MMPA MGT 2090

A CPA’S INTRODUCTION TO AI
– WHAT YOU NEED TO KNOW*

*From A CPA’s Introduction to AI: From Algorithms to Deep Learning, What You Need to
Know, published by CPA Canada and AICPA, 2019
 OVERVIEW
1. Introduction.

2. What is AI?

3. Why CPAs Need to Stay Current with AI.

4. How did we get here? What does it mean for business?

5. How does AI work? How is it being used?

6. What’s next?
INTRODUCTION
Right now...

• a smartphone app called Climate Basic is telling farmers where to

plant crops based on local temps, expected rainfall, soil quality etc.

• Someone is sitting on their couch scrolling through Netflix’s

recommendations unique to them generated by an algorithm.

• Manual accounts payable tasks are being replaced by systems that

match purchase orders & flag invoices for payment, then categorize
the expense using Optical Character Recognition & Machine
Learning algorithms.
INTRODUCTION
• We have virtual assistants (Siri), navigation apps (Waze),
autonomous cars, humanoid robot bank ambassadors &
hospital receptionists, Google Assistant makes
appointments with your hairdresser.

• AI is everywhere!

• And, we’re still just scratching the surface of how it will

change how we live & work!
WHAT IS AI, REALLY?
Here’s the short answer & broadest definition:

AI is the science of teaching

programs & machines to complete
tasks that normally require human
intelligence.
 THE NEW SPACE RACE: GLOBAL INITIATIVES
TO WIN AT AI
• $15.2 Billion invested globally in AI start-ups in 2017 (48% in China &
38% in the US).
• 2018, US Defense Advanced Research Projects Agency announced
$2B campaign to develop next wave AI focusing on contextual
reasoning capabilities.

• 2018, EU launched a $1.8B investment in AI research, France $1.8B

on AI & data sharing, UK 300M investment brining 50 leading
businesses to set a 1B deal in AI.

• 2017, Canada invested $125M to develop an AI strategy and is now

investing $950M in the Innovation Super Cluster initiative that
includes SCALE.AI (Supply Chains & Logistics Excellence.AI) to
shape a new AI-powered global supply chain platform.
 AI FUELING INVESTMENTS & RESEARCH
• AI-related research will grow from $12B in 2017 to close
to $58B in 2021.

• Since 2013, machine-learning-based research has been

the 3rd fastest growing category for patent filing.

• On the job front, employment search engine Indeed

reports the share of jobs in Canada and US requiring AI
skill has grown significantly since 2013.
WHY IS IT IMPORTANT FOR CPAS TO
STAY CURRENT WITH AI?
• Simply put, AI has the potential to significantly impact accounting &
assurance jobs & businesses in the near future – if it hasn’t already.

• Transactional tasks are increasingly being automated.

• Algorithms can spot subtle clues in data sets to pinpoint fraud,

identify opportunities for savings, improved efficiency & profitability.

• All while leveraging & making sense of large volumes of new

information coming from multiple sources (Big Data).
WHY IS IT IMPORTANT FOR CPAS TO
STAY CURRENT WITH AI?
1. Everyone has an opinion about AI.

2. Positive: how it will improve everything from transportation to

medical diagnostics.

3. Negative: the potential elimination of jobs and violation of privacy.

4. Ethical: how does AI resolve self-driving car’s dilemma on whether

to crash into an oncoming car or a pedestrian when an accident is
unavoidable.

5. Question of decision transparency. Will AI systems be able to

explain in understandable terms, the reasons for its decisions?
WHY IS IT IMPORTANT FOR CPAS TO
STAY CURRENT WITH AI?

• CPAs have the opportunity to focus on AI as an enabler

of innovation & increased productivity.

• Consider the Climate Basic app helping farmers.

• This use of AI across the industry has produced the

largest crops in the country’s history.
WHY IS IT IMPORTANT FOR CPAS TO
STAY CURRENT WITH AI?
• Netflix says its sophisticated recommendation engine
has saved the company $1B in lost subscribers.

• CPAs have the opportunity to improve efficiencies &

quality of day-to-day work through AI that reads
contracts, performs automated tasks (e.g. accurately
identifying and booking expenses), or identifying risky
transactions in real time.
OVERVIEW OF TERMINOLOGY
HOW WE GOT HERE
• It all started with Data.

• Historically, companies used data to improve

performance (e.g. customer surveys).

• However, data gathering was an explicit experience.

• And, steps weren’t always taken to verify data collected

represented what was actually happening.
HOW WE GOT HERE
• Difficult to determine whether survey respondents truly represented
their customer base (maybe respondents were far different).

• The internet & evolution to a digital world have changed data

acquisition from an explicit, often manual process, to an ongoing,
automated passive activity.

• Websites, apps & social media can track everything from your
shopping behavior to your heart rate.

• Many business processes are being digitized leading to the

collection of info to help reduce costs.

• Real data in real time with no surveys & response bias, just facts.
FROM DISTRIBUTED TO CLOUD COMPUTING
• Passive data collection has made data sets so massive & complex that
traditional software just can’t keep up.

• So companies like Google, whose business is data, began looking for

ways to increase computational power in order to process Big Data.

• The answer: Distributed Computing – moved from having one

sophisticated computer handling all data to 100s or 1,000s of
commodity computers working in concert but independently cutting
down overall processing time.

• That solved the problem for large technology companies but, what
about the smaller organizations that still had no way to access &
process large volumes of data effectively.
FROM DISTRIBUTED TO CLOUD COMPUTING
• Enter Cloud Computing – a rentable network of remote
servers hosted on the internet to store, manage and
process data.

• Scalability of Cloud Computing eliminated the need to

buy and maintain costly computer systems.

• It provides access to more on-demand resources &

processing power for everyone…as a result, all
organizations can now benefit from Big Data.
BIG DATA MEETS AI
• Basic idea behind AI is to let the machine statistically
analyze all data being collected to drive insights much
faster & more accurately than otherwise possible.

• So, rather than a simple rule like >12 hrs. of daylight =

summer; <12 hrs. = winter, AI analyzes the parameters of
numerous summer days and uses algorithmic models, to
determine which parameters and at what levels those
parameters constitute a summer day (e.g. temp > 18
degrees, humidity of x%, etc.).
BIG DATA MEETS AI
• It will then leverage that understanding & based on
probabilities determine whether a particular day is a
summer day or not by assessing all those relevant
parameters against what is has ‘learned’ to be a typical
summer day.

• Experts agree that AI has now evolved to the point where

it has a real, practical value.
BIG DATA MEETS AI
AI has evolved to having real, practical value because of
the convergence of four trends:

1. Digitization of business processes, smart phones, Internet

of Things (IoT) and social media/web 2.0.
2. The advancement (and availability) in processing power
to store & compute data.
3. The maturity of algorithms & AI models.
4. The huge upsurge in investment in AI.
DATA WAREHOUSES, LAKES & SWAMPS
• How is all this data stored?

• Typically, via a Data Warehouse or Data Lake.

• Data Warehouses are designed to store data and provide

businesses a central repository to integrate, manage &
analyze data at many levels.

• Data can come from internal and/or external sources

and is largely structured.
DATA WAREHOUSES, LAKES & SWAMPS
DATA WAREHOUSES, LAKES & SWAMPS
• Data Lakes are basically a way to both organize huge
volumes of widely diverse data, structured & most often
unstructured, and store it.

• When a Data Lake becomes unmanageable, it turns into

a Data Swamp.
DATA WAREHOUSES, LAKES & SWAMPS
DATA WAREHOUSES, LAKES & SWAMPS
WHAT DOES THIS ALL MEAN FOR BUSINESS
TODAY?
• Two concepts: improved productivity & innovation.

• Includes simplification and automation of tasks that free up

employees to focus on more stimulating & higher value-
adding work.

• New innovative business models have been established as a

result of AI (e.g. Alexi, Google assistant or x.ai).

• Although not popular before AI, there is now a multi-billion $

industry at the centre of the home automation market.
WHAT DOES THIS ALL MEAN FOR BUSINESS
TODAY?
• Capital intensive sectors, such as manufacturing &
transportation, likely to see largest productivity gains from AI,
given many repetitive operational processes ripe for AI-
enabled automation.

• AI robots are already taking over warehouses across the

globe.

• AI poised to unlock efficiencies in professional services, such

as legal, HR, accounting/audit where data plays central role.
GLOBAL ACCOUNTING FIRMS & AI
• BDO successfully piloted BDOLexi translation app, using
neural network technology, to manage info in multiple
languages during global audits.

• Deloitte launched Omnia AI, a multi-disciplinary practice

focused on delivery of AI services & solutions.

• EY embedding AI technology across its end-to-end audit

process, including using machine learning in EY Helix & EY
Optix for predictive analytics, using drones for inventory
observations & the development of business document-
reading & interpretation tools.
GLOBAL ACCOUNTING FIRMS & AI
• KPMG launched its AI toolkit called Ignite & is working
with IBM/Watson.

• PwC is investing in data platforms to securely capture,

organize & facilitate analysis of data & is working with
H2O.ai, a leading Silicon Valley company, to develop
Audit.ai, which will provide enhanced capabilities to
provide assurance.
TYPES OF AI
There are two types: Narrow AI & General AI.

1. Narrow AI, also known as ‘weak’ AI, is made up

of narrowly intelligent systems that can exceed
humans in specific tasks (e.g. playing chess &
making medical diagnoses).
These narrow capabilities are not transferrable
(i.e. an AI chess player can’t make a medical
diagnosis).
TYPES OF AI
2. General AI or strong AI refers to human-level
intelligence that is able to transfer knowledge
between domains.
While narrow AI is all around us in language and
vision recognition systems & recommendation
engines, General AI is the stuff of science fiction
& movies…for now.

•
HOW AI WORKS
• For AI programs to navigate through situational complexities,
different approaches on creating software, with the ability to
determine different outcomes, are necessary.

• The Logic & Rules-based approach uses conditional

instructions and defined rules to carry out a task or solve a
problem (e.g. ‘if this, then that’).

• It has been in practice for a long time & has been the
underlying premise for AI until recent advances in machine
learning & deep learning, which are techniques within AI.
HOW AI WORKS
• E.G. to create systems to calculate tax through
a digitized decision-making process, the Logic
& Rules approach employs a detailed series of
computer instructions, including ‘if, then, else’
or probabilities/weights.

• These detailed series of instructions at the heart

of any AI system are known as algorithms.
HOW AI WORKS
• Machine Learning is the more prevalent AI
approach using algorithms to guide its
predictions.

• Its name comes from the ability of its

algorithms to ‘learn’ from experience (i.e.
use data sets) rather than relying on a
rules-based system.
HOW AI WORKS
• The algorithms create computational models
that process large data sets to predict
outcomes & make inferences.

• More data, more examples, help algorithms to

finely tune its output over time.

• The algorithm adjusts – learns – by trial & error.

HOW AI WORKS
• Each novel example is an opportunity for the algorithm
to guess correctly or incorrectly & learn from its
mistakes thus usually providing better insights.

• The insights are then fed back to refine the algorithmic

models further & make them more accurate over time.

• Examples of current ML use – email spam filters, text,

image & voice recognition systems, search rankings,
spell check, etc.
A CLOSER LOOK AT MACHINE LEARNING
• ML models can be grouped into two types of
problems: regression & classification.
• Regression problems involve prediction of a
quality or value (e.g. taxi fare to go from Pearson
to downtown Toronto during rush hour).

• Classification problems involve classifying input

(e.g. determining whether a transaction is typical
or anomalous).
A CLOSER LOOK AT MACHINE LEARNING

3 different techniques are commonly used

for a machine to ‘learn’ the problem and
become smart at providing the answer…
A CLOSER LOOK AT MACHINE LEARNING
1. Supervised Learning is the method to teach AI
systems by example. Many points of data
provided, each point tied to expected
outcomes so that an understanding can start
to develop of how data relates to expected
outcomes.
Once trained, systems can ingest points of
data & give an output that fits the learned
model.
A CLOSER LOOK AT MACHINE LEARNING

2. Unsupervised Learning – the system is

told which loans were repaid
successfully and which ones were not
and asked to identify what trends most
affected the outcome.
A CLOSER LOOK AT MACHINE LEARNING

2. Unsupervised Learning – If only raw

insights into the data are wanted, the
computer is asked to find insights
effectively through unsupervised
learning.
A CLOSER LOOK AT MACHINE LEARNING
2. Unsupervised Learning – for example, Netflix may
ask its system to find other customers whose
viewing habits ae similar to yours.
It can then use those results to recommend
movies those similar customers liked.
Here there are no trained data sets such the
Good & Bad Loan labels; it is just looking at data
to find clusters of customers that are similar even,
though the human customers did not specify
what they meant by ‘similar’.
A CLOSER LOOK AT MACHINE LEARNING

2. Unsupervised Learning – Detecting fraud

and anomaly in transactions often falls in
this realm because you are essentially
looking for transactions that are ‘not
typical.’
A CLOSER LOOK AT MACHINE LEARNING
3. Reinforcement Learning – an AI system learns
under its own supervision without needing
trained data sets.

It makes predictions, validates those

predictions against reality & continually
adjusts to improve its output next time.

It’s inspired by behaviourist psychology &

refers to goal or reward-oriented algorithms.
A CLOSER LOOK AT MACHINE LEARNING
3. Reinforcement Learning – Algorithms are
trained to perform tasks by maximizing the
rewards or points for the expected favourable
outcome & continue to explore new
possibilities, essentially through trial & error,
until that maximized reward is found.
Rewards could be linked to increase values in
stock picking, evaluating trading strategies,
points in a simulation game or beating an
opponent in an adversarial game.
A CLOSER LOOK AT MACHINE LEARNING
3. Reinforcement Learning – In DeepMind’s
AlphaGo Zero, the system learned to play
simply by playing games against itself starting
from completely random play.
It quickly surpassed the human level of play &
defeated the previous champion-defeating
version of AlphaGo by 100 games to 0!
https://deepmind.com/blog/alphago-zero-
learning-scratch/
A CLOSER LOOK AT MACHINE LEARNING

Regression problems are solved mostly with

supervised & reinforcement learning while
classification problems, depending on
specifics, may be answered by using any
one of the three learning techniques.
 AI ALGORITHMS
• There are 3 different strategies used to implement
the 3 different techniques described previously.

• These strategies are at the heart of what allows the

AI system to ‘learn’ insights from the data.

• As CPAs, you may never have to deal with

technology to this level but a cursory look may
provide an appreciation of what AI algorithms do.
 AI ALGORITHMS
• Consider this very simplistic example. You
have data on sales per day at an ice cream
shop as a function of daily average
temperature.

• An AI system could identify a regression

equation that provides the best fit for the
provided data set & use it to predict sales for
next week using projected temperatures.
 AI ALGORITHMS
• Mathematically, the problem is now about
finding a curve that fits the geometry of the
correlation.

• If the AI system finds the parameters that

defined the curve, it can predict any Y ($)
values for a given X (temperature).
AI ALGORITHMS
 AI ALGORITHMS
• In classification problems, the curve would
typically divide two classes (e.g. good & bad
loans).

• AI algorithms are essentially all designed to find

the parameters that define the curve.

• These simple curves are only based on one or two

dimensions while, the more complex and
multidimensional the task, the more complicated
the AI algorithm.
 WHAT IS THE BLACK BOX OF AI?
• As algorithms become more complex, it is
not always clear how or why a decision has
been made by the AI system.

The lack of interpretability & transparency

of results to human users of the AI system
could lead to a lack of trust & loss of
confidence in the AI system
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• Deep Learning, also known as hierarchical
learning, is a subset of machine learning.

• It is an emerging & exciting form of AI that

can identify relationships & linkages in vast
volumes of data that would be impossible
for humans to process & apply them to
similar situations.
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• While still in early stages of development,
Deep Learning has the potential to take
automation to new levels of effectiveness &
improve decision making.

• How? By using algorithms that roughly

approximate the structures & functions of
the human brain.
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• The idea is to create algorithms that can
simulate an array of neurons in an artificial
neural network that learns from vast sources of
data.

• The interconnected layers of nodes or neurons

provide data to each successive layer as a
nested hierarchy of concepts many layers
deep.
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• The connections between these virtual
neurons are represented by a number or
weight.

• Those weights reflect the importance

attributed to the input data; the greater the
weight the more important the input is to
carrying out the desired task.
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• In simple terms, Deep Learning is the
stacking of algorithms (both supervised &
unsupervised models) on top of each other
into multiple layers.

• This allows for creation of extremely

sophisticated & finely tuned input/output
models.
 AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• During the training of neural networks,
weights are varied & adjusted until the
desired outcome is reached.

• E.G., when processing thousands of images

of human faces, the neural network
recognizes objects based on a hierarchy of
simpler building blocks:
AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
1. Straight lines & curved lines are at the
basic level
2. Then eyes, mouths & noses
3. Then entire faces and,
4. Finally, specific facial features.
AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• Once the 1st level learns primitive features,
this data is fed to the next layer, which
trains itself to recognize more complex
features.

• This process is repeated in successive

layers until the system can reliably
recognize a nose or a scar.
AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• Training deep learning systems requires vast
amounts of data; however, the learnings from
theses systems can sometimes be replicated
to other situations.

• For example, a neural network trained to

recognize facial features can be replicated in
multiple situations such as within different
security systems.
AI & THE HUMAN BRAIN: A TECHNICAL DIVE
INTO DEEP LEARNING
• In addition to recognizing images &
patterns, deep learning appears to be a
promising way to approach complex
challenges, such as speech
comprehension, human-machine
conversation, language translation &
vehicle navigation.
HOW AI IS BEING USED
The most promising & most advanced
applications of AI are in the areas of:
1. computer vision
2. natural language processing &
generation, and
3. audio processing.
HOW AI IS BEING USED
1. Computer Vision – a big reason why self-driving
cars and drones are possible.

• AI-based pattern detection pulls relevant

information from an image or sets of images in
order to further group & analyze them.

• Computer vision is also being used in retail

(Amazon Go store), agriculture (crop & soil
monitoring), medical diagnostics and, for
surveillance.
HOW AI IS BEING USED
2. Natural Language Processing (NLP) – is
everywhere.

• We interact with NLP in the form of autocorrect on

our phones, spell checks, grammatical error
detection & when we Google something.

• In essence, it is the ability of a computer system or

device to understand spoken or written natural
language in order to provide a range of insights or
answers.
HOW AI IS BEING USED
3. Natural Language Generation (NLG) –
transforms visualized data such as charts
or graphs into a narrative in an
understandable language.
• Quill by Narrative Science is an example of
how NLG can transform insight at the data
level into concise, easy-to-read, intelligent
reports.
HOW AI IS BEING USED
• Advancements in NLP & NLG have also given rise
to an era of chatbots, AI-based programs able to
carry on meaningful conversations.

• Chatbots allow us to have conversations with

virtual assistants such as Siri & Alexa.

• They are increasingly found on company websites

& are used for conversational marketing, customer
support & post-sales support.
ANALYZING BIG DATA
• Thanks to the IoT, social media & the power & outreach
of cloud & mobile computing, there has been an
explosion in the amount of data produced & the
amount is only growing.

• According to the IDC, the world is on track to create 44

zettabytes (or ~44 billion terabytes) of data by 2020!

• Organizations are tapping into increasingly

sophisticated analytics techniques to get closer to
customers, set strategy, innovate & grow.
ANALYZING BIG DATA
There are 4 main categories of analytics:
1. Descriptive Analytics provide insights into
events of the past.

Organizations use this technique to

evaluate their annual performance &
gather insights.
ANALYZING BIG DATA
2. Diagnostic Analytics examine data to
answer why an outcome happened.

It is characterized by techniques such as

drill-down, data discovery & correlations..
ANALYZING BIG DATA
3. Predictive Analytics look into the future to
anticipate outcomes such as demand
forecasting for supply chain operation.

In this case, existing data is used to train

machine learning models to forecast what
will probably happen.
ANALYZING BIG DATA
4. Prescriptive Analytics go even further by providing
possible outcome solutions that guide predictions into
actions such as generating ways to optimize
production or inventory.
E.G., when manufacturing a product, as system may
detect a lack of material to complete the order.
Prescriptive Analytics will analyze the inputs needed &
launch a production run for any missing material to
complete the manufacturing of the ordered goods.
INCREASING SOPHISTICATION OF BIG DATA
& ANALYTICS (PWC 2017)
ANALYZING BIG DATA

Bottom line, AI needs Big Data just as much

as Big Data needs AI because, as has been
established, AI needs lots of Big Data to
learn.
WHY RPA IS NOT AI
• RPA, or Robotic Process Automation is a set
of capabilities of software automation that
can handle high-volume, repeatable tasks
such as answering questions, making
calculations, maintaining records &
recording transactions.

• All are tasks that previously required

humans to perform them.
RPA IS NOT AI
WHY RPA IS NOT AI
• RPA can mimic tasks performed by
humans & automate those tasks digitally.

• RPA systems are easily programmed &

modified by non-technical workers.

• However, implementing dozens of process

robots requires significant involvement
from human technical specialists.
WHY RPA IS NOT AI
• RPA is a waypoint in the journey of organizational
automation which started off as basic scripting.

• E.G., basic automation could involve a factory-

line robot performing one simple repeatable
action.

• An orchestration activity within RPA could involve

downloading a file, taking key figures from the file,
transferring them to another file & saving the
document.
WHY RPA IS NOT AI
• All of these tasks are actions performed using
pre-programmed instructions.

• No additional decisions need be made & no

learning occurs as a result of performing the
tasks.

• Therefore, it is important to note that RPA on its

own is not AI.
WHY RPA IS NOT AI
• As technology progresses, RPA systems are
being coupled with algorithms to work with
unstructured data related to vision, images &
NLP in order to work on more judgement-
related activities.

• E.G., a bot in an HR recruitment team may

help with background checks from multiple
systems such as motor registration, credit
agencies, police verification, etc.
WHY RPA IS NOT AI
• Cognitive automation, which refers to AI
techniques applied to automate specific
processes, creates more capabilities.

• E.G., companies using cognitive

automation to perform signature
verification, through digital image
processing, for contract assurance in
telelcom operators.
TRANSITION FROM RPA TO RPA WITH AI
WHAT NEXT?
• If you think AI is impressive now, wait until it is
powered by Quantum Computing!

• Unlike classic computers which rely on circuits

or bits that are either on or off to perform
computations, Quantum computers are based
on quantum mechanics & use qbits (or qubits).

• These qbits have the ability to be both on & off

at the same time.
WHAT NEXT?
• The result; qbit-based computing is inherently
parallel (i.e. all possible outcomes from a
calculation may be calculated at the same
time).

• This inherent parallelism means quantum

computers of sufficient size could very quickly
factor large numbers & manipulate vast arrays
of numbers to detect subtle patterns, a core
requirement of making a powerful AI system.
WHAT NEXT?
• While the science & potential of AI is exciting,
the more businesses explore & embrace AI,
the more aware they become of the
challenges it poses.

• A 2017 PwC survey found that 76% of CEOs

were concerned with the potential of bias &
lack of transparency in AI systems.
WHAT NEXT?
• In order to overcome bias going forward,
organizations will need to develop an
approach to responsible AI.

• They will need to establish teams &

processes to look for bias in data & models
& closely monitor ways malicious actors
might ‘trick’ algorithms.
WHAT NEXT?
• On the other side, how will the black box
of AI impact the assurance professional’s
work effort?

• It is not just data bias & malicious actors

that have to be worried about; think
privacy.
WHAT NEXT?
• What are the privacy implications when an AI
algorithm suggests a product to a consumer
based on the individual’s past preferences &
purchasing habits?

• And, what might the fallout be when individuals

are categorized for mortgage acceptance or
flagged at airport security?

• These questions & concerns are increasing just as

our use of AI in our everyday lives is increasing.
THIS IS JUST THE START
• CPA Canada & the AICPA are committed to
exploring the implications of AI.

• Future companion publications will address the

impact of AI on CPAs; from CPAs in business to
assurance professionals.

• It is important that you continue to read,

collaborate with colleagues & learn about AI &
the other categories of technologies which may
impact you as a CPA.
MARVIN: HITCHHIKERS GUIDE TO THE GALAXY
QUESTIONS?