Robot apocalypse ..The war has started

mwpt

kabayiri wrote: »

Lofti heights eh?

(Apologies to Mr Zadeh for a really carp pun).

I lost faith a bit in AI, I confess. There wasn't a shortage of self promotion, but a lack of meeting expectations. I introduced the first expert system on to the manufacturing shop floor of a company, and the boss there was disappointed that it didn't "teach itself to solve new problems". No pleasing some people...

:-)

Yes, expectations have always been too high, naturally. Same for most tech - where are our actual flying hoverboards? But I do think this is the era with the most promise. Deep Learning and related tech is going mainstream, anyone can download public domain code and play around with it.

We all know the consequences of AI are enormous, but consider this:

Do we have an imperative to develop AI at the fastest pace possible in a controlled fashion? I believe we do. The alternative being that someone develops it in an uncontrolled fashion, or worse, with malicious intent.

I think strong AI is inevitable. Even something approaching strong AI could be incredibly harmful if done with malicious or naive intent. Best we are ready for that.

mwpt

Here is another random persons blog thoughts about AGI (artificial general intelligence) in the form of deep learning.

https://medium.com/basic-income/deep-learning-is-going-to-teach-us-all-the-lesson-of-our-lives-jobs-are-for-machines-7c6442e37a49#.oyqu3grr7

This graph is very interesting:


One of the important points to stress here that may not be obvious to people not that interested in AI, is unsupervised learning. This for me as always been the most interesting part of AI because it means the start of something quite general. We've long had techniques where scientists/engineers/programmers can design weak AI algorithms to do things faster than humans. Fine, but those algorithms are very limited, to their specific task.

As the article says, we are now at the point where - in many cases - we no longer have to code algorithms to every task. We have a pattern/design/algorithm that is general in nature. The algorithm, in the form of a specialised neural network, behind AlphaGo is called deep learning, and it knew nothing about the game of Go or the Atari games, when the team fired it up to start learning. This same code can be run against training data from any of those tasks and as long as you feed it enough training data and the correct fitness function (winning, score, whatever) it will reshape it's internal structure to perform the task, and perform it extremely well.

As mentioned previously, people have been predicting the rise of world changing AI for a long time now, and deep learning isn't quite the panacea yet, but it's the start. My opinion is that we are now hitting a steep part of the exponential curve upward. I'm even more confident that before 50 years we'll see human level intelligence and before that we'll see general artificial intelligences doing, well, a lot more of the jobs we do today.

One thing I'm really hoping materialises out of this is interface to human brains. AI doing our jobs is all very well, but a combination of us and AI would be really interesting (and hopefully in the good sense). Want to learn to fly? No problem, download the module.

mwpt

Not sure why the graph didn't work. Try again:
https://cdn-images-1.medium.com/max/800/1*yPLHq5HEBTIs0VCdpe7KYA.jpeg

padington

I don't think even in a thousand years we'll have machines that can make better new innovative machines than us. The many many many years of our evolution just won't be able to be recreated in our lifetime.

We'll make pretty decent tools though in 50 years that do some things amazingly well and look pretty much like us and the things they excell at, they'll completely kick our ar*e at.

Mistermeaner

A stepford wife would be useful. Until some liberal gives them the same rights as human wives and then mistermeaner has to buy his house for a 3rd time

kaflinkle

I'm still a little ambiguous around what AI actually means. I also never think we are going to get what I would call artificial intelligence on the same level as humans.

Think about an idea or a problem that you solved. I myself whilst repairing something in the house or coming across a vexing problem at work have solved it by a complete leap away from logic and the solution has worked, an idea that has come out of the blue based on experience, curiosity and perhaps something else I cannot put a name to. This I don't believe could have been replicated by any AI.

Human intelligence can also produce great works of art be it music, paint or dance and again, how would AI accomplish this, it is something we feel.

But perhaps there are different levels of AI. As I said before, I am still a little confused as to what AI actually means.

mwpt

kaflinkle wrote: »

I'm still a little ambiguous around what AI actually means. I also never think we are going to get what I would call artificial intelligence on the same level as humans.

Think about an idea or a problem that you solved. I myself whilst repairing something in the house or coming across a vexing problem at work have solved it by a complete leap away from logic and the solution has worked, an idea that has come out of the blue based on experience, curiosity and perhaps something else I cannot put a name to. This I don't believe could have been replicated by any AI.

You are an AI. Your brain is a collection of atoms that has evolved over time to be trainable. There are layers upon layers of the brain, it's unbelievably complex, yes, but it's just a collection of atoms. We can eventually replicate that. Either through evolving other atoms or improving our own brains through hardware.

At what point does thinking life disappear and become a calculator as you start removing atoms from your brain? If you can answer that then you're smarter than everyone else. If you can't then you can't say we'll never create AI.

Basically it will happen.

Mistermeaner

We already know how to make ai machines that are at least as good as humans, theyre called babies. They learn stuff and can do things.

Why do we need another type of machine?

mwpt

Mistermeaner wrote: »

We already know how to make ai machines that are at least as good as humans, theyre called babies. They learn stuff and can do things.

Why do we need another type of machine?

What do you mean need?

Why do we need jacuzzis?

Mistermeaner

replace need with want

but your anology is also not logical; jacuzzis performance a service that is not otherwise obtainable (except for in geothermal areas), they are not a substitute for something else.

AI is arguably a substitute for humans