AI and humans

[The Groke]

Problem is that even a really stupid person, or indeed a young child could "outperform" an immensely powerful and complex computer in many of the areas in which it would be necessary to cater for when attempting to create a human-like AI.

Yes of course they wouldn't be able to calculate PI to a million digits etc, but the ability of a real person to react, and interact with any given environment, unexpected situation or to engage with another person will always be far superior than anything we can achieve with current technology.

Think of the subtleties of language and non-verbal communication which would need to be mastered to create a convincing human AI......staggeringly complex!


Even when considering a non-human AI - attempting to create an artificial replication of a slug, crab or indeed some kind of "low-end" mammalian intelligence, all of the above applies.

 

[lobster]

Think of the subtleties of language and non-verbal communication which would need to be mastered to create a convincing human AI......staggeringly complex!

Even machine translation is not that good, of the problems I have noticed while testing on-line web translators is that
when entering text from English to German and then back, what came back
was not what I originally entered. Surely if machine translation is to work
properly, I would get back the original text or something very similar?
Lastly word processing software offer grammar checking, however they are generally not very good.

 

[samk]

Surely if machine translation is to work
properly, I would get back the original text or something very similar?

I doubt you would get the same back in a human translation given the ambiguity in language, different ways of dividing up the world in different languages, and many ways of saying the same thing in one language. A human would hopefully give something that wasn't gibberish, tho

 

[jayeola]

yah, I've used bablefish very sparingly to translate some Spanish tech papers. Just out of curiosity I fed the Spa translation back to Eng and the results were shyte.

 

[gurrier]

yah, I've used bablefish very sparingly to translate some Spanish tech papers. Just out of curiosity I fed the Spa translation back to Eng and the results were shyte.

To some extent this is a function of the nature of natural language - think chinese whispers. There is a many to many mapping between words and meanings - there are more than one way to express a meaning and there is more than one meaning that can be attached to a phrase. Automatic translation is a lot better than it used to be but it's still a paradigm shift short of what it could and will be.

 

[jayeola]

hrm... that's interesting

 

[MadDruminFerret]

I wouldve thought our best bet at creating a human like AI would be making use of the decoded human genome... Our understanding of what the code actually means is almost non existant but what if we were to use it as input for some kind of growth simulation? its never going to be as simple as that though is it.. since you would need a simulated mother to grow it in

The only reason i come to this line of thinking is that our brains are so complex that reverse engenieering them seems to me an impossible task... so maybe a bottom-up approach is more feasible

 

[untethered]

Computers seem to have time on their side as they evolve a great deal more quickly than we do. Imagine if Moore's Law applied to human intelligence!

I can confidently predict that computers will be wiping the floor with humans in every area of intellectual endeavour within a hundred years at the outside, thirty years at a minimum.

Quite what hardware or indeed software they'll run we're yet to discover, but the foundations have already been laid.

 

[MadDruminFerret]

Microsoft Genius.. your digital superior

 

[samk]

I wouldve thought our best bet at creating a human like AI would be making use of the decoded human genome... Our understanding of what the code actually means is almost non existant but what if we were to use it as input for some kind of growth simulation? i

The problem with that is you would need to go down to the molecular level to run that simulation, and simulating a single protein molecule accurately needs more powerful computers than we have now.

Copying the brain without understanding it would be pointless anyway, there are cheap and established ways of making babies with the same kinds of flaws and limitations we already have

 

[Fez909]

I don't think we'll ever get to the stage of simulating human intelligence. I don't see the point, either.

AI should be about freeing up humans to do things which machines cannot do, such as the subtleties of language, and things which require abstract thought that is difficult to program for machines.

The good thing is that in trying to simulate human intelligence, we create these technologies anyway.

 

[xenon]

Slight tangent. but rather than getting machines to operate more like peple. In the BBC doc, Visions of the Future, Part 1. The Future of Intelegence.

Some speculate that within this century, humans will be fittted with intelegence enhancing chips. Things like Memory expansion, hardware acceloraters, heads up displays built into the eyes. Which begs the obvious question. At want point if this does transpire, do peple and robots become significantly comprible.

persoanly I think the idea of implants such as that, will take a long time after the tech exists, to catch on. Raises all sorts of questions about equity, species, rights and so on.

 

[Fruitloop]

Isn't the normal translation of MIPS Meaningless Indicator of Performance for Salespeople?

 

[dylanredefined]

Somedays i feel a zx81 without the memory pack could out perform me .

 

[jayeola]

Isn't the normal translation of MIPS Meaningless Indicator of Performance for Salespeople?

http://en.wikipedia.org/wiki/BogoMips

 

[samk]

Some speculate that within this century, humans will be fittted with intelegence enhancing chips. Things like Memory expansion, hardware acceloraters, heads up displays built into the eyes.

Except for that last one, doing these would need detailed understanding of how the brain works, which would help enhance AI, and probably with less leadtime than is needed to convince people the implants are safe.

 

[Jonti]

Computers seem to have time on their side as they evolve a great deal more quickly than we do. Imagine if Moore's Law applied to human intelligence!

I can confidently predict that computers will be wiping the floor with humans in every area of intellectual endeavour within a hundred years at the outside, thirty years at a minimum.

Quite what hardware or indeed software they'll run we're yet to discover, but the foundations have already been laid.

Transistors are The Emperor's New Mind

Point being that Turing machines just cannot do math like people do. Not now. Not ever.

 

[Hocus Eye.]

I expect that the first computer driven robots that we see in real life will be in the form of club bouncers. There will be a simple chip not unlike a calculator, a bit of techy stuff to enhance it, a speech analyser and a fist driven by an electric motor to knock out anyone who wants to argue.

More advanced versions will come in the form of robotic tabloid journalists employed to interview survivors of plane crashes etc. to ask standard emotional tear-jerking questions like "How does it feel to be the only member of your family to survive this disaster?"

The most advanced ones will be politicians who will run the country with one eighth of their computing power while fielding inane questions from the above robotic tabloid journalists with the rest of their faculties.

I think that the bouncers will be the most useful. There will be no need for further research.

 

[jayeola]

Hey - what about them working as civil servants? They could end up taking the flak for everything. "The Right Honourable Gentleman would like the house to join him in condemning R2d2/8 for losing the entire DNA database in the post"