Artificial Intelligence through emulation of 'real' intelligence is a valid and interesting field in areas where a "fast, good" answer is better than a "right" answer.

For example, there's an experimental neural-net powered system that uses photographic images captured from video cameras to look for known suspects (let's not beat around the George W. here, we're talking about terrorist suspects). We simply don't have the computing power for it to come up with a perfect, "conventionally" computed answer every time: that would take too long, and wouldn't cope well with crowds of people.

An artificially intelligent system modelled on one the the ways that we theorise the human brain works will come up with a good answer in acceptable time. Sometimes it will miss a suspect, and sometimes it will produce a false positive, but it's still got several advantages over a human: it works every hour of the day, watches many screens at once, never sleeps or takes a toilet break, and produces accurate, hight-speed output, to name but a few.

Artificially intelligent systems are useful everywhere that a human would be more useful than a computer: in making decisions based on past experience or based on unknown variables, in providing "fast, good" answers rather than "slow, perfect" answers, and - of course - in emulating human or animal behaviour for computer games.

They are, as you say, more "stupid" than conventional algorithms. But then, so are humans, but we still use them for all kinds of tasks, too. And to answer your question: a machine that displays emotions has, potentially, exactly the same purpose as a human that does the same. And unless humans basically are "RNGs that spew out emotions", then the difference is clear.

"people who can put aside (or appear to put aside) these personal values when making decisions are frequently considered to be more rational and more intelligent than people who can't."

...which I have a problem with. Putting aside your emotions when making a decision doesn't mean the outcome will automatically be better. Emotions are a form of intelligence too. And at the end of the day a judgement which neglects or undervalues your own emotions or the emotions of people who'll be affected by that judgement is in itself flawed - if you don't value people what do you value?

I know this isn't quite related to your post. And I know you weren't making the point I'm arguing against. I just find this discussion interesting.

Perhaps what people developping AI are trying to do is make things which can combine the efficiency and power of a computer with the instinctive intelligence and compassion of a person? Maybe getting a computer to understand abstract variables would still result in a 'flawed' judgement, just perhaps a more efficient flawed judgement. Rather than a rigid one which doesn't take into account or value human feelings. After all, whatever ethical implications there are in making a copmuter/robot/whatever that can 'feel', we're making these things to serve us. How can they serve us if they don't understand us?

Maybe getting a computer to understand abstract variables would still result in a 'flawed' judgement, just perhaps a more efficient flawed judgement.

Exactly what I was saying, but not so concisely.

and I didn't even see your comment before I posted.

AI isn't just about simulating the way people think. It's much bigger than that.

The US air force developed a system which knew how to fly an aircraft, one of their jet fighters. It was designed to learn how the airflow was working over the wing and how to interpret the pilot's input (pull back, aircraft should lift the nose, that kind of thing). The point of the system was for when an aircraft was being shot at and the wings got damaged, this system would quickly learn (ISTR it took 0.2 seconds) how the wing had changed and modified itself to let the pilot carry on flying the plane normally. The plane would correct everything needed, so as far as the pilot was concerned, the aircraft behaved just like normal. The gaping hole in the starbord wing suddenly wasn't a problem.

Another example was what my PhD was supposed to be about. Say you've sent a flying robot to Mars, it can take anything between 20 minutes and an hour for a signal to get from Earth to Mars and the same amount again for you to get any data back. There's no way you can fly the thing in real-time - the plane will have crashed before your first signal's even got there. What you need is a program to fly the aircraft for you, with you just giving high-level commands, like "fly over there for me". The problem comes when an aircraft encounters something you'd not expected. In traditional programming there's no room for something outside of the bounds you're expecting. What happens if the aircraft suddenly gets caugt by a gust of wind and finds itself flying upside-down? If you've not expected that to happen, things could go wrong very quickly. If instead you've taught the computer to fly the plane, it can hopefully recover from this situation itself.

The point is that most AI is actually about allowing a computer to inteeract with the real world, not interacting with other computers. The real world *does* act unpredictably and trying to to programmatically trap every possible outcome is impossible.

AI isn't about getting a computer to act in a human way (although there are branches of AI that look into that), it's about using the ways in which biology solves practical problems for interacting with the world around us and applying the same methods to computers, and seeing what happens.

Gareth

"You might as well have an RNG that spews out emoticons..."

I think you've just found your final year project...

--Jon

heh, 'spews'.

heh, 'emotions'.