AI character

I have an idea to create an AI character who is roughly human-level in intelligence, but has realistic computer-style strengths and weaknesses. And I'm pondering what he/she will be like.

Probably will be The Rain Man. Poor sensory processing - difficulty recognizing objects, especially in unexpected orientation; needs to learn how each person talks and is very poor at understanding speech from unfamiliar people; recognizes people by their eyes, so can't recognize you if you wear sunglasses or something. Very good at rapid, brute-force processing, but has trouble seeing patterns and making meaning out of things. What else?

If I'm asking for advice on a story idea, don't tell me it can't be done.

Well it depends what kind of sensors it's hooked up to, these days there are computers stat can just about recognise your face moving through a crowd at an airport, so I wouldn't say sensory processing would be particularly weak. An inability to prioritise maybe (you could program it so save lives, but as in 'I, Robot' it would calculate the chances first and act according to them, not according to human values), a lack of emotion, even a lack of understanding of emotion.

Try grabbing a copy of Robert Heinlein's 'The Moon Is A Harsh Mistress' and read the bits about Mike, it might give you a few clues (more knowledge than an family, any community even, could hope to be able to remember at a whim, but all the understanding of a toddler).

It's a difficult question. However, I'd say that pattern-matching is one of the essential aspects of consciousness, so I'm not sure about making that neutered.

What kind of physical structure are we talking here? Camera network? Android?

[1] This facsimile operated in part by synAC.

I dabble in AI myself as a hobby, not that I'm some guy working to put us into a singularist but the first question you should ask because I find this an issue with everybody who portrays AI... what was the original AI's purpose? Was it building cars at a factory? Managing spaceships at a starport? Making sure the traffic lights are correct?

Was there even meant to 'be' an AI (there certainly wasn't in 'The Moon Is A Harsh Mistress', or at least not from where the characters stood)?

I'm not sure that I'd trust Heinlein for actual information on how computers work, Matt. Like most sci-fi writers of his time, he was more focussed on telling a story.

[1] This facsimile operated in part by synAC.

Yeah I would have to agree with Tzeze here. I've coded a lot of A.I.s and if you want to say "What would it's personality be like if intelligent", it can get pretty esoteric if you want it to be accurate. The most important thing to ask, if you want to know how it *might* act if sentient, is what its original purpose was.

Hey, he posited nothing about now they work (or not much anyway), but Mike was at least as plausible as any 'accidental AI' before or since. Oh, and Heinlein probably knew as much about how computers work than any sci-fi writer these days (given that all we seem to get these days are Asimov type robots).

Well it's more or less a vehicle for someone who thinks different from the society of the author so they can reflect on some stuff.

Like playing up the logical part to act as a commie-expy... or playing up the evil controlling AI to play up the fear of authoritarian governments or fascism. Or perhaps as a robot it can't have any feelings and that makes it sad. (Hey how do you turn a phrase into a link to another wiki page?)

If you wanted more realism, it gets weird. Say you've a paper clip factory AI. All it cares about is making really awesome paper clips. It might not even be able to talk properly. Just speaks in factory-order forms. But it's really adamant you get that paper clip right!

Haha. No. Try hard sci-fi writers sometimes. They don't know how to shut up about realism.

[1] This facsimile operated in part by synAC.

Heinlein is a hard sci-fi writer, her can hit level 5 or 6.

Yeah, well most of Heinlein novels were more about his crazy social views that went from being a hippy to a fascist dictator. That's quite a switch for anybody. But I guess when you make a novel, even if it's a hard sci-fi, focus on the story since that's why you read the novel.

Well at least he didn't go for Asimov's Three Laws, those are absolutely poisonous (define 'human', define 'harm', then try to work in criminal laws, no wonder VIKI from the 'I, Robot' movie was so friggin stuffed up). You also have given no reasons why the AI protagonist in 'The Moon Is A Harsh Mistress' is at all unrealistic. Oh, and a paperclip factory AI is not going to become sentient because it's never going to get expanded to the point of having sufficient 'brain-power', whereas the computer 'Mike' from Heinlein's aforementioned novel was running everything, the catapult, the life support system, the communication system, etc. so even a 'what you're programmed for' approach can't be used.

I object to Heinlein's views being called fascist. "Authoritarian" perhaps, if we go by Starship Troopers, but I'm simply not going to accept the use of something as ill-fitting as "fascist" applied to anyone whose ideology (at one point; the back cover of my copy of the Moon is a Harsh Mistress describes it as a "great libertarian novel", and his latter works were increasingly Reagan-esque, while the guy was a self-proclaimed liberal who loved The Forever War) is not based around some flawed nationalistic version of social Darwinism.

Now that I've opened that can of worms, I seem to recall that Elezier Yudkowsky thought that AGI personalities, as they go, will be as diverse as human personalities, because they'd evolve from genetic learning algorithms (or whatever it is you kids use these days) and heuristic learning, and therefore be as diverse as human babies.

However, they do have the advantage and disadvantage of fundamentally different hardware, and what is generally difficult for a modern computer to do (speech recognition, image recognition, facial recognition, detecting emotions, etc.) would probably be generally harder to do for an AGI, just like humans beings generally are not as proficient as computers at mathematical and logical processes.

Of interesting note though, is that given the correct hardware or software, an AGI-based intelligence will be a lot better at measuring things, especially size and distance, and everything that can be derived from that; speed, acceleration, etc.

Things I like: Ghost In The Shell |Serial Experiments Lain |Eden: It's an Endless World! |Sid Meiers Alpha Centauri |Aeon Natum Engel

^^ Wasn't the whole point of Asimov's robot stories that the Three Laws were inherently flawed anyway?

Blind Final Fantasy 6 Let's Play

^Well he originally came up with them to avert the cliché Robot War plot (didn't work), but, yeah, a lot of them are about that.

What does "AGI" mean?

And basically impossible to implement practically.

It's been a while since I've read the book, but how about, how does he know how to talk without social peers? He can tap phone lines, sure, but that ought just be noise to somebody who doesn't even have a concept of language.

edited 6th Oct '10 7:48:36 AM by Tzetze

[1] This facsimile operated in part by synAC.

Yeah, they were, but the biggest issue would not be taking them to their logical conclusion, it would be trying to make them work in the first place. Oh, sure, physical harm might be easy, but what other forms of harm are there?

'He' (for want of a better term) was programmed to understand Loglan, and probably could translate certain phrases of other languages. Of course, this wouldn't set him up for talking in English, but his intelligence was soon discovered by the local repair-map (after he paid a cleaner 10 trillion credits over what he should have done).

edited 6th Oct '10 8:01:56 AM by MattII

@Tzetze: AGI is for Artificial General Intelligence, and it's basically a way of saying "strong AI" or "human-level AI". I'm sure prescience could tell you more, but he doesn't seem to be around anymore.

*sigh*

Things I like: Ghost In The Shell |Serial Experiments Lain |Eden: It's an Endless World! |Sid Meiers Alpha Centauri |Aeon Natum Engel

I just had an interesting thought. One of the major ways of optimizing is to turn algorithms into recursive functions.

So a recursive function is one that calls itself repeatedly to solve a problem.

Maybe you could have an AI character that thinks in terms of recursion (not easy but you know it'd be interesting).

Well, a recursive function isn't necessarily better than any other, and the computer doesn't have infinite stack space. (In most environments, nested calls around 1000 deep will break something.)

Da Rules excuse all the inaccuracy in the world. Listen to them, not me.

Right, I forgot about that... but that makes the "accidental" a lot harder to swallow. Programming a computer to "understand" a human-usable language would probably result in an AGI anyway.

That seems a bit to esoteric to smoothly integrate into a story.

And Yej, all the cool kids use tail recursion.

Or more seriously, recursion that deep doesn't usually come up outside of math problems and stuff.

edited 9th Oct '10 2:57:05 PM by Tzetze

[1] This facsimile operated in part by synAC.

He wasn't originally intelligent, it was just that when they finished hooking the entire life-support, communications, banking, etc systems up to him he had 1 1/2 times as many neurones as the human brain. He died at the end, got lobotomised by a surgical missile strike, and even after they hooked everything back up he never came alive again.

Turning something recursive does not and cannot optimize it.

Anything that can possibly be expressed using recursion can also be expressed without it.

The main reason recursion is actually used is that in a lot of cases, it leads to more elegant and readable code. However, there is no difference from the machine's point of view.

Blind Final Fantasy 6 Let's Play

Heh, I didn't mean for you guys to interpret recursion being itself the optimization but that optimized formulas in a lot of cases are recursive solutions. We express them in iterative format for good coding practices. Like, bubble versus merge sort.

Anyway, author thinks my idea is too esoteric I'll think of something else. I would always figure that the AI character would be really good at a few things but really bad at others. For instance if the AI were a soldier, it'd be a brilliant strategist, marksman etc but terrible at understanding civilian life.

OK, more about my AI character:

A teenage boy got into a car accident and his brain was completely destroyed. His parents donated his body to a research laboratory that installed an AI brain in him. So this character has a human body, but a mechanical brain.

The main plot is that he and two psychic kids (The Empath protagonist and a Seer) have to flee when terrorists attack the lab. They team up with two other kids and take down the guys who did it.

If I'm asking for advice on a story idea, don't tell me it can't be done.