Notes from New Sodom

... rantings, ravings and ramblings of strange fiction writer, THE.... Sodomite Hal Duncan!!

Saturday, October 07, 2006

AI - Mark 1 & Mark 2

As a break from the whole Strange Fiction series, I thought I'd take brief sojourn into a realm I'm truly unknowledgeable about, but am going to opinionate on anyway. There's a great debate going on about it here between Ben Rosenbaum, Ted Chiang, David Moles and others, who clearly have more grounding in the literature than I do, so go here if you want something more than my wild riffing. Otherwise, here goes...

1. Artificial Intelligence

That would be like the intelligence that told us Saddam had WMD, right? Intelligence that was, um, artificed? Like:

-- We've got some intelligence, sir; it says that Sodom Hussein does not have WMD.

-- My fellow Amerkins, can you excisify the negatorial... for the sake of this great nation of ours... and freedom... and democricy?

-- Well, sir, then the intelligence would be wrong.

-- My fellow Amerkins, no, it would just be artificialisated... for the sake of this great nation of ours... and freedom... and democricy.

No? That's not what you mean? Shit, cause that makes sense to me, in a way that most talk of AI doesn't. Let me try again. So is it, like, robot spys... cause that would be cool! Like, a separate robotic intelligence network, working alongside Military Intelligence and the Central Intelligence Agency?

-- Sir, the CIA tells us-

-- My fellow Amerkins, I don't care what the Central Intelligence Agency tells us. I don't care what Military Intelligence tells us. All I care about is what Artificial Intelligence tells us... for the sake of this great nation of ours... and freedom... and democricy.

(I would hope that this AI is better designed than the animatronics they're using for President Talbot Dubya Yancy, but I have my doubts. Sorry, I digress.)

So, yeah, OK, I know that's not what we mean by AI. My point is that intelligence is a word that means a whole lot of things to a whole lot of people. You can be called intelligent for knowing lots of shit about the world. You can be called intelligent when you don't know squat about the world but you can figure shit out. Knowing shit and figuring shit out, memory and reason, can be simulated in databases and chess programs, so strictly speaking we already have AI. The machines are already clever.

But the whole cleverness thing is a smokescreen. That's clearly not what we mean by AI. So I'm going to avoid the semantic dust bunny of intelligence as much as possible, try and look at those different things we mean when we use the term in other ways, by other names: having knowledge; being able to apply it; being able to glean it from the world around; processing one's environment into information; reacting appropriately to different circumstances; and whatever else comes up.

First then, let's look at the most basic and physical of those -- reacting appropriately. After all, the question of whether a machine can interact with us as an intelligent being surely depends on the question of whether it can interact at all. So...

2. Artificial Interaction

Here's a thought experiment in AI. You're sitting with a laptop in front of you, with a word processing program open. You point and click to open a document. The document opens. You hit the key marked "A". The letter "A" appears on the document. And so on. If the laptop's behaviour is appropriate to your actions -- it doesn't open up random files, fail to carry out commands for no obvious reason, crash when you go off to get a cup of coffee -- is it interacting with you?

Seems like a no-brainer to me, here. Of course it's interacting. It may not be exactly proactive in terms of the interactions between you and it. It may not suggest the resolution to that story you've been pissing about with for weeks but, well, you do shit to it and it does shit to you, even if the latter is just showing you the words you've typed. Hell, I've used plenty of programs that have interacted with me a whole lot more than I'd like ("Yes, we'll switch off all those automatic correction options, thank you very much, Mr Microsoft Word."). Move things on to the level of handwriting-recognition or voice-recognition functionality and you're very much talking about a machine that interacts with you. Hey presto, AI... Artificial Interaction.

But, but, but... I hear you say. What about that whole proactivity thing? This is just a machine responding to your actions, albeit in a fairly complificated way. It's still a passive object rather than an active agent.

Well, we'll get to that. See, that's largely a question of whether a machine can interact with its world the same way we can, which is to say intelligently. Shit, though; there's that word again, the one I wasn't going to use -- intelligent. Let' me rephrase. If the interactions between my laptop and me are those of an object -- stimulus-and-response, cause-and-effect -- rather than an active agent, what do we need to add so they qualify as those of an active agent? What else are we looking for in AI? Well, seems to me we're looking for evidence of this... thing we reckon that we've got, where something weird goes on between stimulus-and-response, cause-and-effect.

The Turing Test and the Chinese Room thought experiment posit communication as good evidence for this... thing. The interactions they focus on are linguistic exchanges, where a human constructs this thing called a question, feeds it into a machine, the machine analyses the semantic units and their syntactic structure, and responds, using a set of semantic units and syntactic structures from the same language to construct this thing called an answer. Most of the actual communication (the hearing and speaking part) for the machine is just mechanics, same as for us humans with our vibrating ear-bones and wiggly tongues. So what we're talking about is the analysis-and-response part. What we're talking about is Artificial Interpretation...


Post a comment

<< Home