A.I & super computer cloud.

bogwalrus

Just reading an article on how scientists are building a supercomputer to work exactly like a human brain. Its costing 1 Billion and a ten year program.


Pretty amazing and it got me thinking about A.I


Techincally if you build a super computer with a brain and combined that with cloud computing you could have hundreds of android robots wirelessly connected to this super brain.


So basically just building one super brain is enough to give us our own personal robot slaves (or masters)


thoughts?

Capt'n Midnight

AI has always been 20 years away

they've even developed silicon that mimics the way neurons work at a hardware level rather than programming


the real problem is that we still don't understand how the brain works

look at the AI chatbots there hasn't really been a dramatic leap since Eliza

syklops

AI is here already, its just not what we were promised. To quote a recent Wired article on the matter:

Today’s AI doesn’t try to re-create the brain. Instead, it uses machine learning, massive data sets, sophisticated sensors, and clever algorithms to master discrete tasks. Examples can be found everywhere: The Google global machine uses AI to interpret cryptic human queries. Credit card companies use it to track fraud. Netflix uses it to recommend movies to subscribers. And the financial system uses it to handle billions of trades (with only the occasional meltdown).

Zombrex

bogwalrus wrote: »

Just reading an article on how scientists are building a supercomputer to work exactly like a human brain. Its costing 1 Billion and a ten year program.


Pretty amazing and it got me thinking about A.I


Techincally if you build a super computer with a brain and combined that with cloud computing you could have hundreds of android robots wirelessly connected to this super brain.


So basically just building one super brain is enough to give us our own personal robot slaves (or masters)


thoughts?

The problem with AI is that it will always rely on the software and as yet we don't even know what the "software" of our own brain is, so we cannot replicate it in a computer.

It always amuses me in sci-fi when AI devices attempt to kill humans under the idea that "it is trying to protect itself, it is evolving".

Why is it doing that? Did you write it to do that? Biological life evolved defense by virtue of the fact that if it didn't it wouldn't still be there. This is not an inherient property of life. A Skynet situtation is not going to happen unless we program a Skynet situation, the AI will have no more a strong desire to live than your toaster will.

So ultimately we need to seperate out the idea of artificial life and artifical intelligence.

One interesting way to go with AL is by setting up an evolving system and just seeing what happens. We may evolve virtual life that behaves more like real life this way if we put in selective pressure that forces evolution.

But given how unpredictable this would be it need to be separate from machine learning and AI.

Capt'n Midnight

Zombrex wrote: »

It always amuses me in sci-fi when AI devices attempt to kill humans under the idea that "it is trying to protect itself, it is evolving".

http://xkcd.com/1046/ :pac:

Morbert

Zombrex wrote: »

A Skynet situtation is not going to happen unless we program a Skynet situation, the AI will have no more a strong desire to live than your toaster will.

This is true, but there is a (hypothetical but by no means probable) danger that is similar to the Skynet situation.

The more sophisticated and adaptive you permit an A.I. to be, the more control you relinquish over possible solutions to posed problems. If we ask an A.I. to develop an optimal set of socio-economic relations to minimise war and conflict, for example, it might decide to wipe us all out, or to chop all our heads off and submerge our brains in morphine solutions.

Capt'n Midnight

Morbert wrote: »

This is true, but there is a (hypothetical but by no means probable) danger that is similar to the Skynet situation.

The more sophisticated and adaptive you permit an A.I. to be, the more control you relinquish over possible solutions to posed problems. If we ask an A.I. to develop an optimal set of socio-economic relations to minimise war and conflict, for example, it might decide to wipe us all out, or to chop all our heads off and submerge our brains in morphine solutions.

why would it ?

It's not been exposed to 4 billion years of Darwinian evolution.


We don't have AI we have 'fuzzy logic' , we have programs that mimic complex behaviour. IIRC we can barely replicate the pre-programmed responses of an insect. We still don't have any generic learning algorithm, chess computers are still number crunchers that investigate vast numbers of possible outcomes rather than get a feel for the games. Also in chess endgames are fairly simple since there are less pieces so databases of optimum solutions can be looked up.

It's only recently that a computer, with a handicap, can beat a world class GO player even though it's about the simplest game to describe to a computer.

We still don't have an AI that can look at a game of GO and get an instinctive feel for the relative strengths of white or black have over various areas on the board.

Morbert

Capt'n Midnight wrote: »

why would it ?

That is an optimal solution to the posed problem. 0 people means 0 wars and 0 conflicts. I.e. An A.I. would find solutions to problems that would not necessarily be in our best interest. Similar to the way we use condoms to more efficiently fulfil our genetic "orders" to have sex.

Again, I stress that this is all in the realms of fantastical speculation. I am not expecting the Zen19D Go program to commit mass genocide.

Capt'n Midnight

Morbert wrote: »

That is an optimal solution to the posed problem. 0 people means 0 wars and 0 conflicts. I.e. An A.I. would find solutions to problems that would not necessarily be in our best interest. Similar to the way we use condoms to more efficiently fulfil our genetic "orders" to have sex.

Again, I stress that this is all in the realms of fantastical speculation. I am not expecting the Zen19D Go program to commit mass genocide.

There is a difference between SciFi and Fantasy

I'd consider Asimov's laws of robotics as pure fantasy.

First off an AI wouldn't have any concept of it's own mortality, in that it could survive forever , it could also duplicate itself, it could possibly distribute itself.

An AI would figure out a way to get itself into the space program so it wouldn't be stuck on this planet.

Morbert

Capt'n Midnight wrote: »

There is a difference between SciFi and Fantasy

I'd consider Asimov's laws of robotics as pure fantasy.

First off an AI wouldn't have any concept of it's own mortality, in that it could survive forever , it could also duplicate itself, it could possibly distribute itself.

An AI would figure out a way to get itself into the space program so it wouldn't be stuck on this planet.

I don't think you're following what I'm saying. I said "The more sophisticated and adaptive you permit an A.I. to be, the more control you relinquish over possible solutions to posed problems.". So, for example, if we assigned it a problem like "implement socio-economic policies that will minimise war", it might implement, or at the very least, suggest, solutions that are not in our best interest, even though they are technically valid solutions. This is analogous to the way our brains have devised clever solutions to satisfy our instincts, even though such solutions do not propagate our genetic material.

Capt'n Midnight

Morbert wrote: »

I don't think you're following what I'm saying. I said "The more sophisticated and adaptive you permit an A.I. to be, the more control you relinquish over possible solutions to posed problems.".

That's why we have SciFi and all those leaking nuclear power stations to let us know that we need to be wary of safety and that a fail safe is kinda important (or at least a kill switch)

Bricks are more reliable than people.
But people design systems too.


Oh yeah we don't know how computer chips work anymore.
that is to say they are too complex for a human to understand

Capt'n Midnight

http://www.theregister.co.uk/2013/02/04/human_brain_project/

The European Commission has selected the Human Brain Project (HBP) as one of its Future and Emerging Technologies and will send it up to €1.19b over ten years so it can build a supercomputer capable of simulating the human brain.

...
“With today’s technology, an exascale computer capable of simulating a cellular-level model of the whole human brain would probably consume about a Gigawatt – billions of euros worth of electricity every year.“

so we are still a long way off.

Morbert

Capt'n Midnight wrote: »

http://www.theregister.co.uk/2013/02/04/human_brain_project/

so we are still a long way off.

There is no doubt we are still a long way off. However, depending on your criteria of what qualifies as a human brain, it would not be as expensive to run. For example, you could have a sophisticated neural network, with all the plasticity and efficacy of a human brain, but with an incredibly slow firing of action potentials. I.e. A brain capable of cognition and abstraction comparable to a human, but on a much larger time scale.

Lbeard

Capt'n Midnight wrote: »

why would it ?

It's not been exposed to 4 billion years of Darwinian evolution.

When the first homo sapien, came from Africa to Northern Europe, he noticed it was cold. He saw a bear. He clubbed the bear, and then had a warm fur coat. He didn't need millions of years of evolution to grow fur.

In computer science there's always a lot of hype and vapor. The founders of Google may look like geniuses because they are billionaires, but their original search engine was just a bog standard data base, with a simple scoring system for entries. Cloud computing is the same network computing that's been around since the 70s.

MIT claimed to have mimiced a synapse with 400 transistors. Are we close to Skynet?.....there are an estimated 100 trillion synapses in the human brain. That would be 40,000,000,000,000,000 transistors. Xilinx have a chip that has 6,800,000,000 transistors. That's 58,823,529 Xilinx chips, for one brain.

Using Moore's Law, based on the other figures, I've calculated we're 45 years away from being able to put a human brain on a chip.

But are the computer industry really working on this? When it comes down to it, 99.9% of all software development, is not working on some futuristic technology that will break us through to the next level. It is cruddy data base apps. Google is just a database, Facebook is just a database - even porn sites are just databases . The Cloud is just full of cruddy database apps. And the hardware development is virtually all platforms to run more databases on. Computers are largely glorified filing cabinets, you can watch LOL Cats on.

But to create something that can surpass the human brain may already be already in reach (an array of a thousand Xilinx chips, under Moore's law is only 20 years away). The mechanisms that make the human mind may be simpler than they seem - and knowledge/knowhow of the world may be something that's accumulated through something like the Bayes theorem. Things like curiosity, a will to live, can all be preprogrammed.

amen

Oh yeah we don't know how computer chips work anymore.
that is to say they are too complex for a human to understand

Really ?

Capt'n Midnight

amen wrote: »

Really ?

You could understand the interconnects of a billion transistors ?

Morbert

Capt'n Midnight wrote: »

You could understand the interconnects of a billion transistors ?

Well yes. Quite easily. I am involved in chip research every day.

Capt'n Midnight

Morbert wrote: »

Well yes. Quite easily. I am involved in chip research every day.

Humans can understand a subset of the system. But no one can have a mental image of all the possible interactions.

Or to put in another way you can hand over an ASIC design to a silicon foundry and be pretty certain they won't be able to reverse engineer it. Higher level VHDL description costs more than the Tape Out version.

Morbert

Capt'n Midnight wrote: »

Humans can understand a subset of the system. But no one can have a mental image of all the possible interactions.

Or to put in another way you can hand over an ASIC design to a silicon foundry and be pretty certain they won't be able to reverse engineer it. Higher level VHDL description costs more than the Tape Out version.

That is a very very very different statement to

"Oh yeah we don't know how computer chips work anymore.
that is to say they are too complex for a human to understand"

Squeaky the Squirrel

Capt'n Midnight wrote: »

http://www.theregister.co.uk/2013/02/04/human_brain_project/

so we are still a long way off.

American study ready to go on mapping the brain, hoping to solve parkinsons, Alzheimers and other Mental diseases.