Thursday, September 25, 2014

Review: Progress Toward the Singularity

This focuses on ways in which GAIs could interact with the material world, rather than GAI itself. In either domain I make no guarantee of currency; you will note that most of these links are 2-3 years old and have presumably been developed and found application since then.

1) Automated lip reading. It's not just HAL these days.

2) Remotely reading the position of people in a room based on how our bodies distort WiFi fields.

3) Smart sand.

4) Drones through your mail slot.

5) Computers that can spot microexpressions and lies better than humans, as well as microchanges generally (color and motion changes).

6) Stock trading programs doing things that we can't understand, faster than we could even if we did - termed by researchers in Nature a "new machine ecology" of software.

7) Generation of mathematical proofs that no human understands, but which can be shown mechanically must be correct. P = NP? If we can't understand it, then for human brains, NP doesn't even equal NP!

8) Reading your mind and knowing what you're looking at; they can reproduce images from movies as you're seeing them.

9) Forget chess. They can beat everybody at rock paper scissors too, because you're a lot more predictable than you know (but not than the computer knows). The best you can do is flip coins and come to a draw.

10) A fair amount of sports print journalism has been replaced by computer-written articles already.

For now, all these technologies are very much dependent on large amounts of money and attention from humans, and are in no way self-replicating. But we should expect that those places where the most utility is derived by humans are where we see powerful (read: expensive) new technology first put into place - like the stock market, and warfare. Ask Al Qaeda, who have been getting gradually exterminated by the HKs circling over Afghanistan for over a decade. (If we don't call the first truly autonomous drones HKs, that would represent a missed opportunity.)

Monday, September 22, 2014

Skynet's Ancestors: Are Groups of Humans Also Conscious?

I just re-watched Terminator 2, which is aging pretty well, not least because it used CGI as opposed to live action latex-and-juice effects in the first one that today look fairly bad. A refresher: Skynet sends a new, scarier shape-shifting terminator back in time, this time to kill John Connor as a 12 year-old instead of killing his mother as in the first one. To counter Skynet, the grown-up John Connor sends back a re-programmed terminator to protect him. It turns out that there were parts recovered from the first terminator from 1984, and a tech company has been reverse-engineering them - creating a loop where Skynet essentially catalyzes its own creation. A big part of the movie is the side-mission taken by the Connor family and their pet Austrian to destroy the lab working with the old terminator's parts, thus stopping the creation of Skynet.

Of course this plot was an excuse to get Schwarzenegger in another movie even though the terminator (well, the first terminator) was killed in the first one, and to make him a much more central character. But the most interesting thing to me isn't that, or even the paradox of a self-causing AI. Certainly the idea of fate figures prominently in the second and third movies, and many who think the Singularity concept is a coherent one also think that such an event is inevitable, given a technological civilization with enough time.

And that's why the most interesting thing to me about T2 is the police. (How many? "All of 'em, I think," says the young John Connor.) First of all, the parallelism is interesting: fine Skynet, you try to kill John Connor's mother? Well then we'll kill your mother, what do you think of that? (As a side note, because the shapeshifting Terminator made no attempt to defend Cyberdyne, either Skynet didn't think of this, or had already seen Terminator 3 and knew it didn't matter.)

1970's Colossus, which deserves to be more clearly remembered as introducing us to the idea of defense computers using the nuclear weapons under its command to sinister ends. (Although if you were Skynet, you could do better than something so obvious as a nuclear war.)

But much more fascinating than that: even though there is no fully functioning, self-perpetuating AI inside Cyberdyne systems, a more dilute but still discrete pattern has begun forming around the primitive Skynet's birthplace, a wealthy society's resources marshalled to protect the womb. And much like in our own tissues, Cyberdyne's individual T-cells (investors, police) cannot possibly appreciate the ultimate result of the developmental process they're taking part in. It's hard not to see the invisible hand of Skynet's intelligence, already embedded in the logic of the human scientific enterprise and capital accumulation and the institutions we've created to advance those processes.

To be clear, I don't get the impression that Skynet was somehow magically reaching back through time and hypnotizing LA's finest (and Cyberdyne's investors) to protect its primordial ancestors, or that this is what James Cameron was suggesting. But it's interesting to contemplate not just in light of the concept of the Singularity, but of the serious philosophical debate about the basis of consciousness, and to what kinds of entities it can apply. If you think consciousness comes from a pile of neurons being connected, then does it matter whether some of the neurons are inside separate skulls, connected by speech rather than synapses? That is to say, if you believe that one person is a conscious entity, doesn't that mean two people must form an additional conscious entity, as would any combination of humans? So things like Wyoming, and Honda Motors, and investors in the stock market, are conscious entities! As would be Cyberdyne Systems (and its associated financial and security apparatus) - and Skynet is the eventual expression of the converging currents of that particular self-organizing system.