ArtificialIntelligence

Quora Answer : Artificial Intelligence has been around for a long time with Lisp and Prologue but did not have widespread adoption. What has changed recently in the last decade to make AI more useful?

Apr 3, 2019

Basically two things :

faster and more powerful hardware

larger data-sets for training.

The second of these is probably more important than the first.

I was in academia with people doing neural network research in the 90s. My first job using AI in 1991 used neural networks with a couple of hundred nodes and a few dozen training examples. We didn't have more training examples, and you couldn't run many more nodes than that on our off-the-shelf PCs.

Today people would find that ludicrous. How can you pick up any patterns with that?

Well you can't. (Just toy, very well prepared, examples.)

And today people use thousands of nodes and millions of examples.

That data is largely available thanks to the last 20 years of the internet, where we've been gathering massive collections of human writing, photos, and other media types which we can now build frighteningly plausible models from.

We have so much data today. The internet collects it. And the business models of the internet giants like Facebook and Google and Amazon means that there's a mature and sophisticated market of hard and software to manage enormous collections of data that be fed to the deep learners.

There are also two approaches to AI.

The "symbolic" which Lisp and Prolog used to focus on. And the "statistic" or "connectionist" ie. data-driven AI.

I don't think either is inherently superior to the other. Both have been around since the dawn of computing in the 1950s. And there have been plenty of religious wars.

But they are usefully complementary.

I suspect today we are just throwing more resources at machine learning / connectionist AI. I wonder how powerful our Prolog systems might be if we gave them equivalent computational resources and programmer time and attention. Possibly equally amazing.

But the nice thing about connectionism is that you don't really need smart programmers. You just need a lot of data, and can already get a long way with off-the-shelf free-software packages.

EDIT : See LanguagesForAI

Quora Answer : Will artificial intelligence happen?

May 14, 2020

Originally Answered: Is true artificial intelligence even possible?

I have an MPhil in Artificial Intelligence research. Although it's basically about game theory and agent based modelling of very, very simplistic behaviours.

Like many young impressionable geeks, I went to college to study AI and to be involved in finally creating "true artificial intelligence".

Somehow I got side-tracked into hanging out with philosophers. And through them I came to realize that "true artificial intelligence" just takes you down the rabbit hole of trying to figure out what "true intelligence" really is.

At some point, nearly everyone realizes that you have to give up worrying about whether the computer is "really intelligent". We can't even know if other people around us are intelligent or conscious. Or whether they are just philosophical zombies or an illusion.

AI is about producing intelligent behaviours. Not about "truth".

Quora Answer : In ten years how powerful do you think AI will be?

Aug 23, 2019

We don't really have good metrics.

But AI will be powerful.

Let's put it like this : an AI bot will be able to fool the casual observer that it is a real person via video chat. It will be able to synthesize real time video of a non-existent person talking to you. It will be able to interpret your speech sufficiently to know what you are talking about and talk back to you sensibly about the subject.

It won't always sound like a "smart" person. If you are deliberately trying to catch it out, and make it say nonsense, then you will probably still be able to do that. But if you are casually dealing with it, without caring whether it's a person or not, then it will work smoothly.

Meanwhile, AIs will be routinely and successfully working with people in every aspect of "intellectual work", from accounting to the creative arts.

AI vehicles and robots will be increasingly prominent in most physical / mechanical work, from the construction industry to surgery.

Quora Answer : What are the possible inventions in artificial intelligence?

Mar 22, 2020

Yes.

But the kind of AIs we have at the moment are still very much driven by humans.

We choose to turn them on, feed them data, fine-tune them, ask them specific questions.

What an AI invents at the moment is very much under human guidance.

If by inventing "on its own" you mean, completely under its own motivations, to start with you need to give the AI a kind of "autonomy". Probably its own robot body that it's capable of sustaining (in terms of energy, self-repair etc.) And have its intelligence hooked to its perceptions and attempts to survive.

Such AIs need not be super-intelligent. Insects have autonomy without much of what we think of as intelligence. But their bodies are well adapted to survive in the environment. And their limited intelligence does solve the typical problems they face (how to navigate from here to there, how to find food)

"Inventing things" isn't really much more than problem solving in a new domain. The intelligence of today's AIs is more than sufficient for that, if they are given the right information, configured the right way.

If they are put in a position where their bodies are sufficient for that, long term, we might well start to say they are inventing by themselves. But most AIs won't be in that position, because humanity is building AIs to work for us. So most AI creativity and invention will be at our service.