The most powerful artificial intelligence is a neurotic artificial intelligence. That’s the upshot of a study by a team of Austrian IA researchers, reports New Scientist. The researchers programmed software robots, or bots, to have four different emotional states – aggressive, defensive, normal and neurotic – and then had them play the military strategy game Age of Mythology. The neurotic bot, which “was more likely than the others to distort hard facts … and flip between extremes of behaviour,” outperformed the other three. Like the aggressive bot, it won all seven of the games it played, but it won them much more quickly than the aggressive bot. “Neuroticism is a competitive advantage,” conclude the researchers.
I find this news strangely uplifting. It means that when our computer overlords finally take over, we can at least take comfort in the fact that we’ve created them in our own image.
While this is mere science as fable, I’ve long thought that the test of true AI will be when it has mental illness (Marvin The Paranoid Android?)
In terms of their methodology, I think that it is too bad that they did not test against a simpler (albeit, less glamorous) open-source game. I have played different versions of vms-empire for over 20 years. Such would allow other researchers to more easily confirm the results. vms-empire features “fog of war”, a simpler resource management and transportation logistics. Of course, its AI is very primitive but the same ideas apply: anytime you encounter any of the enemy’s scouts, you “overreact” and send most of your forces in that direction in hopes of finding the main body. Your own immune system works that way. You hoard your military resources (much as the USA hoards fissile material) because there is no big penalty for having “too much” of a global resource. OK, maybe on something like naval ship where space is limited, but not on land. It is somewhat like Andy Grove’s book “Only the Paranoid Survive.” Of course, one does not just want to survive, one wants to thrive. Thus, in our personal hygiene, some of us create near-sterile environments for ourselves and our children and then wonder why those immune systems are underdeveloped (its because the immune system is under-stimulated and has failed to “learn”, but that concept is certainly not intuitive).
“Computer overlords” taking over? I don’t think so. The hoi polloi may be ruled (present tense) by profiles, dossiers, credit scores and the like all stored in computer data banks (just as the films Brazil, 1984, Blade Runner and other intelligent futuristic storylines predict) but the tyrant (or term-limited President or whatever) at the top will be human. He might even have some computer providing advice directly, but the top decision maker will be human. I do not expect our future to be Terminator (even though that Austrian actor runs Cali-fornia these days) because we do not understand enough about what allows most human minds to pass the Turing Test while AI continues to fail. Some aspects of life are like Chess (a game that the best machines now dominate) but real life is still analog. We are amused and even to some degree educated by robots like Marvin, maybe as our maids and litter collectors, but we certainly do not want him performing appendectomies, being the top-level decision-makers in our airliners or having his finger on the button.
Neurotic bots can rule in even more arenas. Think of them as following a contrarian strategy. Wall Street has long recognized that you need an edge or else market efficiency will completely marginalize your results. Being neurotic thwarts logical behaviour and gets these bots past the rank and file.
So it is with the web. The web is a very low friction environment. Inject a new meme and it bounces around and reaches equilibrium very quickly. Hence big leaderboard/aggregator type systems like TechMeme usually don’t tell you what’s interesting. Likewise, the most interesting Facebook applications are hidden out in the long tail.
If you want to play the neurotic bot strategy, you’ve got to create punctuated equilibrium around your idea:
http://smoothspan.wordpress.com/2007/10/01/the-internet-first-breeds-diversity-then-conformity-punctuated-equilibrium/
Go ahead, the bots can do it!
Cheers,
BW
The fundamental flaw in their work is that behind the fancy GUI game interface, its still just electricity flipping switches on and off. These machines simply have not been demonstrated any kind of complicated behavior that suggests to anyone watching them that they are aware of their surroundings.
For all their talk of bots simulating emotion, in reality, simulators are just random number generators filtered through program logic that acts as a human proxy. In this case the random numbers are filtered to produce inputs that the programmers think would be generated by human players with the emotional characteristics of aggressive, defensive,normal and neurotic. And, like one of the previous posters astutely observed, this is probably more a measure of the more extreme randomness allowed to the “neurotic” bot than the others.
Isn’t this what you would expect from a bunch of AI guys in Austria? Back in the 1930s, didn’t Austria produce another neurotic robot that played Age of Mythology all over Europe? As I recall, he initially had a lot of victories due to his neurotic strategy but eventually lost as rational players entered the game. Now, if their little neurotic bot had ran off to a bunker on the screen, swallowed a cyanide pill and blew his brains out along side his girlfriend, I might have been more convinced that intelligence was involved in the experiment.