How to Build Truly Intelligent AI

[neuron]

There is a ongoing joke in our community that if it's written in PowerPoint it's Ai if it's written in python it's Machine learning or deep learning

[bigbuckey]

Really wish there was more substance to this video. "AI needs to be able to generalize" wow excellent observation how do you propose we do that.

[derasor]

Big fan of Dr. Mitchell, she is right about trying to find a path from fundamental storytelling in the brain to simulating a current model of a situation at hand in the machine, to predict the next 'frame', in order to tackle intuition and common sense. How to map 'abstract thought' to define situations with math is the big conundrum IMO. If this could be done, a similarity score could be used that takes into account different analogy dimensions (of action, subject, etc)... Anyhow, she is a TRULY brilliant and beautiful human being. Cheers!

[ErixSamson]

"Godel, Escher & Bach" by Douglas H. is a MUST READ!!!

[SoCalFreelance]

So nobody is going to ask where the glowing forest is located?

[freddyfourfingerz]

For great A.I. examples in literature and television (humanized A.I.'s), I recommend the show person of interests, the novels involving SIMPOC, and the expeditionary force novels.

[yourpersonaldatadealer]

Surely it’s just a case of looking for similar groups (or two separate groups that have a union subsection) of neurons lighting up to beyond a certain threshold (e.g. these movement neurons fire with 80% of the pattern of these other neurons, therefor there is an abstract relationship between the stimuli).

[XOPOIIIO]

But machine do make analogies, otherwise they couldn't play Go, because brute force is not working there. We just can't crack how to make them to do this more efficiently. The breakthrough could be far away, or it could be made tomorrow. Deep learning did not reach limit yet, it's actually the fastest developing computer science discipline today.

[mikedziuba]

Before you can do analogy, you need to discard all unnecessary information and create abstract models of the things you are comparing. For example, if you ask a small child to draw a man, then he or she will draw a stick figure of a man. And if you ask the same child to draw a dog, then he or she will draw a stick figure of a dog. And if you compare the man's legs with the dog's legs in these drawings, then they will look more or less the same. And when their legs look the same, then you immediately see the similarity between the dog and the man.

The man will have two stick touching the ground, and the dog will have four sticks touching the ground. So, the child might say that the dog is like a man with four legs.

If a computer can create abstract representations of a man and a dog with stick figures like a child, then this computer would also see the similarity between the dog and the man. Because those sticks that represent legs would be the same for both the dog and the man.

So, a child can create abstract models and then reason with these abstractions. And that's what today's AI can't do. Today's AI only deals with concrete objects and doesn't create any abstractions from these objects. And of course, without any abstractions, today's AI can't reason with the objects either.

[SeaJay_Oceans]

560, 000 years ago, A.I. thinking, '' How can I grow truly intelligent natural O.I. ? '' (Organic Intelligence)

[luciojorgelourenco]

I think a good tip I can give, although I am not a scientist, is that if one wants to build a machine to work like us, one should think in quantum entanglement of different particles in a way that it would work like this: One computer makes perfect decisions and perfect actions and the other entangled to it would make a "proper" decision and a "proper" action. Thereafter, both would be dealing with memories about what happened and discussing to each other, what would create consciousness.

[SportDubs]

Like reading the entire internet in a millisecond and knowing what its output should be.

[tomfeiler]

Yep, I always wondered if the scientists are dumb or if I was just to conceited, but shows they really seem to narrow their view on a problem to such a degree that they don't see how they are completely off course. That is a big problem I think, they should sometimes just ask intelligent students or people that work in another field if they would agree with their results. Deep but restricted knowledge is useless if it comes to solving real world problems.

[brown]

Programming psychology and emotion, is analogous. I would be interested to see what you build.
But watch that algorithm close. It could get upset, at least in the beginning of it’s learning phase of how to relate and reconcile things full of emotion.

[morkovija]

2 year olds had a few hundred thousand years of evolution to prop them up. They did not evolve their abilities from scratch. I am certain those features she is referring to will be emerging side effects on sufficiently advanced model

[JeffreyBoser]

I have a problem with this right at the beginning... We do NOT have computers that can recognize objects. We have algorithms that can analyze data we give them in a certain way we tell them to, looking for very specific PRE-DECIDED things. We have yet to make a computer than can take any input and play around with it and learn, a computer needs to be told every step of the way to do everything. It can't even recognize if a plug is plugged in wrong.

A human doesn't have to be told their ears are there, or how to start looking for patterns in what they are picking up, or how to connect them to patterns from what their eyes are picking up, and building and building and building contexts. Even if a human never learned what ears were, they can't help but use them. The biology is active, and initiates this itself, sometimes incorrectly. But function in biology is not implemented, it is inherent.

What we are headed toward is making better algorithms, we aren't making them more 'intelligent'.

[Rahul-dk]

One of the most my favourite person concerning AI and ML.

[alzard]

How can I become a scientist? I am a software engineer, but I did not finish school. Do I have to finish some kind of uni? Or can I self educate myself and get hired somehow?

[illustriouschin]

I'm glad she qualified it with "truly" intelligent because that's what we are, truly intelligent! HAHAHAHA.

[benkasminbullock]

I must have missed the "How to build" part, where is it?