and ground, water and air source heat pumps.
There may also be sources which, while not entirely Other considerations may also play an important role.
renewable, still improve energy efficiency such as For example, if the supply from the grid has been found
combined heat and power (CHP). to be unreliable, then there may be a need to store
energy on-site as a precautionary measure.
A building automation system which has access to
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
renewable energy generated on site normally has three This means that calculations can soon become quite
main options when it comes to the use of this energy: complex, and also constantly changing. An AI-based
It can be consumed directly, it can be stored – most system can use data collected from past experience and
obviously using batteries, or it can be exported to the from data gathered from other sources to ensure that
grid. Stored energy can also be either consumed directly price advantages are exploited while maintaining an
or exported. acceptable level of energy security and resilience.
The decision about the use of this energy involves a

number of factors. These include the price paid for
Use of AI to Manage Onsite

Energy Generation
Onsite solar PV
Onsite
energy Onsite CHP
sources
Onsite wind
Decision criteria:
AI-based building Energy price
energy
Storage needs
management
Future needs…
Other inputs
and outputs
Onsite energy
Grid Input / Output storage input/output
Energy for building
services and devices
22 | © BSRIA
11 CONVENIENCE
Convenience and ease of use are important factors Speech recognition is a classic example of training
in the design of buildings and of the services that by backpropagation. The system is presented with
are provided. Building users, whether employees or repeated examples of people saying something like
customers, wish to navigate the building as quickly and “third floor” and eventually identifies certain features
easily as possible. Time spent getting through security, that all or most of these have in common.
waiting for lifts or even just working out how to use
controls for lighting, HVAC or visual aids is likely to be Voice activation has the potential for a wide range
resented and also represents a cost. of simple tasks from instructing lifts to adjusting
temperatures and lights, opening blinds or even
11.1 AN EXAMPLE: VOICE RECOGNITION instructing coffee machines.
One type of AI which has particularly taken off in
residential applications but has been slower to gain As with other technology in its early stages, there still
traction in the commercial sector is voice recognition needs to be a fall-back in case, for some reason, the
and voice activation. voice recognition fails or misinterprets the instructions.
There are also questions relating to privacy and
Voice activation is a classic application of AI, given confidentiality in a commercial environment.
that people’s accents, pronunciation, speech speed,
vocabulary and grammar vary enormously, even when
they are notionally speaking the same language. The
report of early speech recognition famously failing to
cope with Scottish accents caused both indignation
and amusement, partly depending on what accent one
happened to have oneself.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`

© BSRIA | 23
11.2 AN EXAMPLE: LIFT MANAGEMENT Lift usage patterns vary depending on the time of
Anyone who regularly uses lifts in high rise buildings will day and the day of the week. In an office building, at
have experienced the frustration of having to wait what the start of the working day, most journeys start at
feels like an unnecessarily long time. Programming lifts the lobby; at the end of the day it is the reverse, with
so that they prioritise floors in the correct order and different patterns in between.
adopt the most appropriate waiting position can quickly
become quite a complex process. An AI system can learn typical daily and hourly patterns
and prioritise these to minimise waiting times.
Voice recognition training - a simplified example
“Third floor…”
AI system
Training system
“Thard flarr” to recognise
“Third floor”
“Thirrud floorr”
“Tird floyr” System recognises

“Third floor”
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
24 | © BSRIA
12 OPERATIONAL AND
MAINTENANCE EFFICIENCY
The operation and maintenance of buildings and As air conditioning filters become blocked, the unit
building systems represent a significant ongoing may need increased power to produce the same result.
expenditure. At the same time, the consequences of any Since other problems may produce similar symptoms,
key building systems failing can be severe. These can AI can help narrow down the actual problem. A pump
range from inconvenience to loss of revenue, discomfort may start to vibrate more than it would normally. Doors
of the building occupants or even, in extreme cases, closing on a lift may encounter abnormal resistance.
threats to people’s health or safety.
The ability to make predictions is enhanced if data
Consequently, building systems tend to be checked is collected from more locations, which gives an
regularly, either by employees or by third party advantage to organisations providing maintenance for
maintenance contractors. On the one hand, many multiple sites.
of these visits are unnecessary. On the other, they
often fail to notice faults that lead to failures between A further advantage is that, by identifying the nature of
inspections. The result is disruption, which often the fault, the AI system will often be able to advise what
continues while the problem is investigated, and any action is needed and what replacement parts may be
necessary parts obtained. required and possibly also carry out the action.
This represents an almost ideal opportunity for This reduces the cost and effort of maintenance and
predictive analytics. Devices such as chillers, boilers, also the cost and other risks associated with the failure
lifts or pumps generate large amounts of data about of critical building systems.
their functioning and condition. By collecting this data
over time, an AI-based system can start to identify the
patterns that typically lead to failure.
Smart Connected HVAC with

Predictive Analytics
Perfor-
mance
data
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
Fault IP
report
connec�on
predictive
analytics
Preventative
Support centre or maintenance
automated system

© BSRIA | 25
13 EASE OF CONSTRUCTION
The cost of construction from initial design through to By comparing the actual recorded results to those
its handover represents a major part of the cost of any projected in the original design then accuracy can be
new building. It has often been noticed that productivity further increased in future design projects.
in the construction industry has increased far more
slowly than in other areas of the economy. 13.2 CONSTRUCTION PROJECT
MANAGEMENT
AI has the potential to significantly reduce both the Construction projects for large buildings tend to be very
cost and the time taken to complete major projects as complex and vulnerable to numerous external risks. If
well as reducing risks, from injury to the need to repeat an AI system is provided with data related to the project
work. and also to other projects, this could make it easier to
predict the likely consequences of any delay or other
13.1 DESIGN USING AI problem encountered in the course of construction.
Computer-aided design (CAD) has reduced the need for
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
detailed, painstaking and repetitive manual drawing 13.3 HEALTH AND SAFETY IN
and calculation. AI is bringing this process a further step CONSTRUCTION
forward. While the safety record of the construction industry
has improved enormously, it remains one of the most
AI can experiment with and model different design hazardous occupations. AI can help improve safety in
solutions from the overall building design to detailed construction in various ways. Accidents and incidents
items such as cabling, layout of HVAC systems, across different sites can be recorded and analysed to
safety and security. Accurate estimates of material identify common factors and recommend preventative
and construction costs can be produced along with action.
projections of energy use. These can be further refined
taking account of such issues as position of rooms in AI can even analyse images of construction sites and
relation to the sun. identify signs of risk, such as personnel not wearing the
appropriate protective equipment.
AI can be used to
model designs and
eliminate errors
26 | © BSRIA
14 COMFORT AND WELLBEING

Comfort of building occupants is influenced by AI and machine learning can play a role in learning
factors such as temperature, air quality, humidity, people’s preferences and proactively making
noise levels and the quality and quantity of lighting. adjustments, for example when a person moves
Comfort matters as an end in itself. It also influences into a particular area of the building. If people are
productivity, staff retention and the willingness of assigned desks on a daily basis then people with similar
customers to stay in a building or return to it. preference patterns can be grouped together.
People’s ideas of what makes for a comfortable Buildings are complex systems which interact not just
environment also differ, as anyone who has experienced with the people inside them but also with the external
arguments in an office about opening or closing environment, including such factors as weather,
windows or turning heating up or down will know. pollution and noise levels that vary over time in ways
that are hard to predict. An AI-based system can
In general, the more freedom people have to customise interact with weather patterns and also use weather
their immediate environment within a building the forecast information to adjust blinds, heating and
better. Not only can they adjust their environment, but lighting throughout the day.
simply having the freedom to do so produces a sense
of empowerment. Much of this can be achieved by
providing people with basic tools such as smartphone Buildings interact with their environment
apps that allow them to set heating and lighting and in a complex and changing way...
other services in their vicinity.
Buildings are complex systems

which interact not just with the
people inside them but also with
the external environment, in
ways that are hard to predict.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 27
15 HEALTH AND SAFETY

One of the key functions of buildings is to keep their 15.2 OTHER ONGOING HEALTH AND SAFETY
occupants safe, or at least not to pose any additional RISKS
risks for them. Everyone who has had to pass health and safety tests
at work will be familiar with the issue of obstructions
15.1 FIRE: A LONG-STANDING RISK which people can trip over or which obstruct important
The big risk with which buildings have been associated routes, including even fire doors. If the main hotspots
throughout history is that from fire, and consequently, for such incidents are monitored by CCTV, then it is
fire detection and fire prevention technology was possible to use AI to identify potential hazards, which
one of the earliest applications of automated building can then be investigated and removed.
technology.
15.3 COVID-19 AS A GAME CHANGER
While fire fatalities are much more common in homes, The COVID-19 pandemic has hugely increased the
over the past decade an average of some 20 people per prominence and the scope of health and safety. This
year have died in non-residential fires in England alone. is important enough to be dealt with under its own
section in this paper.
Fire detection is a classic example of balancing
risks. False alarms cause disruption, economic cost
and annoyance and also encourage people to take
future alarms less seriously. On the other hand, the
consequences of failing to identify an actual fire can be AI has a role to play in
catastrophic.
the development of more
Alarm systems, therefore, face the challenge of effective fire alarm systems as
identifying genuine fires while minimising the number of it can use multiple test results
false alarms. AI has a role to play in the development of to model likely responses in
more effective fire alarm systems as it can use multiple
test results to model likely responses in different
different scenarios.
scenarios.
There is likely to be much more caution in deploying AI

in actual fire scenarios because of the perceived risks of
any bad decisions.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
28 | © BSRIA
16 SECURITY
Virtually all buildings have a requirement to restrict A person’s gait has also been shown to be a reliable
access to all or some areas of the building. Technology identifier, even when the person tries to disguise it. AI
has long been deployed in order to achieve this, from enables the defining characteristics of a person’s gait to
movement detectors to CCTV and smart entry systems. be recorded and then to be recognised.
Some buildings have a heightened need for security. There is also the potential to identify more complex
They may be seen as representing a higher risk of aspects of behaviour. We are all familiar with the term
terrorist attack because of their high profile or the “suspicious behaviour”, even if it is hard to define.
number of people that use them. Others may house AI can help to identify suspicious or at least unusual
particularly valuable objects, as with art galleries and behaviour which can then be investigated. This could
museums. range from potential shoplifters to people thought to be
loitering outside a building or in a certain area of it, or
Such places will need to deter a more sophisticated and perhaps returning repeatedly in ways that do not match
motivated type of intruder. In some cases, the intruder’s expected behaviour and which could indicate malicious
presence may be permitted but they may have malicious intent.
intent, for example, to carry out an attack.
It should be clear that we are entering into potentially
AI has the capacity to enhance security in a number of contentious territory here, embracing what has been
ways. A key aspect is quickly and accurately identifying called the “surveillance society”, with the risk of
individuals who are to be either admitted or excluded. impinging on people’s freedom and privacy, and also the
unfair targeting of certain individuals or categories of
Facial recognition focuses on key aspects of an people.
individual’s face to link them to a unique individual.
Given the different ways in which a face can be This in turn takes the discussion back to an important
presented, a degree of AI is needed in order to adapt debate about AI which was touched on earlier in this
these. paper. One claim often made for AI is that it works on
objective data and avoids the biases, both conscious
and unconscious, to which humans are prone.
Gait Recognition can Accurately
Identify Individuals
© BSRIA | 29
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

However, even where the data used to train AI is Thirdly, once a building management system has been
accurate in itself, it is usually selected based on certain compromised it can be used to access other systems
human assumptions as to what is or is not relevant. within a building such as key business systems. This
Whether by design or by accident, this data is quite problem has often been exacerbated by the fact that
likely to reflect some form of human bias. building management systems have formed the weakest
link, especially where those responsible for running
Biases, once identified, can be corrected. Indeed, this them lack access to cybersecurity skills or tools.
is probably easier with AI than with humans, for whom
evidence presented which appears to contradict their The source and nature of cybersecurity threats is
biases can sometimes actually entrench them further. continually evolving, making it hard to predict how
and when they will occur. This makes combatting them
Accordingly, AI is likely to play an increasing role in especially challenging.
building security.
AI can help identify and deal with potential cyber
16.1 CYBERSECURITY threats. Breaches and attempted breaches tend to
Cybersecurity is an important aspect of smart, involve changes in patterns of behaviour. For example, a
connected buildings for three main reasons. building system may start to transmit abnormal volumes
of data and in different structures.
Firstly, cybersecurity breaches that give access to
building systems or their data can threaten the AI is well suited to identifying normal behaviour over
operation of the building’s fire safety and physical a period of time and contrasting this with abnormal
security systems, endangering life. activity. This can then be flagged and investigated
further, typically by human operatives, and action
Secondly, they can cause serious disruption which, while taken as appropriate. This can include isolating suspect
not necessarily causing injury, can prove extremely systems and updating security patches.
costly. This can include interfering with lifts or with
HVAC systems. Raising or lowering the temperature
can make a building effectively unusable. Raising the
temperature in a computer room or data centre can
cause IT systems to crash and, potentially, to become
permanently damaged.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
30 | © BSRIA
17 ENTERTAINMENT
Some buildings focus specifically on entertainment. Robots are also likely to provide entertainment as well
Obvious examples include theatres, cinemas, as perform services in many different types of building.
amusement arcades and buildings in theme parks.
Robots designed to interact with humans tend to
However, many other buildings include at least some entertain, partly because of their anthropomorphic
services intended to entertain. Retail centres seek to appearance. This may lead to their being deployed even
entertain, partly to encourage reluctant shoppers and when less striking building services could carry out the
their children to remain there. Some offices may include same or a similar function.
entertainment facilities, perhaps as a way of improving
employee satisfaction, wellbeing and loyalty.
If a retail centre features a variety of entertainment

facilities then surveillance, linked to an AI system, can
monitor the behaviour of people in different vicinities to
establish which are the most effective.
Robotic butlers, communicating with the

building could provide entertainment as
well as convenience
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 31
18 STATUS AND REPUTATION

Ever since humans began to live in fixed settlements, Building owners most likely to wish to use AI to impress
buildings have often been associated with confirming are those associated with advanced information
the status or credentials of those occupying them. technology. Major corporations such as Apple, Google
Private homes, the buildings of rulers and governments, and Microsoft use their global headquarters to
the temples of religion and, today, the offices of showcase a vision of how they see their role as leaders
corporations all tell us something about how the in an ongoing technological and social revolution.
occupants and users see themselves and wish to be
seen.
Much of the drive for status and prestige has, for

fairly obvious reasons, concentrated on what is most
visible. AI by its very nature tends to be less visible, less
tangible, operating mostly behind the scenes as the
“brains” of a building.
Google’s proposed new HQ at

Mountain View, California
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
32 | © BSRIA
19 OTHER SPECIALISED BUILDING

REQUIREMENTS
Almost every industry segment is likely to have buildings Various gambits have long been deployed to influence
with distinctive requirements related to their use. The the mood and behaviour of shoppers. One long-
following example relates to retail. standing example is the use of “muzak” to set a mood
most conducive to spending. With AI, the behaviour
Buildings have long played an important role in of shoppers can be analysed in response to different
retail. While this is most obviously reflected in the tracks. The retailer really can discover whether their
imposing architecture and décor of some major stores, shoppers are more responsive to Mozart, Mantovani or
technology has also been used especially for security, Meatloaf.
comfort and to help people navigate the premises.
Similar behaviour analysis can be carried out to reflect
COVID-19 has hit physical retail stores especially hard, changing displays or lighting schemes. If the store has
accelerating the long-term shift towards online retail a means of tracking mobile phones or other mobile or
that was already well underway. If the notion of what wearable devices, and the entitlement to do so, then
is loosely called “the high street” is to survive then this can provide additional intelligence.
there will be a need for changes to make the process of
shopping more engaging. While theoretically powerful, such tools raise both
practical and ethical issues. Would a technology that
AI has a particular role to play in helping store and retail is powerful enough to collect and analyse images that
centre managers understand the behaviour of people in point to people’s emotional states be acceptable? At
and around their stores. present that seems to be unlikely.
Facial mood recognition can provide

information about the reactions of visitors
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 33
20 AI AND PANDEMICS
The COVID-19 pandemic, starting in late 2019, was the At present (June 2021) it is unclear when the COVID 19
most serious pandemic to sweep the world since the pandemic will be brought under control. This is likely to
influenza pandemic of almost exactly a century earlier. vary considerably from one country to another. It is also
unclear whether there will be major recurrences (for
One factor which became apparent at an early stage example resulting from new virus mutations). There are
was that the virus was much more likely to spread inside also a number of other viruses which have come close
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
buildings than in the open air. Consequently, much to causing global pandemics, including SARS and Ebola.
of the impetus to control and suppress the virus has They and others could cause major outbreaks.
focused on controlling and limiting access to buildings,
especially those where people tend to gather in large The building services world cannot assume that major
numbers and in close proximity. pandemics will remain a once-in-a-century occurrence.
This is likely to lead to continuing measures to make
Given that so much of human life in modern developed buildings safer in the face of such health threats. AI
societies takes place in buildings, including work, retail can play a significant role in this. The fact that different
and leisure activities, this has had serious economic pandemics may spread in different ways to COVID-19
and social consequences. If the world is to return to also implies scope for modelling different routes for
anything resembling “normal” conditions following the contagion within buildings.
pandemic, then people need to be persuaded that it is
safe for them and for others to return to buildings.
AI can combat infection

by improving air quality
Things the virus loves.... The solu�on Sensors monitor air status
throughout building....
High CO2
High occupancy Low humidity
Poor ven�la�on Can facilitate the
aerosol effect that AI-based
spreads the virus building management
High ar�culate
ma�er
Viruses can
a�ach
Ven�la�on and humidity
delivered to zones where most
needed
34 | © BSRIA
20.1 SAFER SERVICE AND MAINTENANCE 20.2 REDUCING TRANSMISSION THROUGH

Even during lockdowns, buildings have to be serviced IMPROVED INDOOR AIR QUALITY
and maintained. Personnel and staff carrying out this COVID-19, like many viruses, can spread rapidly through
work are potentially exposed to risks both on site and airborne transmission within buildings. However, certain
travelling to and from the site. Remote monitoring and specific conditions appear to be particularly conducive
predictive maintenance brings benefits in terms of to the spread of the virus.
reduced costs and downtimes which have already been
described. However, it also helps to keep service and High levels of CO2 suggest a concentration of people
maintenance staff safe. within an area combined with poor ventilation, a
situation conducive to the spread of COVID-19. The
With this solution, service staff will normally receive virus also survives longer and spreads more readily in
information about the nature of any problem, what conditions of low humidity. High levels of particulate
action they need to take and what, if any, parts are matter in the air are not only bad in themselves but also
likely to need replacement. The better the AI system, help the virus to distribute itself.
the more accurate this information is likely to be.
All of these things can be measured by existing sensor
This should reduce the frequency of site visits ‒ only technology. By feeding this data into an intelligent
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
one visit will be necessary where before an initial visit building management system it is possible to adjust
to diagnose the problem was often required. Staff will levels of ventilation and humidity dynamically. The more
arrive on site with an understanding of the problem, granular the HVAC system the more precisely conditions
they will be carrying the requisite parts, and will need to can be controlled.
spend less time troubleshooting.
Smart wearables can

relay information
about physical states
to AI systems

© BSRIA | 35
Mixed Reality allows

people to view 3D
objects in different
locations
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
AI can enhance this process, for example by predicting While this can address problems as they occur, if the
when unsafe conditions are likely to occur and taking information is collected and analysed over time it
action before they do. may be possible to change the use of the building, for
example by altering the flow of people in the building or
Remote monitoring linked to AI can also help maintain adjusting working hours.
air quality, by identifying problems caused by blocked
filters which need changing, or other malfunctioning Widespread use of CCTV within buildings for non-
equipment. security purposes is also likely to raise privacy concerns.
AI greatly increases the power of CCTV to help address
20.3 USING CCTV ANALYSIS TO MONITOR safety and security concerns but it also provides tools
SAFETY which can be used to put people’s lives under much
CCTV systems linked to AI also have the potential to closer scrutiny in ways that they may not always
address some problems. Fairly basic AI can identify welcome.
people not wearing face masks when they are supposed
to do so and warn either them or the building operators.
Images, perhaps combined with other sensors, can also
identify unsafe gatherings of people in areas of the
buildings such as the classic pinch-points of staircases or
kitchens.
36 | © BSRIA
20.4 USE OF MOBILE AND WEARABLE 20.5 ADAPTING OFFICE WORK TO A POST-
DEVICES COVID WORLD
Mobile devices such as smartphones and wearable It is still far from clear what the “new normal” for
devices such as smart watches have a potential role to human interactions and work patterns is likely to
play in combatting viruses like COVID-19. be. However, it is increasingly seen as likely that the
traditional pattern of commuting to a fixed place of
Both can potentially communicate the location of work five days a week will, in many cases, be replaced
the wearer within a building. This can identify unsafe with a more mixed pattern of work. Business travel,
concentrations of people. Users can also be alerted if including face-to-face meetings, became all but
they are closer than recommended – for example within impossible in 2020 in many parts of the world and looks
two metres of another person. The “safe” distance unlikely to recover to previous levels soon, if ever.
can be dynamically adjusted as the situation changes
or knowledge of conditions changes. In the event of While basic web conferencing services have proved an
an individual becoming infected, the data can identify invaluable tool, most users have also become aware of
those people with whom they have been in close their limitations. There are, however, more powerful
contact so that they can take appropriate precautions, technologies becoming available to address this.
such as isolating and being tested.
One example is mixed reality. This can present objects
Some smart wearables also collect information about in the form of three-dimensional holograms, which can
the wearer’s physical health. This could include then be shared and examined by people in different
information about symptoms that suggest possible locations. This could make remote conferencing feel
infection. By analysing data on the physical state of closer to the experience of sharing real objects in the
people subsequently shown to have the virus and same room. This type of technology is likely to prove
comparing this with other cases it may even be possible especially relevant in disciplines like building services,
to identify new pointers to the virus. This is a classic where discussions frequently centre on complex three-
application of AI and is likely to be particularly useful for dimensional objects and systems.
a disease like COVID-19 where the physical symptoms
appear to vary so much from one individual to the next.
Even more so than CCTV, use of wearables to collect

and analyse information on the wearer’s health is likely
to provoke concerns about privacy and consent. This is
particularly true in that such processes are likely to be
effective only if all or most people in the building wear
active devices. An employer may find it hard to justify
obliging employees to comply, either legally or ethically.
Enforcement would be even harder if the owner of,
say, a theatre or a restaurant tried to “require” such
monitoring of their customers.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

© BSRIA | 37
21 THE IMPORTANCE OF A
HOLISTIC APPROACH
One thing that should be apparent from this white be coordinating well with other departments and may
paper, if it was not already, is the extent to which all even be unwittingly working against them.
aspects of a building tend to be interlinked. Some
building goals, such as wellbeing, comfort, safety and Similarly, to find the optimal balance, a building system
security, while each being distinctive, tend to work in needs to be able to access data about as many aspects
harmony. Lighting, HVAC and devices such as blinds all of the building as possible, establish how they interact
affect the indoor temperature. There will be exceptions and also assign a value or weighting to them. This
to this synergy. For example, increased ventilation to weighting may change over time. For example, during a
reduce the risk of infection may make some building pandemic, measures to reduce the risk of infection are
occupants feel less comfortable unless countervailing likely to assume a higher priority relative to comfort or
measures are taken, such as heat exchange. convenience.
In other cases, there may be a conflict or trade-off to be AI provides a means of monitoring all of these
had. For example, between energy efficiency and the interactions and even, potentially, identifying new
maintenance of comfort. Similarly, systems that make a ones. The movement towards AI and to the integration
building more secure may also prove less convenient for and convergence of building systems thus go hand in
people moving about quite legitimately. hand. BSRIA’s global research has identified a long-term
movement towards convergence of building systems. In
The problem is in many ways analogous to the oft a converged building, different building services share
repeated cliché of people within an organisation information and may in many cases have an overarching
working “in silos”. Each may be acting intelligently based control system which takes decisions or delegates as
on the information and resources available to them, but appropriate.
they are missing a lot of the bigger picture. They will not
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
38 | © BSRIA
To date, the cost and complexity of converged solutions controllers manage non-HVAC functions. Europe is in
has meant that convergence tends to occur most often second place with 28%. This suggests that these two
in larger and more prestigious projects. The following regions of the world are seeing the most convergence,
figure shows that, globally, some 72% of controllers though the Asia Pacific and the Middle East and Africa
are allocated to HVAC, which has long been the core are not far behind.
competency for building automation systems.
As convergence increases, an AI-based building
However, 13% of controllers address security, 8% management system has access to more information
lighting, 4% fire safety and 3% some other building about the state of the whole building, which it can
service. The share of controllers linked to non-HVAC analyse over time, to coordinate and fine-tune all of the
services is highest in the Americas, where 36% of building’s services.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
Building automation controllers by application and region, 2020
Source: BSRIA –Building Automation Controls World Overview 2020
World 72% 8% 13% 4% 3%
Middle East
/ Africa 78% 10% 8% 5%
Europe 72% 9% 15% 2% 1%
Asia / Pacific 76% 2% 9% 6% 7%
Americas 64% 7% 19% 7% 3%
HVAC Lighting Security Fire Other

© BSRIA | 39
22 AI AND BUILDING SERVICES:

THE CULTURAL DIVIDE
There is already a problem with building services which Suppliers of building systems need to bring together
results from the fact that knowledge and understanding AI expertise and building expertise. Currently, leading
about how buildings are designed, constructed and suppliers of building systems are trying to acquire
operated tends to be found in different organisations greater AI know-how, while some of the leaders in AI
to those with in-depth understanding of information are attempting to enter the building services market.
technology.
So far, the focus of AI leaders, which includes such
Those with expertise in installing and supporting corporate giants as Google, Amazon and Apple has been
buildings frequently struggle when it comes to issues mainly on the residential market. This has the attraction
like cybersecurity or the most effective way to configure of being considerably less complex (if one compares
a complex building management system. Conversely, the demands of a typical home with a typical office
those who understand information technology may block or retail centre). However, commercial buildings
have scant understanding of the complexity and quirks offer potentially higher margins than the highly price-
of physical buildings and their support systems. conscious residential market. The greater complexity
of most commercial buildings also means that there is
AI can accentuate this divide further, but it can also, more scope for AI to achieve its full potential in helping
in some cases at least, help to overcome it. Systems to address complex problems.
using AI tend in general to require less oversight and
supervision. In many cases they can use machine The advance of AI is likely to see both competition
learning to work out the best response to a problem, and alliances between suppliers with established
without having to call on human expertise. In this strengths in commercial buildings on the one hand and
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
scenario the need for human skill may actually be AI expertise on the other. Past experience suggests
reduced in some cases. that technology changes have the potential to be
momentous enough to change the supplier landscape.
However, with AI still in its early stages, significant input
from human experts in order to configure and guide Building operations and facilities management are also
the system and also to check up on it is likely to remain likely to need to acquire new skills in order to make the
necessary. This presents a challenge for suppliers, most of the new technology.
installers and for building operations.
40 | © BSRIA
23 AI WILL REQUIRE ITS

OWN STANDARDS
When one discusses standards for artificial intelligence, 23.2 STANDARDS FOR THE PROCESSING
one is alluding to several distinct types of standard, AND TRANSFER OF DATA
all of which matter, whether directly or indirectly. As should be clear by now, AI is heavily dependent on
Some standards are enforced by regulation. Other access to large volumes of data. Systems have been
standards emerge as the result of commercial pressures developed to handle “big data”. Big data implies not
encouraging most users to adopt the same standard in just very large volumes of data, but data that is rapidly
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
order to sell and support their products or services. changing and which is presented in many different
forms and formats. This is encapsulated in the “Three
Given that AI is still in its early stages, it is not surprising Vs” of big data: volume, variety and velocity.
that there are few standards specifically relating to AI.
Indeed, there is a clear problem when it comes to any However, just as humans absorb information more
legally enforced standards for advanced technology. The easily when they are presented with a structured report
technology is hard enough for lawmakers to understand rather than simply being given access to a library, so AI
to legislate in an informed way. It is also evolving so fast can process information more quickly and accurately
that it is difficult for legislation to keep up. Legislators when there is some consistency to the structure and the
and those responsible for standards are likely to have to taxonomy of the data presented to it.
rely more on the trust that they place in subject matter
experts. In building services, there has been a long-term trend
away from proprietary protocols towards standardised
23.1 STANDARDS THAT ENCOURAGE ones. In the case of larger buildings, BACnet has
DEPLOYMENT OF AI IN BUILDINGS emerged as the de-facto standard. BSRIA research
As mentioned in section 7.2, legislation such as the EU’s indicated that in 2020 it was used in 55% of new
Energy Performance of Buildings Directive encourage building automation installations, with a much higher
the deployment of building automation systems as share in many countries.
a means of improving energy performance. Also, as
mentioned in section 7.2, the SPIRE standard evaluates However, many devices, such as air conditioning units,
smart buildings against six “dimensions”. make more detailed data available in a specialised and
proprietary protocol than can be interpreted by more
open protocols like BACnet. To some extent, this is a
question of priorities. A protocol that deals only with a
particular type of chiller can go into more detail than
a protocol like BACnet which is used across building
services.

© BSRIA | 41
Proprietary protocols also give manufacturers a 23.4 STANDARDS SPECIFIC TO ARTIFICIAL

competitive advantage when it comes to offering INTELLIGENCE
services, including intelligent predictive analytics of The International Standards Organisation (ISO) is
their device’s performance. A further consequence currently working on the development of standards
is that not all relevant information may be shared to cover various aspects of Artificial Intelligence. The
between systems. This is especially the case, given that process was initiated in 2021, and those currently under
it is not always possible to predict what information will development are listed opposite.
actually turn out to be relevant.
Most of these are still at a draft stage or in preparation.
Standards such as Project Haystack set up a common However, ISO/IEC TR 24030 Information technology
taxonomy for building solutions which enable different – Artificial intelligence (AI) – Use cases is due to be
building systems to have a common understanding published in May 2021. The breadths of specific aspects
of what a term means. It functions rather like a covered is a further indication of the likely scope and
multilingual translation dictionary. impact of AI.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
23.3 STANDARDS FOR DATA PRIVACY AND As AI develops and gains a stronger foothold it is
DATA PROTECTION inevitable that both standards and regulations will
In Europe in particular there are many concerns need to develop further in order to ensure that the
about data protection and data privacy. These relate undoubted benefits are balanced by proper controls.
especially to personal data, with measures such as the
European Union’s General Data Protection Regulation
(GDPR), which imposes strict controls on when and how
personal data can be stored, used and accessed.
This creates a potential conflict, as AI-based building

systems can benefit from information about the
location, preferences and state of health of individuals
when it comes to goals such as comfort and wellbeing.
While this can be in some cases addressed by
anonymisation of personal data, this is likely to become
an increasingly contentious issue as AI becomes more
powerful.
42 | © BSRIA
International standards under development for artificial intelligence

Information technology – Artificial Intelligence – Assessment of machine learning
ISO/IEC WD TS 4213
classification performance
ISO/IEC WD 5259-1 Data quality for analytics and ML – Part 1: Overview, terminology, and examples
ISO/IEC AWI 5259-2 Data quality for analytics and ML – Part 2: Part 2: Data quality measures
Data quality for analytics and ML – Part 3: Data quality management
ISO/IEC WD 5259-3
requirements and guidelines
ISO/IEC WD 5259-4 Data quality for analytics and ML – Part 4: Data quality process framework
ISO/IEC WD 5338 Information technology – Artificial intelligence – AI system life cycle processes
ISO/IEC WD 5339 Information Technology – Artificial Intelligence – Guidelines for AI applications
Information technology – Artificial intelligence – Reference architecture of
ISO/IEC WD 5392
knowledge engineering
ISO/IEC AWI TR 5469 Artificial intelligence – Functional safety and AI systems
Information technology – Artificial intelligence – Objectives and methods for
ISO/IEC AWI TS 6254
explainability of ML models and AI systems
Information technology – Artificial intelligence – Artificial intelligence concepts
ISO/IEC DIS 22989
and terminology
ISO/IEC DIS 23053 Framework for Artificial Intelligence (AI) Systems Using Machine Learning (ML)
ISO/IEC CD 23894 Information Technology – Artificial Intelligence – Risk Management
Information technology – Artificial Intelligence (AI) – Bias in AI systems and AI
ISO/IEC DTR 24027
aided decision making
Artificial intelligence (AI) – Assessment of the robustness of neural networks – Part
ISO/IEC AWI 24029-2
2: Methodology for the use of formal methods
ISO/IEC TR 24030 Information technology – Artificial intelligence (AI) – Use cases
Information technology – Artificial intelligence – Overview of ethical and societal
ISO/IEC AWI TR 24368
concerns
Information technology – Artificial intelligence (AI) – Overview of computational
ISO/IEC DTR 24372
approaches for AI systems
Information technology – Artificial intelligence – Process management framework
ISO/IEC CD 24668
for Big data analytics
Software engineering – Systems and software Quality Requirements and
ISO/IEC AWI 25059
Evaluation (SQuaRE) – Quality model for AI-based systems
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
Information technology – Governance of IT – Governance implications of the use

ISO/IEC DIS 38507
of artificial intelligence by organizations
ISO/IEC AWI 42001 Information Technology – Artificial intelligence – Management system

© BSRIA | 43
24 CONCLUSIONS
Artificial intelligence is still at a comparatively early AI will also present big challenges to everyone operating
stage when it comes to buildings and most other in or with building services. On the one hand, it will
areas. However, as this paper has set out to show, even increase the already growing importance of knowledge
the relatively weak forms of AI available today have of software and IT processes. On the other, particularly
practical applications which are relevant to virtually all for more basic aspects of building services, AI may help
of the major aims that building developers, designers, out by being to a degree self-configuring and self-
owners, managers and users have for their buildings. managing.
However, while AI can certainly help when it comes to AI has for a while been seen as posing a potential threat
pursuing specific goals such as comfort, safety, security to employment in building services. However, the
or energy efficiency, where it really comes into its own comparatively weak form of AI that is currently available
is in evaluating the different ways that all aspects of a means that, for the time being, the threat will be mainly
building interact with each other, whether intentionally to lower-skilled positions. The longer-term implications
or unintentionally. are less clear.
This in turn can help the building stakeholders advance AI will accentuate and accelerate the long-term shift
from simply doing things more efficiently, to finding away from building services that are “siloed” into
new ways of doing things or even new aims for their particular product and function areas, towards one
buildings and the way that they are designed, built and which favours those who understand how AI can be
used. deployed to make all aspects of the building function
more effectively. The process goes hand in hand with
The successful development of AI and the avoidance of the move already being seen towards convergence of
potential pitfalls depends on a number of key factors. different building services to one managed by a single,
Data needs to be available from as wide a range of networked building system
sources as possible, preferably in a format that can
easily be interpreted. This requires both fast, robust This could well lead to far reaching changes in the
communications and accepted standards. industry as the comparative advantage resulting from
different skills changes.
Much of the data that could potentially be fed into AI
to improve building functionality is highly sensitive. This It is hoped that this paper will encourage everyone in
applies particularly to personal information about the any area of building services to look more closely at how
building’s occupants. AI may well improve their comfort, AI is almost certainly going to change reality in their
productivity, satisfaction or safety, but that raises the particular area, what products and services they offer
ethical question of whether they should be required or and how they work with others.
encouraged to share it.
Artificial Intelligence is already part of building services
A related question is how AI in buildings should be and is certain to become even more central to it.
regulated and standards enforced both to ensure that The choice for everyone in the industry is whether to
it functions reliably and safely and that there are no take steps to understand it better and to work with it
abuses. A key problem here is that with AI rapidly proactively, or whether to become its passive recipient.
evolving into something ever more complex it will be
increasingly difficult for regulators to understand it, let
alone control it.
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
44 | © BSRIA
--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

Engineering services: Test, compliance and consultancy support services for
clients, designers, constructors, manufacturers and facilities managers
Instrument solutions: Hire, sales and calibration of instruments
Market intelligence: Off the shelf reports, bespoke studies and management
consultancy for global markets
Information and membership: Bookshop, library, training, networks and events --``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---
Head Office: Old Bracknell Lane West, Bracknell, Berkshire RG12 7AH UK
T: 0800 254 5700 (UK) or +44 (0) 1344 465 600 | F: +44 (0) 1344 465 626
E: bsria@bsria.co.uk | W: www.bsria.com/uk/


y Bb f Vga Aaaaaaaaa

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

y Bb f Vga Aaaaaaaaa

Uploaded by

Copyright:

Available Formats

--``,`,,``,```,```,,,````,``,,,`-`-`,,`,,`,`,,`---

A BSRIA White Paper

Copyright BSRIA Limited

Date: June 2021

Issued by: BSRIA Limited

Feedback on BSRIA publications is always gratefully

3 WHAT IS ARTIFICIAL INTELLIGENCE?................................................................................... 6

4 WEAK AND STRONG ARTIFICIAL INTELLIGENCE............................................................... 9

5 WHEN IS A BUILDING SMART OR INTELLIGENT?.............................................................. 13

6 THE KEY AIMS OF BUILDINGS.............................................................................................. 16

7 REGULATIONS AND STANDARDS SET THE SCENE FOR BUILDINGS........................... 17

10 ON-SITE RENEWABLE ENERGY........................................................................................... 22

12 OPERATIONAL AND MAINTENANCE EFFICIENCY............................................................ 25

14 COMFORT AND WELLBEING................................................................................................. 27

15 HEALTH AND SAFETY............................................................................................................. 28

18 STATUS AND REPUTATION.................................................................................................... 32

19 OTHER SPECIALISED BUILDING REQUIREMENTS.......................................................... 33

21 THE IMPORTANCE OF A HOLISTIC APPROACH................................................................. 38

22 AI AND BUILDING SERVICES: THE CULTURAL DIVIDE..................................................... 40

23 AI WILL REQUIRE ITS OWN STANDARDS........................................................................... 41

Copyright BSRIA Limited

2.3 THE IMPACT OF COVID-19

Copyright BSRIA Limited

McCarthy defined intelligence thus: “Intelligence is the

3.2 MORE THAN JUST FOLLOWING Howard Gardner's nine

Intelligence implies an ability, however limited, to

3.3 INTELLIGENCE IS A MATTER OF DEGREE

3.4 INTELLIGENCE IS OFTEN TASK-FOCUSED

Copyright BSRIA Limited

At the same time, there has been a similar downward

Copyright BSRIA Limited

4 WEAK AND STRONG ARTIFICIAL

Other examples of weak AI in building services would

Weak AI may well perform its specific tasks much more

THE LIMITATIONS OF WEAK AI

Copyright BSRIA Limited

4.3 THE SINGULARITY DIVERSITY OF DATA SOURCES AND DATA SHARING

Humans, in contrast, often cope with their own MACHINE LEARNING

A neural network can help an AI system learn what the

Copyright BSRIA Limited

COLLABORATION WITH HUMANS or university background have tended to be successful

hypotheses, and also identify relationships and patterns than in others.

5 WHEN IS A BUILDING SMART

10 Drawing from the wider world Strong

9 Iden�fying new goals

5 Finding excep�ons / anomalies Weak

Copyright BSRIA Limited

Copyright BSRIA Limited

6 THE KEY AIMS OF BUILDINGS

7 REGULATIONS AND STANDARDS

In a rapidly advancing field, there is also a risk that any

Regulations and standards regarding buildings and their

Regulations generally include penalties for non-

Standards range from those which are incorporated into

Copyright BSRIA Limited

Provided by Accuris Licensee=LEEDS BECKETT UNIVERSITY/7100928001, User=Leyton Diaz, Carlos

Copyright BSRIA Limited

Where a smarter building

Using AI to Identify and Correct

Energy use over time Energy use adjusted

Device A Heating Heating

Comfortable conditions Comfortable conditions

Abnormal energy Normal energy use

Copyright BSRIA Limited