Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 8

Jobes 1

Nathan Jobes
CST 300 Writing Lab
20 February 2021
Artificial Text Generation in Media

The realm of artificial intelligence is vast and filled with unexplored ideas and

applications. Various forms of machine learning have come in and out of popularity for decades

but they grow ever popular as uses for inference models continues to grow. This of course means

companies and researchers alike are constantly pushing the boundaries of what artificial

intelligence models are capable of and consequentially bring up debates surrounding their use

and development (Hagendorff, T. 2020).

Currently, artificial intelligence models of all kinds are finding homes in many parts of

our everyday lives. For example artificial intelligence has found its way into the majority of

social media applications, Facebook, Twitter, Instagram, etc all use some form of artificial

intelligence to filter for problematic users, make recommendations, etc. Even journalists and

news media outlets have begun using language models to speed up their workflows (Rangaiah,

A. 2019).

However, sometimes researchers push the boundaries and bring up questions that have

not yet been asked or whose lack of a solid answer has been ignored in the name of progress. For

example the GPT-2 model produced by OpenAI was one such controversial project. The model

is an iteration on the first language model (GPT) and generates the next word after being shown

a variable amount of preceding text (Radford, A. 2019). This doesn’t seem to be very impressive

but that impression is possibly misleading. Using this model, the researchers were able to have a

computer read and write stories. Provided a well worded opening prompt the team was able to

have GPT-2 produce relatively convincing articles as well. The models abilities were good
Jobes 2

enough to break records in multiple language related tasks such as the Winograd Schema

Challenge (Radford, A. 2019).

OpenAI was concerned enough with the repercussions of such a capable and versatile

model being publicly available that they brought up the issues they had in their original blog post

about GPT-2. The team lists a few uses for their model; the benefits being related to speech

recognition, translation and writing tools. But the blog also calls out the possible abuses. OpenAI

supposed the GPT-2 could be effectively abused to benefit malicious persons seeking to spread

misinformation or quickly generate convincing texts for phishing scams. So the OpenAI team

initially refused to publicly release the final model and even suggested politicians begin

considering enacting policy regarding this kind of technology (Radford, A. 2019).

This then brought into question how far the community was willing to take projects like

these. Suddenly the automated media practices already in place were poised to become

legitimately fully automated in the near future. No artificial intelligence model had ever written

articles like GPT-2. With a fully automated press within site, it became a real concern whether or

not models like GPT-2 should be made available and if they should be used at all by non-

researchers (Wakefield, J. 2019).

Technology like this is more than a few steps up from writing a script to take the last

night’s basketball scores and put out an automated tweet with the statistics of who won and by

how much. Artificial intelligence capable of writing complete articles, with cohesive points and

structures spawned from nothing but a few sentences with key points, is a great tool to the

virtuous and potentially a weapon to bad actors (Wakefield, J. 2019).

So who cares? Well the people affected by the presence of such technology would be

effectively everyone. The majority of the world has some amount of access to the internet, and
Jobes 3

news media is relatively accessible with some obvious exceptions. The general public and world

governments are concerned about the possible abuse of such capable text generation models;

whereas journalists and news media stand to benefit greatly from the benefits of such tools.

Journalists want to get information to the public. The more information they can verify

and publish for the education for the masses the better. Pursuing information and breaking down

complex issues takes a lot of time and resources. This is why journalists and other media

companies would very much like to keep moving towards automated text generation tools. Tools

like GPT-2 will allow them to produce the article to share the information they have gathered

more quickly, reduce editing time, and allow journalists to spend more time finding more

information on the topic or cover additional topics they otherwise might not have the time to

cover (Carro, B. 2018).

The media of course makes claims of policy to support their position as they seek

acceptance for the use of “robot journalism”. They assert the benefit of more robust news

coverage is immeasurable to public discourse; thereby justifying the use of the tools. They

reference the current paradigm that has been allowing them to do so, and other entities to do so

for an extended period of time. The claim that these current policies should remain mostly as

they are and restricting them would have negative effects.

Conversely, the public and governments rely on trust and expect honesty, especially from

news media. The public needs to be able to trust the information they are getting from the news

as it affects their understanding of the world around them and their relationship with the world.

Automated news poses a threat to that relationship and could destabilize it.
Jobes 4

They argue that technology like this should be regulated or not allowed at all as well as

saying automating journalism contradicts journalistic standards. These are claims of policy and

value, as one desires government oversight and the others calls the practice inherently unethical.

What the community is left with is one serious question; should this type of artificial

intelligence ever be developed to begin with let alone be used in media. The possibility of an

automated press poses significant dangers to the world if allowed to be abused. What benefits, if

any, could make creating such a tool worth the risks or what would society be losing if we

didn’t?

The news media tend to apply Kant’s ethics to justify their desire for access to these

tools. Kant’s ethics is a sect of ethics pioneered by Immanuel Kant in the later 1700’s around the

idea of a categorical imperative. That is, one acts morally if any other person would do or must

do the same (Jankowiak, T. n.d.). The standard by which “others would do the same” comes

from rationality and universally accepted laws, for example murder is a generally accepted

wrong.

So the news media argues that it is irrational to expect a significant change in the honesty

of their publications from using text generating tools. Current “robot journalism” standards

include human review of stories and data (Rangaiah, A. 2019). It is unreasonable to think the

integrity of the institutions would change because they used artificial intelligence to automated

the more tedious processes involved in reporting. Furthermore, every other business is expected

to find ways to increase productivity and reduce costs. Doing so is essential to staying a thriving

business, so seeking out ways to significantly increase reporting is what people should expect

them to do.
Jobes 5

They also argue that it is irrational to think the technology will never be created. The

rational thing to do is to research and understand the limitations of the technology so the world is

not blindsided by its use in a malicious way. These tools do not create a new problem, but they

can be used to solve current ones. Yes, the tools could be used to worsen current problems like

“fake news” but that already exists. Denying the world the benefits of the technology for the sake

of maintaining the status quo is also unjustifiable.

The media and researchers have a lot to gain from being able to have access to these

public models. Sharing information facilitates advancement in the field and automation in

journalism stands to make media sources more profitable and potentially allow the media to

expand their coverage of complex topics. These tools also provide needed user monitoring for

social media that would put start up social media platforms at a disadvantage.

For the public and others, they take a utilitarian approach to their argument. The

utilitarian argument comes from as early as the 1600’s but is formally attributed to Jeremy

Bentham and John Stuart Mill who formalized the concept in the 1800’s. The framework

chooses to look at the net effects of an action and asks if it causes more harm or good the all of

those it affects (Driver, J. 2014)? If an action were to negatively impact some people greatly and

benefit less people barely, then that action would be clearly unjustified.

So, the public argues that it is wrong to develop these text generators because of the

potential they have to cause harm to the masses. The benefit that journalists might get from

automated statistical posts or roughing out entire articles is far outweighed by the huge numbers

of people that could be swindled, or convinced of falsehoods by the abuse of such automation.

Because the artificial intelligence tools have the ability to do substantial harm and limited good,

they are an unreasonable pursuit and should be banned from being used or even produced.
Jobes 6

The automation of inherently human jobs, such as reading people’s posts and

investigating and reporting the facts and objective analysis of current issues also takes away jobs.

Moderators and some journalists stand to lose their jobs as the automation of language

processing means less people are needed to screen people’s posts or report the basic news. The

loss of those jobs hurts more people than the benefit to the company as a whole. This again

violates utilitarian ethics and therefore deems the use and development of these text generators

unethical.

The public and governments have a lot to lose if abusable software like this is made

available. Misinformation can really slow down progress and poisons public discourse. Tools

like this could easily overwhelm the less aware members of the public. This would sow mistrust

and could escalate tensions in politics and other forums.

Personally, I agree more with the arguments against artificial text generation within

journalism, however I do not think stifling research is the correct course of action. I think

policies should be enacted banning the use of text generators for the writing of any content that is

not strictly trivial or statistical in nature. For example automated stories about the current scores

in the World Series of baseball are acceptable; an article on the constitutional implications of

Richard Nixon pardoning himself should not be allowed to be automated.

The potential for abuse with a technology like GPT-2 is huge, such as the disruption of

civil discourse in politics. Therefore the ease at which nefarious entities could use this tech to run

extensive disinformation campaigns presents a huge problem. If ethical organizations do not do

the research into this technology, someone else with a lower standard of ethics will and the world

would be caught off guard.


Jobes 7

It is also worth noting that governments will portray themselves as wanting to prevent the

abuse of the technology. In reality tools like these are very attractive to the NSA for profiling

suspects, and could be used to wage huge misinformation campaigns against a nation’s enemies.

For example, the U.S. has a history of getting involved in other countries affairs since the cold

war, and a remote media attack is a lot more economical than creating exploding cigars.

When the scandal around OpenAI’s GPT-2 model was going on, other researchers were

able to show that they could teach other machine learning models to distinguish between

automated articles and authentic ones. Researchers should be allowed to probe the capabilities of

the technology to better prepare to defend against bad actors and misconduct. And actions like

OpenAI refusing to make the model public should not be strictly condemned as making

something so powerful publicly available right away as doing so in the future could protect the

public from misuse before countermeasures are available.


Jobes 8

References

Carro, B. (2018, October 10). Pros and cons of robot journalism. Mediego.

https://www.mediego.com/en/blog/pros-and-cons-of-robot-journalism/

Driver, J. (2014, September 22). The History of Utilitarianism. The Stanford Encyclopedia of

Philosophy. https://plato.stanford.edu/entries/utilitarianism-history/

Hagendorff, T. (2020, February 1). The Ethics of AI Ethics: An Evaluation of Guidelines. Minds

and Machines. https://link.springer.com/article/10.1007/s11023-020-09517-8?

error=cookies_not_supported&code=b5371862-68d4-46e4-9d5d-4156f0b0392a

Jankowiak, T. (n.d.). Kant, Immanuel | Internet Encyclopedia of Philosophy. Internet

Encyclopedia of Philosophy. Retrieved February 18, 2021, from

https://iep.utm.edu/kantview/

Radford, A. (2019, February 14). Better Language Models and Their Implications. OpenAI.

https://openai.com/blog/better-language-models/

Rangaiah, M. (2020, February 4). Role of Artificial Intelligence (AI) in Media Industry |

Analytics Steps. Analyticssteps. https://www.analyticssteps.com/blogs/role-artificial-

intelligence-ai-media-industry

Wakefield, J. (2019, August 27). “Dangerous” AI offers to write fake news. BBC News.

https://www.bbc.com/news/technology-49446729

You might also like