Ethical Issues in Using Artificial Intelligence?

So I decided to go to the source and ask ChatGPT, “Is it ethical to copy and claim as ones own text generated by ChatGPT or other AI?”

And ChatGPT replied:

No, it is not ethical to copy and claim as one’s own text generated by ChatGPT or other AI without proper attribution. Just like with any other intellectual property, the original author or creator of the text generated by the AI has the right to be credited and recognized for their work.

Furthermore, it is important to note that the ethical use of AI-generated text should be guided by the same principles that apply to human-generated content. This includes avoiding plagiarism, respecting copyright laws, and giving proper credit to sources. In addition, it is important to consider the potential consequences of using AI-generated text without proper attribution, such as damaging one’s reputation or violating intellectual property rights.

Therefore, if you are using AI-generated text, it is essential to make sure that you are doing so in an ethical manner by giving credit where credit is due and respecting the rights of the original creators.

Of course what is missing from this answer is that ChatGPT doesn’t credit its sources, so it isn’t practicing what it preaches. I followed up with this question, “How does ChatGPT justify demanding that it receive attribution when copying what it says when ChatGPT doesn’t reveal its sources?” ChatGPT replied:

As an AI language model, ChatGPT is trained on a large corpus of text data, which includes a diverse range of sources. While ChatGPT itself does not reveal its specific sources, the model’s training data includes a wide range of published and publicly available text data.

When ChatGPT requests attribution for its responses, it is not necessarily asking for attribution to its specific sources, but rather for attribution to the model itself. Attribution to ChatGPT acknowledges the work that went into training and developing the model, as well as the intellectual property rights associated with the model.

Furthermore, when users copy and paste responses from ChatGPT without attribution, it can create confusion about the original source of the information. Attributing the information to ChatGPT helps to clarify the source and provides proper credit to the model’s developers.

In summary, while ChatGPT does not reveal its specific sources, attribution to the model itself is still important to acknowledge the work that went into developing the model and to clarify the source of the information.

Seems pretty much “weasel worded” to me.

4 Likes

I’m reading people reporting how they use it, and articles about how people use it, and I’ve experimented a lot with it myself. I know at times it’s tempting to take, for instance, a summary, and just plug it into my work as a conclusion.

Clean copy is clean copy, but is it my clean copy?

The answer is not dependent on how clean the copy is. That was my point. For an editor, clean copy just means less time spent making all the annoying routine corrections and more time spent on the substance of the work.

If your narrative is weak or your sources nonexistent, it doesn’t matter whether you used the right verb tense. Any editor worth their salt will notice that.

And if you don’t have an editor worth their salt, or any editor at all, neither have most of the people who have been stealing other peoples’ work or just outright lying since the beginning of time. I don’t see how AI changes that.

People seem to forget that machine learning and machine intelligence are not human learning and intelligence. ChatGPT cannot create anything new, it only copies the things it has been trained on to produce a response that is statistically similar to a predictable human response to a question.

So when you ask ChatGPT for a statistical analysis of scientific studies on a specific subject and it comes back with references to a number of completely fictional papers it’s not lying or doing anything other than it was designed to do. That is a response that imitates a ‘real’ response to that question. ChatGPT doesn’t understand what a source is, it doesn’t understand attribution or analyse those (fictitious) papers. It is an imitation engine based on a composite of thousands of source samples.

All ChatGPT will ever do it imitate. All ChatGPT responses are plagiarised.

You see this in image generation “AI” too, when an image looks amazing but then you notice the hair of the subject seamlessly merges into the subjects shirt. The “AI” does not understand what a shirt is, or hair. It doesn’t even comprehend the concept of separate discrete objects — only the probability that based on the prompt this specific pixel is highly likely to appear next to that pixel. All based on the images it was trained on.

When you ask ChatGPT to write a function in javascript it’s just returning probablistic composites of other people’s code.

People have anthropomorphised these AIs wholly inappropriately because they straight up don’t understand them. They are not intelligent. They do not understand any of the content or context of their prompts or responses. They are incapable of producing anything new; they are not creative or generative and honestly the most unethical practice is their creators basically selling them as such when they are not. Unfortunately a lot of people that understand how to train a machine learning model don’t really understand how they work.That’s how you get Google engineers trying to start lawsuits to ‘prove’ their AI is sentient when it absolutely is not and the suggestion is in reality hilariously foolish. It’s not even wrong to use one of my favourite phrases.

4 Likes

Therein is the rub. Copyright law is very complex and not very explicit in that it is legally largely built on a mass of previous cases rather than a clearly written exposition.

In my own experience trying to deal with copyright, it interests me that stealing “ideas” and reproducing them in your own book or whatever, is not the problem. Copyright is primarily about stealing the actually wording of how someone else expressed those ideas. If you had some particularly clever explanation of why Ben Franklin liked living in France, that could be copied/stolen freely as long as the language used to express this fact was not itself copied. It might be “frowned upon” in some circles but it is not a copyright issue.

How that works with ChatGTP is a confusing for me, and I suspect it will be confusing for the law to adjudicate. It does not, in the examples and discussions that I have seen, just grab paragraphs of text from sources and just copy them. For example, just gluing paragraphs of text together. That would run into copyright restrictions,

Rather it is a master of rewording things according to its own lights of how text should be put together. I don’t think that is a really a copyright problem.

In many fields, like academic research, copying ideas with out crediting the sources is frowned upon and not permitted in most academic journals. But this is not really an issue of copyright. This is outside of copyright.

2 Likes

Which explains how it can write such eloquent BS.

Mostly “not” it seems. That’ll change, of course. Programming being an exercise in logic, I won’t be surprised if progress in that area comes faster than with classical prose.

3 Likes

It’s even worse than that. It keeps thousands of IP lawyers busy! :slightly_smiling_face:

Everything You Need to Know About Becoming an Intellectual Property Lawyer - NEL

In an attempt to hit the original issue:

  1. Energy consumption - these services use massively more energy than a google or DEVONAgent search. (These services are orders of magnitude worse than any other related tool).
  2. Trained on other people’s copyrighted material - this is more nuanced for me. Likely OpenAI has violated the copyright of myself and millions of other people. If we let that go. When I write on a subject, I read what other people have written first and then argue with them. So for me - using LLM hallucinated text so I can argue with it - awesome
  3. There are no real citations so you can’t credit anyone else? Most interesting ideas are built on top of other people’s ideas, I’m a big fan of referencing other people’s work.
  4. I would ask ChapGPT itself but don’t want to waste the electricity.
1 Like

Thank you for all the great comments. Most of you know a lot more about the functioning of AI than I do. I appreciate the opportunity to learn from you. :smile: :smile:

1 Like

Set aside whether it is your clean copy. Can you trust it? Have you verified if its claims are valid. In area of your own expertise, I suspect this is easy enough read it and decide if it’s correct. In areas outside your expertise how do you know if it’s correct?

1 Like

I’m concluding that “trust” is a major issue here, especially since there is no attribution or any hint of where the data comes from.

A few days ago I asked ChatGPT to summarize four novels that I have recently read for a book club I’m in. In every case it got some basic facts (not opinions) wrong, including characters that didn’t exist (and since I read Kindle E-books it’s easy to search for names), wrong relationships, plot points, and locations. Nothing to be trusted here!

4 Likes

That’s good to know and a little bit scary.

It has asserted books and articles exist by known writers; they were completely invented.

These bots are not AI; they are LLM. They are a very sophisticated version of what we used to do with linguistic corpora and Perl.

2 Likes

This isn’t a mistake on ChatGPTs part - it’s exactly what it’s designed to do.

It doesn’t read, understand, and analyse the books to produce a report - it only produces text that is highly probable to look like a book report for a book with the titles you provided. It’s infinite monkeys producing Shakespeare - it is not an intelligent mind doing analysis.

4 Likes

Hence basically useless as an information source.

2 Likes

If it had an abundance of reviews of those specific books in it’s training data it might actually have been more accurate. If you ask it to summarise an event that got a lot of news coverage you’ll probably find the responses are pretty accurate.

I’m still only in my early thinking on this topic, though I think you are too, hence your question. Although let’s be honest, sci-fi has been been providing food for thought on this issue for decades and there are plenty of genuine experts who have dedicated careers to pondering the mysteries of these ethical questions.

For me, it boils down to two questions (I dither over whether one should be weighted more than the other, but today as I write this I think not. Writing is a contract between the writer and the reader, and both questions should be weighted equally).

  1. Is it ethical for me to pass something off as my own writing/thinking when it was generated by an app?

My answer is no.

  1. Is it ethical not to warn the reader when writing has been generated by an app, not a human?

My answer is no, and actually it’s important to note here that we’re all online, techy people, but of the polls I’ve seen for France, U.S., and UK, the majority of the public in those countries do not agree with / are uncomfortable with the rise in the use of AI over the last few months. No matter what Silicon Valley is saying, most people at present do not want this in their lives and are worried about it when polled on it. We have a duty of care to inform people when it’s being used, at least until we’ve got some kind of understanding about what we’re all doing.

Final parting thoughts:

I’m a scientist so my interest is particularly from a science perspective, and I agree with the prevailing views now starting to emerge in the “science community” (as much as that can be considered one entity!) that it is unethical (and “unscientific”) to use any AI that is not open source, since there is no way to understand its algorithms, inherent biases, etc., and that alters what data comes out the other end. At a super-basic level: don’t bring anything into your lab/ research that you don’t understand. (ChatGPT fails this test and should not be used, so although I’ve mentioned it below as scenarios I like, I don’t support the use of ChatGPT itself for this, it’s just what people are playing with currently.)

I much prefer the longer-term vision of AI as co-pilot, as demonstrated by Star Trek, but of course already having real-world applications, e.g. checking developer code for errors, saving you if you try to hit a lorry in your car, etc.

A couple of my colleagues have played with ChatGPT as a way to bounce ideas (can be handy for asking iterative questions and thinking through an issue), I’ve seen an ad agency use it to veto designs (I.e. have a human discussion for ideas, then ask ChatGPT and dump any idea that ChatGPT offers as they’re clearly not original, and everything that’s left is worth considering further). I like how Readwise Reader have implemented it as a tool with a specific use case (called Ghostreader - it is there to co-pilot with you as you read: it can summarise a paper so you know whether to spend time on it, help you interpret statements you don’t understand, define terms in context, etc.). All these uses still require human thinking, and the AI serves to complement the human endeavour rather than supplant it.

I suppose my basic ethic code here is “does the AI help a human reach further?”. If you’re using AI to write an article you haven’t researched yourself, then no. If you’re using AI to help improve your article by working through your thinking, testing weak points in your argument, etc, then yes.

(For what it’s worth, the only AI I am aware that I’ve deliberately used is Ghostreader, and to date that’s mostly just been for fun to test how it works. I can see its value, but apparently I’d still rather just read something and figure it out on my own. That might change with time.)

1 Like

This is what will matter in the long term - whether or not most people have a reason to adopt a technology. The tech world has been excited about an endless list of technologies over the years that are basically dead on arrival - 3D TV/Movies and VR are good recent examples. Sure they may still see some niche use, and some areas of industry and the tech world are still pretty obsessed with them, but the public at large has pretty much forgotten about them.

The metaverse is another good example. I like the quote from the legendary John Carmack, Meta’s now previous head of the metaverse, talking about Meta’s own implementation at it’s own conference about the metaverse:

“This here, this isn’t really what I meant. Me being an avatar on-screen on a video for you is basically the same thing as [just] being on a video.”

All the tech industry has given us of the metaverse so far is a way to do the same things we already do better elsewhere. Why should anyone care?

The same is very much true of current AIs right now. Once the novelty wears off, will it really have improved our lives?

I’m mostly in agreement with this statement, but I’m trying to reconcile with the similar condition that we have no real “open source” way of understanding how the CPUs that process our data work either. We have to rely on a very closed box that accepts our inputs and produces outputs that we then have to find some way of trusting,

ChatGPT does a great job of bluffing when it doesn’t know the answer! A truly intelligent being knows when to say “I don’t know.”