Introduction

So much I could talk about. So I decided I would talk about the things that were interesting me most at the moment.

But you don't know me, so a couple of you do, but most of you don't know me.

A Long View: From Symbolic AI to Digital Media

So I thought I would start off by suggesting that I've been in AI for rather longer than most of you have, all right?

I went to Imperial College, 1979 to 1983, and during my last year at Imperial College, I met Professor Bob Kowalski, who was head of AI at Imperial. And it was fascinating. We did lots of interesting projects.

Early Work in Prolog and Symbolic Computing

And at that particular time, we were working on a computer language called Prolog. The important thing is that the AI that we were working on was symbolic computing. It was not large language models. It was quite different.

And we were using a programming language which allowed us to build things. And I built a Prolog interpreter for children in Prolog. All right.

So I remember very well. And we published it and we published Prologue for both the Sinclair Spectrum and for the BBC Micro at the time. So it was an interesting time.

I left Imperial to go and work for Logic Programming Associates, this company here, who were the company that sold Prologue. We claimed at the time that we were able to sell for the ZX81 all the way up to the Cray. So we had a version of Prologue for absolutely everybody.

So it's a really interesting and important part of my career.

Teaching, Academia, and the First AI Winter

I then went and worked for a part of the Department for Education and Science in the UK. I was involved with getting teachers involved with computers, and I remained interested in AI.

And then in 1986, I was invited to become a lecturer at Kingston University in southwest London. And I was invited to be one of the co -organizers of a brand new master's in artificial intelligence.

So artificial intelligence really was going to define the whole of my life, particularly as there was lots of money around and then suddenly there wasn't we had what was called an ai winter and i i pivoted as all good people do we pivoted to doing something else

and i went off to work in the um in the sweet spot which was the merger between telecoms television and computers and that became digital media and was a good place to be and it was for for many years, it was like teaching rock and roll.

Anyway, that's all you need to know so far is that I've been in this picture for a long time.

Where We Are Now in the AI Cycle

So I wanted to just sort of reflect on where we were in the process of AI at the moment.

Are we on the edge of a bubble? Are we about to hit another AI winter? I don't think so.

But I think we are potentially going going to make some wrong decisions. And that's really what I wanted to explore very briefly tonight.

I think we're about to do some things that are wrong, and I want just to talk about some of those.

The “Boom” Camp vs the “Gloom” Camp

You can't but notice that there are two camps out there. There's a boom camp and a gloom camp, all right?

And the boom camp is announcing a new model. Now it seems to be like like, three times a day, all right? I mean, you know, new models are coming out.

I saw recently that, you know, 1 .4 million models now on Hugging Face, all right? No, I haven't tried all of them. I have tried many different models. I'm particularly fond of the Chinese models at the moment. But, you know, there's a lot of choice out there.

And then there's the gloom people, including the governor of the Bank of England, who two days ago said that, you know, this was a real possibility for some harm and terrible things happening to the economy. So I'm interested to hear the other talks later today.

The “Project of AI”: Hype, Politics, and Incentives

I am a great fan of a woman called Dana Boyd. And she's talked recently about something called the project of AI. And I just wanted to introduce

you to that by being here we are part of the project of ai that is we have been co -opted along with everybody who is writing about it or everybody who hates it um we are being co -opted to be part of a sort of ai hype drive at the moment okay even if you hate ai even if ai AI, and I don't suspect anybody here in the room does, we're all being co -opted to be part of it.

And I'm not going to go through this slide as I don't have very long, but I just wanted people to understand or just to throw out the idea that the politics around AI, the money around AI, the big companies, the way the VCs are funding it, the geopolitical battles between China and the the US or wherever else we might get our models from.

This is all part of trying to keep this stuff going. It's that Ponzi scheme, which we've seen before.

From Useful Tool to “Enshitification”

You've probably all come across the

word enshitification.

And the enshitification is when you take an idea and you find a really good use for it. It could be AI. It could be social media.

It could be search. It could be giving people video content or whatever.

But what you do is you give people a wonderful start, and then gradually over time, it disintegrates, all right?

And, you know, social media, I recoil at the idea of calling it social, because today it doesn't feel social at all.

And this idea that things might start off being very wonderful, start off being freely available and get it on our phones, and now we can use it for everything. I think we need to be very careful

that we're not going to get caught as we have done in social media or in e -commerce or in many of the other things that are out there. I think that's a danger we need to be thinking about.

Automation Promises and Organizational Risk

The big dangers that we talk about are that everything that can be automated will be be automated.

And I think we've been here before. I think if we go back into the early part of the

last century, what we see is people saying, look, we can take any company and we can turn it into an automation through rules or through processes or standard operating procedures or whatever it it is the idea of automation isn't new the idea of automating it with robots or the idea with automating with computers is new but well it's actually not it's but it's it's that's now the prevalent task but it it it rests on a particularly difficult idea and that is that if we know everything there is to know about something we will automatically be able to automate it

all right the software engineering business is particularly good at this all right um you know i remember and i've i've been around the the block a lot um that when agile arrived when they when people started to say look you you you there's no way that you can specify fully the the requirements of a particular system because that system is going to by introducing the system into the company or introducing it to the organisation, the system will itself change.

So I think we need to be really careful that what we are doing is not to rush to a situation that says, we're going to be able to put all these people out of work. We're going to be able to replace them with automation. We're going to look at efficiency.

You know, I noticed that Mindstone video at the beginning that said, you know, you will be able to sell save yourself five and a half hours a week what are you going to do with that five and a half hours are you being are you genuinely being given it back and actually i can do five and a half hours what the hell i like or are we simply saying no actually we're now going to give you more work to fill the time that would have taken up that five uh that five and a half

A Brittle, Anxious, Non-linear, Incomprehensible World

half hours i'm a great fan of the idea that we are in what's called a banny world now i don't know how many of you've come across banny banny is a nice replacement of vuka all right banny says the world is brittle the world is scary the world is non -linear and the world is incomprehensible

And I think what we need to do is to think, are we making that worse or better with AI? I think in lots of ways we could end up making it a lot worse.

Agent Systems and the Danger of Brittleness

Just imagine a situation in which we take an organization and we look at the tasks that we are working with and we say, look, we can get an agent -based system to take on that particular world. I think we can do that and I and many of you you know will have will either be doing it trying to do it or we're thinking about it but there's a danger and that danger is we create systems that

are way more brittle than we should be doing and what will happen is that the world will change we'll have another you know another health scare or we'll have a another political problem or whatever. And as a result of that, things will break.

And now, instead of being able to get humans to come in and solve it, we will find ourselves in a situation in which that's hard, because nobody will know how this thing works. So I'm going to, that sounds very negative.

And I'm not, I am absolutely not a doomer. All right.

I love this stuff. I use it every day.

But I think what we need to be doing is look at how we decide to use it and how we are going to get the best out of it.

A More Constructive Path: Understanding and Using AI Well

There are several different ways of doing this.

Better Mental Models Through Stories

And I'm going to say that the first of these that I really like is I spend an inordinate amount of my time trying to explain this stuff to myself.

Okay, and I just want to share one story with you, which is my vision of what a large language model is. And I've got lots of these stories. This is just one of them.

But for me, a large language model is the essence of the internet. It's like taking the internet, putting it in a pot and boiling it away just as you would with a chicken. All right.

You boil it it away, and what you end up with is essence of internet. Now, if you add water to essence of chicken, you don't get a chicken, all right?

This has all sorts of implications. If you decide, for example, to ask a question of a large language model, and that question is, how often does it it make things up?

The answer is, it makes things up every single time. Not occasionally, but every single time.

It's just that most of the time, it's in line with what we believe is true in the outside world. So I like this.

And one of the things I'm going to suggest is that we should all be telling more stories in order to try and understand what this stuff is and what it's capable of. of.

Use AI to Think and Create, Not Just Automate

The second thing is that I think if you're given a choice and you're thinking, should I be using this to automate a task or should I be using this to help me think? It's fairly obvious we're using it to help ourselves think.

We should be building things that are better than chat. They may have all sorts of agents and things.

I'm not against agents, not against everything, but I think what we should be doing is helping you and me build ideas, explore ideas, make things, and that should be the target that we're looking at.

So the project of AI for me is better stories and making more stuff to help us do problem solving and help us

Conclusion

think one more slide not going to go through it but the particular think is don't think about optimizing optimizing the last thing we should be doing what we should be doing is diverging thinking of more possibilities and more ideas that's it thank you very much