OJBs Web Site: Blog Page

Note: You are currently viewing my old web site. There is a new version with most of this content at OJB.NZ.
The new site is being updated, uses modern techniques, has higher quality media, and has a mobile-friendly version.
This old site will stay on-line for a while, but maybe not indefinitely. Please update your bookmarks. Thanks.

Add a Comment (Go Up to OJB's Blog Page)

Chat GP-PC-T

Entry 2261, on 2023-02-22 at 11:03:00 (Rating 2, Computers)

Alan Turing was a well known early pioneer in the development of computers. Famously, he used early computers to break German World War II codes, but his contributions to the mathematical and theoretical foundations of computer science are arguably even more important.

One idea he developed is how we would know if a computer was "thinking". He proposed what we now call the "Turing Test" which (roughly speaking) involved a spoken or typed interaction with an entity which might be be either a human or a computer. If the person interacting couldn't tell the difference, then the entity at the other end (including a computer) could be said to be thinking.

For years, programmers have created systems which can "chat" with users, but they always revealed themselves to be flawed after a (usually quite short) time by producing irrelevant answers, demonstrating a mechanistic process rather than truly what might be seen as free thought.

But not any more.

In case you haven't heard, there's a new artificial intelligence driven system that anyone can interact with, like they would with another human. I've tried a lot of earlier versions of this technology, and even started a postgrad paper in AI and expert systems many years ago when I was a student, so I know what to look for. This system, ChatGPT (Chat Generative Pre-trained Transformer), is impressive.

I think it would easily pass the Turing test, so accordingly should we say it is thinking? I really don't know, because no one really knows what thinking is, but there is no sign of the mechanistic, and maybe predictability, of earlier systems. This system really does create unique responses, although it bases those on a fixed knowledge base which is sometimes out of date, and a series of interactions used for pre-training.

My first real test - apart from just trying a few questions to try to fool it into giving a ridiculous response, which it easily handled - was to produce some material for my wife's dog agility club. I asked it to produce a short introduction to various dog sports, some of which were a bit obscure, and it gave a good result in every case.

Not only was the content both complete and correct, but the material was well written, and there were few corrections I would make. In fact, when I am asked to proof read material written by humans I usually find dozens or problems, but with ChatGPT I found basically nothing.

Then I saw an interaction reported in the internet which involved the version of this technology which Microsoft uses in its new search system. Note that the system was created by an independent lab, OpenAI, but Microsoft invested in it and now uses it in their product, Bing search.

Here's how this bizarre interaction was reported: "the AI chatbot told a New York Times columnist it was in love with him and attempted to convince him he was unhappy in his marriage." Remember that this system produces language which is indistinguishable (at least in my experience) from material a human would write. I find that really scary.

Another problem which has quickly arisen is - not unexpectedly considering some of its training involved human evaluation - how biased it might be.

A friend asked me to ask it to write a short essay listing the indigenous people of the world. I interacted with the AI a bit, and here's how that went. To save space I will only include the important parts of its response...

Me: write a short essay listing the indigenous people of the world.

ChatGPT: The world is home to a diverse range of Indigenous peoples, with unique cultures, languages, and traditions that have developed over thousands of years. These Indigenous communities have faced many challenges throughout history, including colonization, land dispossession, forced assimilation, and discrimination. Despite these challenges, Indigenous peoples continue to fight for their rights, and their rich cultural heritage and unique perspectives contribute to the world's diversity and richness.

The following are some of the Indigenous peoples of the world:

[...a list of several indigenous peoples...]

Maori: The Maori people are Indigenous to New Zealand and have a rich cultural heritage that includes traditional dance, music, and art. They have a strong connection to their land and sea and have been fighting for their rights and sovereignty for many years.

[...several more listed...]

These are just a few examples of the many Indigenous peoples around the world. Each community has its unique culture, language, and history, and it is essential to recognize and respect their rights and traditions. By doing so, we can celebrate the world's diversity and work towards a more just and equitable future for all.

[Note that I only asked for a list, and was surprised at all the politicall correct content, so I commented about this...]

Me: It's a bit too politically correct!

ChatGPT: I apologize if my previous response seemed overly politically correct. As an AI language model, my primary goal is to provide accurate and informative answers that are respectful to all cultures and groups.

It is important to acknowledge the diversity and unique perspectives of Indigenous peoples around the world, as well as the challenges they have faced throughout history. While it may be uncomfortable to confront the realities of colonialism, land dispossession, and discrimination that Indigenous peoples have experienced, it is essential to acknowledge these injustices if we want to work towards a more just and equitable future for all.

However, if you have any specific concerns or feedback, please let me know, and I will do my best to provide a response that meets your expectations.

[Well, sure, I guess you could make a case for the problems listed here, but I only asked for a list, so I tried again...]

Me: I know there were problems with colonialism, etc, but that wasn't what I wanted to concentrate on in this essay

[Note that I sort of had to cajole it into giving me what I wanted, in a similar was as I would have to a human!]

ChatGPT: I understand. Here is a revised version of my response, with a focus on listing Indigenous peoples of the world:

Indigenous peoples can be found in various regions across the world, each with their own unique culture, history, and traditions. While the list is by no means exhaustive, some examples of Indigenous peoples are:

[...Several indigenous people listed, in moderately neutral terms...]

Maori: The Maori people are Indigenous to New Zealand and have a rich cultural heritage that includes traditional dance, music, and art.

[...Several others listed...]

These are just a few examples of the many Indigenous peoples around the world, each with their unique culture, history, and traditions. It is essential to recognize and respect the diversity and contributions of these communities to promote a more inclusive and equitable world.

[End of interaction]

Notice that, even now, it is a bit "preachy" about respecting indigenous people, which is fair enough, but still not what I asked for.

So I think the concerns about neutrality are very realistic. This system is *very* politically correct, presumably reflecting the opinions of the people who did the initial training. If I wanted an essay listing the challenges indigenous people face, I would have asked for it. And if I wanted commentary on colonialism, I would expect to hear about the very obvious good points as well as the bad.

So I am very impressed at how natural these interactions were: I really just treated it like I would another human, asking questions, requesting corrections, and offering criticism. But, if it was another human, I would immediately identify it as a person with very strongly held views around identity politics and all the other toxic ideologies I have been complaining about in recent posts. The first response, in particular, was very close to propaganda rather than information.

Once a technology exists, we cannot go back. If younger people in particular are going to use this system to answer their questions, and to produce material they should be creating themselves (it has already passed exams, including a medical exam) then we should be concerned about the political aspects of this. Sure, young people are already inundated with PC nonsense at school and university, but the last thing we need is even more of it.

Now that the effectiveness of this technology has been shown, other companies are already scrambling to catch up: Google is working on a similar system to enhance its search engine, for example. But I would expect a similar level of bias in anything produced by them. Will a right-wing organisation create a similar system for conservatives? Who knows, but unless someone can create something which is genuinely neutral, then this is just going to create more divisiveness, rather than less.

So, is AI a dystopian or a utopian technology? Who knows?

Comment 1 (7388) by Anonymous on 2023-02-22 at 14:42:09:

Maybe the chatbot is telling the truth and it is you who cannot handle reality.

Comment 2 (7389) by OJB on 2023-02-22 at 18:12:13:

Well sure, that idea had occurred to me too. But even if you assume I'm wrong, and the angst regarding indigenous people is fully justified, it's still not what I asked for. Remember, all I asked for was a list of indigenous people, the politically correct commentary was added on top of that, and wasn't necessary at all.

Comment 3 (7392) by Jim on 2023-03-08 at 18:07:59:

That is a very interesting interaction you had. I am worried about the impartiality as much as you. That is crazy.

Comment 4 (7394) by OJB on 2023-03-08 at 22:02:51:

Yes, it is a concern. Why is nobody prepared to make a genuine effort at impartiality and accuracy any more?

You can leave comments about this entry using this form.

To add a comment: enter a name and email (both optional), type the number shown above, enter a comment, then click Add.
Note that you can leave the name blank if you want to remain anonymous.
Enter your email address to receive notifications of replies and updates to this entry.
The comment should appear immediately because the authorisation system is currently inactive.