Bits The story of artificial intelligence

The story of artificial intelligence, from chess to the possible end of humankind as we know it...

http://www.bit-tech.net/bits/2012/03/19/aisearch/1

 

My Miele washing machine has more intelligence than some people I know

 

I finished my degree in AI and CS many years ago. Almost always the main bottleneck in creating useful programs was coding all the rules, limitations and options.
would be ideal to have a new programming environment where the user already has this done and can drag and drop common rules like lego.
eg. find fastest route to top of building, you shouldnt have to code gravity limitations, and walls etc. every time.

 

@PingCrosby - Lol. So true.

@ greigaitken - But, doesn't that negate the meaning of AI? 'Intelligence' implies the ability to integrate current knowledge to solve problems and increase that knowledge. If you just have a 'list reader' then it's not really intelligent, just a large storage of data with an efficient recovery mechanism. (Sounds like Google).

Sentient beings learn by doing (mostly). I think it is this ability of our brains to be plastic that makes us intelligent. IE - Learning to walk is a constant alteration of neuronal connections until our reflexes and central nervous system balance the movements required with our weight and gravity. No one tells us the value of gravity, or the weight of the limb and it's fulcrum.

 

@fodder:
Essentially, it's not all that fundamentally different.
We learn the use of our legs as well. Theoretically, given a few base rules, an AI is able to uild HUGE repositories of "learned data", which it could use to come up with new solutions.
Procssing that data is what computers struggle to do.
Just the basics of walking took years to get right, as there are so many factors to consider.

 

I don't think we'll get close to mimicking true human intelligence in a computer whilst not really understanding how human intelligence works.

Compared to a computer, we're unbelievably strange and illogical. Many of our decisions are made unconsciously by a section of the brain we have limited control over. Our memories fluctuate and change, and our opinions endlessly shift based on stimulus that we receive in an enormously varied number of ways.

The fact our brains are biological, that they are built of cells that live, die, change and change again means our intelligence is in a constant state of flux. Add to that endless chemical influences from internal and external sources, and the result is so endlessly illogical and random it'd be impossible for a non-biological entity to ever replicate it truly.

I'm not saying you won't ever get intelligent machines, they just won't have the same intelligence as humans.

 

A while ago, the Skeptic's Guide to the Universe podcast did an interview with Michael Vassar, the president of the Singularity Institute for Artificial Intelligence (linky: http://www.theskepticsguide.org/archive/podcastinfo.aspx?mid=1&pid=218). It was pretty interesting stuff.

Interesting read; it's not something that I would have expected to find on bit-tech... Professor Sharkey may be right, however; at least for the time being. It's important to remember that no matter how much computing power you have or whatever algorithms you program, computers are fundamentally just big adding machines; albeit machines that can add up at inhuman speed. We really don't know how the human brain functions on such a fundamental level; until we know that, it may well be impossible to design a machine that can truly replicate emergent and "intelligent" behaviour. We've come a long way, but we're nowhere near that goal yet.

Incidentally, on the subject of Alan Turing, you should all go and visit Bletchley Park at some point. My girlfriend and I visited it around two weeks ago, and it was fascinating; if you do go, make sure to take one of the guided tours - our tour guide worked there during the war and knows so much about the entire site and all the work that went on there. There's quite a bit about Alan Turing there (including Gordon Brown's official apology letter), as well as the (fully working) rebuilt bombe, the Colossus and a huge collection of Enigma machines...

Plus you can also visit the National Museum of Computing and marvel at the four-foot diameter hard drive platters from the 60's/70's which only store 4mb per side; or get all nostalgic about all those old computers that you grew up with, now consigned to museum exhibits.

 

completely off topic and train of thought de-railing but...

anyone know where i can get a bigger/full-sized version of this:




... carry on

[i'll read this and comment appropriately when i'm a little less dead...]

 

would this one do?

 

Agree 100% with this.
There are a few things that are IMHO not simulatable: Emotion and Intuition to name but two.
Computers can pretty much by definition make "cold" decisions quicker and more logically than us (humans), but they lack a human factor that is so very important. Just the speed at which, without stepping through all the logical paths, a human brain can draw a conclusion about something based on nothing but what is describied as "gut feeling" is VERY hard, if not impossible to simulate with typical logic-based decisions.
In pure rationality, though, a computer (especially a self-taught system) would out-do us, though... /2c

 

Ahem.. "Most recently, the construction of the Large Hadron Collider hastened rumours that an artificially created black hole could destroy they entire planet"

I think you mean THE entire planet

 

Unless you count "has legs" as a defining characteristic of AI, this happened in the late '90s with Kismet, or even earlier depending on your definition of AI. Which is one of the biggest problems in AI research: the definition of AI, and it;s ever changing nature.

In reality, we have LOTS of AI. We use it all the time. Let me introduce the concept of 'hard AI' and 'soft AI'.

Hard AI is an artificial consciousness. This is immediately made a woolly definition by the choice between an artificial human consciousness (i.e. the only sort of consciousness we have encountered) or any general sort of consciousness, which thus far the definition of which has been hard to pin down.

Soft AI is a learning system. Oh, wait, we have those SLNN things, and they're certainly not really AI so that can't be it.
Soft AI is a system that can learn and produce emergent behaviour that is not programmed in or trained. Oh, wait, the Seven Dwarves and other swarm robots can do that, and they're certainly not really AI, so that can't be it.
Soft AI is a system that can recognise visual stimulus and classify them as known objects. Oh, wait, Google Goggles can do that, and that's certainly not really AI so that can't be it.
Soft AI is a system that can recognise and respond to natural spoken language. Oh, wait, Siri can do that, and that's certainly not really AI so that can't be it.

And so on. The goalposts of 'soft AI' move continuously as the next frontier is reached, achieved, and becomes mundane.


And once again, I must take issue with Sharkey's odd view on machine emotions and machine empathy (specifically, the lack thereof) and of the non-computational nature of the mind.
I'll address the latter first: If the mind is not computational (i.e. cannot be simulated) this implicitly requires you to accept Cartesian Dualism; that there must be an intangible something, a 'soul' or the like, that makes up the mind. This is silly.
On machine emotions: we have emotional machines. A machine can be 'afraid' or 'angry' or similar. We encounter these often: game AI. These aggregate states are machine emotions (e.g. when 'scared' an AI will attempt to hide from a player's sight, where otherwise it might wander looking for 'food', unless hiding would put it in danger in which case it would overcome it;s 'fear' enough to move into the player's sight and flee, before hiding again).
But what about empthy? We can make machines that recognise facial expressions (you can even download openCV and play with it yourself). We can make AIs that 'feel'. Input the detected emotion as an influence on the machine aggregate state (it's emotions) and Voila! You have an empathic machine.


I leave you with a final thought on the 'sentient machine apocalypse': a conscious AI would have a creator, and thus a parent/parents. When we grow up and become independent of our parents, we do not go on a murderous rampage. Why would we? And if not, why would an AI? We would have taught it all it knows, so unless we teach it that Destroy All Humans is good, then it's unlikely to think that way.

 

oh man the old ai debate.. of course it could only be thought of by people who have never programmed anything.. programming and the hardware behind it are extensions of the programmer.. it cannot learn or do anything outside that box it's given

to put bluntly- when trying to create an ai out of non-living parts.. you've already lost.. course people with thier head in the clouds will never accept this.. there must be a way for a machine to think and write it's own code..

think of it like this.. when you write a program- and your trying to simulate the thought process.. your missing the one thing we have, that it can never have.. a soul if you will

sure you can have it process input and react to that input.. but to have it actually do something inspired outside of it's programming isn't going to happen.. you could have it act spontaneously with a random number generator here and there but it isn't thinking for itself- true or false, on and off, if then- that's what your dealing with

oh well I'll leave this here.. think can see a reflection of the guy who built this thing in the visor

Image Removed

How many times TheHippoz. You know that stuff like that isn't OK, so why do you do it?

 

The fact Noel Sharkey was quoted makes this article awesome.

Anyone else remember Robot Wars?

 

Having seen thehippoz's input into topics of a similar nature in SD I think it can safely be said that he doesn't really have much, if any, of a clue when it comes to these sorts of things. You're best just letting him have his say and moving on without acknowledging it tbh.

 

bah had this conversation before a long time back.. I came to the conclusion everyone in the field has finally come to- only this was years before

http://forums.bit-tech.net/showpost.php?p=2597022&postcount=20

just because you read it in a science fiction novel or watch movies doesn't make it true.. it would be cool to tell you the truth- but really not much of a chance imo.. now mixing the two is interesting but considering we only use a small portion of our brain as it is.. probably better time spent figuring out how to access all of it instead of going that route..

you guys can believe otherwise.. we need people like that anyway- it's like the chupacabras keeps your brain turning when your out in the field late at night.. then you see it get in the car with some dood

 

To put it bluntly: you are completely wrong. Not only is it possible to programs to write their own code (polymorphic viruses and other malware will demonstrate this), but code certainly can 'learn and or do anything outside that box it's given'. Take Boids for example: You program an AI with three simply rules of avoiding crowding, aligning with neighbours, and aiming for the average position. These are pretty simple behaviours, but the behaviour that emerges is complex: flocking.
That's not to mention things like SLNNs and learning classifiers. These do bugger all when programmed, you have to train them with input data to get them to classify correctly. Different training data will result in different outputs, for the same code written.

 

Do we have some all singing all dancing AI tech that is equivilent to a person?

Nope.

Are we close to that?

Nope.

Do we have some very impressive techniques that could be perceived as intelligent?

Yup, plenty.

One of the reasons A.I is so complex is that we dont have a definition of what we mean by intelligence, which makes it an even harder goal to achieve. Certainly we have things like the Turing test but that is by no means a definitive test of intelligence in a computer system.

 

Just what I was about to write on, having cogitated over the reply to my last post.

Maybe, what we are trying to think of as AI, is self awareness. Once something is self aware, it becomes more self guided in it's decisions about the world it inhabits. You could program conditions into an AI to cause itself harm, but if it was self aware would it then follow that programming or decide to follow another string of code instead? (I am not a programmer, so that's my take on it).

In terms of the '1's and 0's' being all a computer can do, we are the same. A neurone fires or it doesn't. It may take several other neurones firing at the same time at the synapse to get that neurone to fire, but it is still 'on' or 'off'. Put inhibitory neurones into the equation and you can start to get a system that starts to look analogue despite being 1's and 0's at the operating level.

 

arh yes the miele washing machine is a bit of engineering genius, well mine is

http://www.neatorama.com/2008/09/05/10-most-fascinating-savants-in-the-world/