Human level AI in two decades

[walker 0]

Hi all

Worth bringing to everyone's attention.

http://tech.uk.msn.com/news/article.aspx?cp-documentid=7553206

With the way the AI in ArmA can beet some people in ArmA I think for some Human beings the day has already arrived. "It is not cheating it is just brighter than you are!" will start to become some thing we hear more and more in the future.

AI abilities seem to be expanding rapidly I would say at the moment ArmA AI is about at the level of a slug. COD 4 is about intelligent as an amoeba.

In simulated environments the real problem is sensing the world. We need AI that can sense from the same data a player in an FPS does that there is a door in front of them and that those are stairs. Then we need a AI that decides on the balance of its wants and needs whether to go through the door or up the stairs.

Kind regards walker

[bn880 5]

In two decades quite possibly. It would have been achievable today if we developed parallel/concurrent processing (especially analog systems) instead of branching off into higher sequential digital only performance.

Those will be the days, we may re-discover everything.

[sanctuary 19]

Sounds nice ... until the 3rd decade in which Skynet takes over the world.

[dm 9]

With the way the AI in ArmA can beet some people in ArmA I think for some Human beings the day has already arrived.

The AI beats people not because it is smart, but because it has none of the cognative problems people do.

It isnt looking for you via a textured model, but via a plain cube in an equally plain background. If its line of sight happens to interscet your cube, bam it spots you, rahther than having to double check that what it is seeing behind that bush is actually a camo uniform and not just the leaves moving.

When the AI aims, it has a pre deterimined spot to aim at, which it does perfectly every time. Dispertion and a few other AI tricks should mean that for the most part, rounds will whizz by, but more often than not you end up in the "superman AI headshots you at 500m with the first round of AK fire" situation.

"Human AI" is more about the entity "thinking" and "acting" rationally/irrationally. Hiding, fleeing, flanking, using cover, scavenging correct equipment, setting up traps and ambushes, pre-planning events and then changing those plans according to the flow of the event, thinking up tactics on the spot, doing unexpected things which arent pre-programmed into its "brain" (i.e. thinking like a human). And that is still a LONG way off.

[dmarkwick 261]

And that is still a LONG way off.

What, like 20 years off?

[wamingo 1]

Another nostradamus wannabe I say...

I predict ... That advacements in processing power will end before AI will be able to interpret its surroundings at the same speed as humans (I'm a pessimist). Afterall, there's an astronomical amount of things that humans can observe every moment. Like facial expressions. The minute detail in complex things and what to make of it.

We don't always make the right decisions though... So maybe it won't be hard at all, heh..

In terms of computer games, well, AI cheats, it doesn't interpret pixels like we do. ArmA's AI is more about simulating the faults and limitations of humans than anything else.

Making it miss in a convincing manner. Getting it to respond to your decisions. Fearing death like humans would. Getting it to make decisions other than just standing still, even though lying in some bush would probably be the most effective way of beating you. Etc.

[ebns72 0]

I hardly can see how human-level AI is achievable in 20 years when we just barely understand how our own "AI" (The brain) works and how complex it is.

[bonko the sane 2]

never underestimate the power of human stupidity vs artificial intelligence.

But being a great admirer of Neuromancer and alikes, im really looking forward to this level of AI, so my espresso machine and toaster can make me breakfast exactly the way i like as im getting up

[ryankaplan 1]

Yeah, bollocks. We were supposed to wear silver suits while we were driving our hover cars powered by a Nuclear microplant in New New York in Mars by now. Catch my drift?

In 20 years? Maybe AI that acts believably human in certain videogames due to millions of lines of code, but not 'human' level ai which can learn from its mistakes/successes. Although, the former would be enough for me anyway.

[Sc@tterbrain 0]

Hmm a "futurist". These are the same people that predicted that today we would have:

-Flying cars

-A Moon/Mars base

-Solved world hunger/disease

-Robots to do your chores (no the Rumba isn't even CLOSE to counting!

-Peace/End of the world

-A perfect high-speed mass transit system

-One minute microwavable Mac&Cheese, and frozen pre-made PB&J sandwhiches

....thank god we have food for lazy people at least!

[Halochief89 0]

You also have to think communication with the AI. I mean one of the first communicating with AI was "Hey you Pikachu" (other then its lack of understanding curse words, I said it was a f**king disgrace and it blushed) But now such as Rainbow Six Vegas and its sequel the AI will actually talk with you.

In 20 years maybe they'll be like Eddie from the movie stealth; asking us questions and learning ways to speak against us.

That also gives me the idea future consoles/computers will speak with you. Turning on a Xbox-(whatever) and having yawn and say hello "your name" having intelligent conversations. I mean within 20 years gaming has become very impressive from 1982's Donkey Kong with no AI to F.E.A.R.'s extremely intelligent AI that actually tries to attack you when your near death and not just hide behind a pillar.

[Sc@tterbrain 0]

Characters in video games are a VERY simplistic example of true AI.  But what you describe shows how far "AI" has come in this very specific and limited use.  

When dealing with AI in the way this article suggests, it is about "thinking" and "learning" in a more human sense.  It's about being able to interact in the physical "real world" environment rather than just in a simple game setting.

If the topic of Artificial Intelligence interests you than I recommend reading "Artificial Intelligence: A Philosophical Introductuction" by Jack Copeland.

Its a bit dated in the technology aspect, but still on target for the theory of AI development.

For a quick introduction, have a look here.

http://www.alanturing.net/turing_....AI.html

*EDIT

bn880,

I was just looking through my copy of the book (been a while since I read it) and saw that it discusses the difference between classical symbol processing and parallel distributed processing.  It seems the author also thought that would be a better way to go back in the mid-90's.  This book really does a great job in explaining the two different approaches.

[baddo 0]

It's just not going to happen.

To me "human level AI" has to mean something which is able to do all the things I have done throughout my life, intellectually, and I don't see a computer algorithm ever being able to follow the same intellectual learning and experience path that I have had (or any one of you others).

Some neural network can probably be trained to be very good. But "the little something" is always going to be missing. A computer is basically very mechanical: a bit is either on, or it is off. This is how simple a computer is. By combining bits you get more complex algorithms. But how many bits can you add. How many bits can you combine. How intelligent those bits are. At maximum, no-where near as intelligent as you are, Mister Computer Scientist. The time of the World will run out before you have added enough bits and combined them intelligently-enough, to reach close to the human brain.

Also consider this. A human brain is an organic system. It is able to recover from damage. If artificial intelligence is said to be at the level of the human intelligence, then it has to be able to fix itself when it breaks down. Without any outside help. When someone grabs a part of your brain and throws it away, your brain will start to re-route its messengers. How is your electro-mechanical device going to do the same.

It is fascinating to think of a machine being intellectually at the same level as humans, but I think it is an utopia, it will never be reached. Great stuff for Hollywood though.

[jakerod 254]

It's just not going to happen.

To me "human level AI" has to mean something which is able to do all the things I have done throughout my life, intellectually, and I don't see a computer algorithm ever being able to follow the same intellectual learning and experience path that I have had (or any one of you others).

Some neural network can probably be trained to be very good. But "the little something" is always going to be missing. A computer is basically very mechanical: a bit is either on, or it is off. This is how simple a computer is. By combining bits you get more complex algorithms. But how many bits can you add. How many bits can you combine. How intelligent those bits are. At maximum, no-where near as intelligent as you are, Mister Computer Scientist. The time of the World will run out before you have added enough bits and combined them intelligently-enough, to reach close to the human brain.

Also consider this. A human brain is an organic system. It is able to recover from damage. If artificial intelligence is said to be at the level of the human intelligence, then it has to be able to fix itself when it breaks down. Without any outside help. When someone grabs a part of your brain and throws it away, your brain will start to re-route its messengers. How is your electro-mechanical device going to do the same.

It is fascinating to think of a machine being intellectually at the same level as humans, but I think it is an utopia, it will never be reached. Great stuff for Hollywood though.

Isn't our entire nervous system similar to a computer. Our nerve endings say "Hey that hurts!" (On) or "Hey that doesn't hurt!" (Off). Obviously there is a bit more to it than that but essentially I think everything in the human body can be pretty much replicated in an on/off manner. The only thing that might be difficult is sight and smell.

A computer could fix itself. If we lose part of our brain our brain reroutes messages. If a computer gets damaged if it has the means to reroute data then it can do that. However it has to be given the means to do it. If we had no way to do it then we couldn't do it either.

Bottom line is if you give the AI and/or computer the means to do something then it can just like humans. If we were never given the means to see, smell, touch or hear (plus the other senses) then we could never do it.

Some of that logic might be faulty. Sorry if it is I just got off work and I am a bit tired.