I'm a software engineer with minimal knowledge of AI ... but I like to think of it as very very sophisticated curve fitting with clustering.

Imagine a 2D graph with a single straight line (vector) and a single cluster, it's very easy to predict what X will be for a value of Y.

Now imagine a graph with 1800+ dimensions with billions/trillions of vectors on it with them clustering around millions of points - that's an AI model.

Well I think people referring to AI as algorithms or models are not appreciating the difference. What AI is doing is effectively simulating brain development - the establishment of neural networks and interconnections that allow pattern recognition and reconstitution in self adapting, probabilistic, Bayesian and heuristic ways.

It still has a long way to develop and we are just going through another new tech hype cycle just like we saw with the internet 30 years ago. Eventually it will find its place as a tool, in the meantime we get all the silly hyperbole and bullshit as every CEO pretends he knows all about it and asks his IT director to create an AI strategy for the company - because...well you know the city scribblers keep asking him and he will look like an idiot if he doesn't follow the fashion. I worked on Knowledge Based Systems and inference engines with tools like Smalltalk, LISP, prolog etc for a few years back in the late 1980's. There was a wave of hype then. We are undoubtedly much further on now and have hugely more compute power but I think still far from high 'intelligence'.

The reality is a good chunk of the human race is not that much different from AI in principle - brains that have absorbed data and patterns and regurgitate them in response to stimulus. Think of how a baby learns to recognise faces and objects.

However we still seem far away from proper 'intelligence' in the sense of real consciousness or the ability to actually generate and logically explain new concepts. It is quite rare in humans and certainly people like Newton, Einstein, Ramanujan and that great list of intellectual giants are themselves rare in humanity. I think of intelligence as understanding and the ability to explain. Like Feynman said knowing the name of something may be useful for communication but doesn't mean you know or understand what it actually is in a meaningful sense.

I can see AI hugely outperforming humans on things like medical diagnostics. In principle a well trained AI would never forget that rare foreign disease long forgotten by the GP or nod off while examining hundreds of scan images. And I doubt that an AI would give you 10 different bullshit answers to a plumbing problem in the same way human plumbers do.

The thing about passing the Turing test is it depends how bright the person asking the questions is.

Agree with this. GPs must be the most overpaid and over educated people relative to their actual output, which seems like any robot AI can do.

I hope this is one area AI really takes over, think of all the cost savings for the UK!

I can help people to understand this AI thing and where we are right now. There is a lot of confusion, hype, concerns, BS, sceptics, fan boys. People like ben ski aren't very good at explaining things in a straight forward and logical way so you just end up with more noise and hyperbole.

Let me give you three examples of what a system like chatGPT (an LLM - large language model) can do right now...

A) I can ask a question like "what is the difference between pine and oak" or "was the Commodore Amiga 500 a good computer?"

It will gather information that is already out there on the Internet and provide me with a useful answer. How good the answer is will be a matter of opinion.

B) I can ask it to draw me a picture of things. An apple, a chimp, whatever.

It will gather data on what these things look like and provide a 'drawing' of them. I could give more challenging requests like "man called ben, who looks like Woody Allen, loves monkeys" and it will provide a visual response. Again, matter of opinion how well it will do in each case.

C) I could ask it to describe an episode of the Simpsons in the style of Shakespeare and it will do that.

It will scan data about that episode, like the script, reviews about it, etc. and then ingeniously make it into a Shakespearean type of poem. It probably won't be perfect and there might be a few mistakes about how it interprets what the episode was about.

The first thing to say is: this is incredibly impressive. It is something we didn't have before but do now. It provides these responses almost immediately.

Two things:

- It is not showing any originality whatsoever. It cannot replicate the human mind for that. It is limited by the available data which has been produced by humans. There isn't any evidence that it will become genuinely original or creative in future but some people claim that it could/might/will.

- It is an incredible data-gathering tool. This is what it really demonstrates. You want information - it will get it for you and present it in the way you want it, which is quite articulate, and even ridiculously so if you just hapoen to want it in the style of Jane Austen or whatever.

So it is a tool. It isn't the master and might never be. It will enhance what people can do in the same way that calculators and phones and cars have. But it isn't intelligent in the way the human mind is. Even if it draws amazing pictures of apples in future - we don't suggest that a digital camera is intelligent just because it can produce an image of reality better than we could ever draw. AI goes further than the camera by answering a request we have made by gathering information and then formulating that into an image. Humans can obviously do this as well, but we are not necessarily limited by the available data online or our core programming (will vary from one person to the next).

Also, you might notice that, despite all the information available to it, it hasn't yet become an Islamic fundamentalist or hard right ideologue - it just happens to have the moral belief system of a woke, technocratic, corporate drone. Because that is what it has been given as its moral code by the people who designed it. Just like the people who designed it :)

---

How much better it gets from here remains to be seen. It might just be a case of incredible progress now and then diminishing returns over time. Think of it like computer games. There was an unbelievable period from around 1990 to 1999 when they just got better and better and better, and the computers needed to keep improving to keep up with the software. A lot of the best games now are arguably still the ones from the mid to late 90s. What has happened since? Well the graphics have kept getting better but that's mainly it as far as I know. In the last ten years, some improvements in functionality but nothing all that ground breaking. Diminishing returns.

It may be so with the AI thing. So it will tail off and we just happen to be in the midst of the really great leaps forward at this time - which capture people's imaginations.

Now, you might say: what's so good about presenting information that is already online into a report in any style you like? Or drawing stupid pictures? Well, this sort of thing is impressive - let's be honest - but it is somewhat of a novelty. I think the main benefit is in the ability to speed up and process huge amounts of information and to recognise patterns. It will ultimately be for a human to use that tool for his own ends and to check the results though. It's like how AI can already drive a car but would you really want to sleep whilst it drives you on a long distance journey? Airline pilots are still needed even though planes can more or less fly themselves, because sometimes things go wrong and the human master must intervene. And those people must maintain and practice their skills to ensure they can meaningfully intervene when needed. This is a key point I will return to.

It will be used by bad actors as well as good. The ability to pretend to be somebody else and adopt their writing style... a boon to the Nigerian economy. Its use for propaganda purposes. An efficient tool to enhance the ability of governments to deliver all the crap we don't want. Also, does anybody really imagine that, even if it were able to, it would be allowed to point out that the so called climate emergency and net zero are a load of bollocks? And it will be useful for research and in myriad ways in people's every day lives. But again, people will ultimately be in charge and the sensible ones will ensure this is so - lazy idiots might not.

In terms of people's jobs. Yes, it will be able to replace at least some. I'm pretty sure the bog standard daily mail articles are already written by AI. It will depend on the jobs though and to what extent there is decision making involved. It will depend on how well the system and process can be done by AI. If there are ethical or political decisions involved, physical activities, etc. Only at the basic end though really. A lot of jobs are already fake and unnecessary though and yet they still exist. So there's more to it all than what makes economic sense. The future will continue to be human and therefore messy.

Some people say medical diagnosis... well, I think we've all had the experience of asking Google about an ailment we are suffering from and recoiling in horror as cancer appears as one of the first results. It seems obvious that software trapped inside a computer will struggle to diagnose somebody saying they have chest pain. It will require human involvement but the AI might enhance the process. Maybe it will mean that doctors with six years of training and earning all that money can focus on the more difficult things and a lower level administrator could be helped by the AI to do a lot of what GPs currently are doing. Or maybe GPs go extinct. There are all sorts of permutations.

There are are some who claim that AI will create a lot of new jobs. I'm not exactly sure how but perhaps if it helps to open up new areas of understanding and industries this might indirectly create work for people. I don't see how it can take manual jobs unless the AI (which is software) is augmented by robotics or other types of machinery (a vehicle would be an example). It may be that the nations that have more service based economies will suffer first and most. Why would processes done in the UK by people need to be done here in future by AI contained in servers? And then: how does this country earn an income to pay for its imports from other countries or the benefits all the unemployed people will supposedly be paid? Perhaps geopolitics and economics will need to be fundamentally reconsidered. Maybe not.

Note that the last few paragraphs have gone into how it might be used and how it might change things. But this is all speculation. Even assuming it develops significantly from here there would need to be the personal and political will of people and peoples to embrace it and adopt the possibly groundbreaking changes that it gives rise to. This would be challenging to say the least. All the normal human drives will remain. It will also be very risky to embrace the speculated advanced technology too whole heartedly because we are already getting more and more stupid and lazy. Outsourcing most, if not all, of our thinking and skills might not be a good idea. The increasingly fat and dopey slobs who follow such a path might end up paying the price in a multitude of ways.

Main takeaway - it is interesting, it is fun, it can do things we don't really need and it can do things that can help us achieve useful goals. It will be used by good actors and bad. It isn't original or sentient. It is limited by the available data. Humans are in control of it. How it changes from here: this is speculation. And as always, there are all kinds of very human factors feeding into that.

Well me being me, I tried this. And his claim seems absolutely true



So there you have it. ChatGPT is racist, in that it differentiates on skin colour.

Ok... lets dig..



Meh...



So... did it 'learn'...?



Now did it actually learn anything...



So no. It learns nothing. Its just a database/search. It clearly has no intelligence at all. The woke programming is fixed and everyone will see the same.

It's obvious that someone, somwhere designed in the bias. You are seeing deliberate human imprinting it's operation.

Finally



Bias, of course, is not 'inclusivity'

I gave up at this point. I learned something, even if ChatGPT did not.


One last try..



So logged out and in.... and guess what...