Slashdot Mirror
DARPA Files Patent On Predictive Simulation
An anonymous reader writes "New Scientist has a post on a patent filed by the Defense Advanced Research Projects Agency (DARPA), seeking to control a new potent predictive simulation. The patent outlines the process, which may someday allow researchers to accurately predict the behavior of observed subjects. They're not there yet, but not for lack of trying. It already works in some military war game scenarios, says the patent. 'Parunak says his model can successfully detect players' emotions, and then predict future actions accordingly. He believes the technique could one day be applied to predict the behavior of adversaries in military combat situations, competitive business tactics, and even multiplayer computer games. The patent application gives an interesting insight into DARPA's goals. The agency has pumped a lot of money into AI in recent years without reaping major rewards. One day computers may find a way to accurately second-guess humans, but I suspect we may have to wait a little longer yet.'"
When a genuinely new way of doing something is patented, I don't see much of a problem. Even if you don't agree with software patents in principle, patents that introduce a new technology tend to expire before the technology matures enough to become profitable. In that case, the patent filer gets the honestly deserved upper hand of having better in-house knowledge about the technology by the time it gets to production stage, instead of having the unfair advantage of forced monopoly over its production.
Some patents are harmful - such as those which either patent a well known technology they didn't really invent, or patentsquatting (patenting something with the only reason of preventing others from using this technology, even if you have no intention of using it yourself either), but it doesn't seem this was one of these cases.
If the copyright system worked like the patent system (requires novelty and expires in a reasonable amount of time (~5 years)) then we'd be living in a much better world.
"have these scientists not watched a single sci-fi movie."
Apparently they havent watched history movies either.
The appropriate response to someone attempting to predict your behaviour is to feed their prediction algorithm false data (aka, feints, lying, etc). That way _you_ can predict _their_ behaviour, and not only that, you can even _control_ their behaviour.
This isnt something particularly new. This is the same old thing 'but on a computer'. I expect to momentarily see another patent application for the same old thing 'but on the web'.
The military people can already make predictions. They need others to make predictions too. They're sick and tired of propagandists and politicians who can't make even the simplest and most evident predictions.
The military people know, for example, that adventures like the invasion of Iraq only serve to fuel terrorism and make everything a hundred times worse. That's simple common sense. But since the propaganda machinery and the politicians lack all common sense, the military people want this predictive technology to become widespread, so that maybe someday common sense will prevail over craziness.
Terrorists can't threaten a country's freedom and democracy. Only lawmakers and voters can do that.
It becomes a bit like land mines: it forces you to use a less optimal route to your target than what you would have preferred. There must be a term waiting to be coined here. Idea space denial?
sigs are hazardous to your health
Someone should tell Hari Seldon his work is already done!
/ The Arrow
"How lovely you are. So lovely in my straightjacket..." - Nny
First: I doubt the alternatives have equal probabilities. In a real world setting, there are hundreds if not thousands of different options at any moment, most of which are highly unlikely. Second, you're not playing against a single person, but against a whole bunch of them, which changes behaviour significantly. Third, you have to take the possible actions of your allies (also a large number of individual agents) into account. That makes computing the probabilities of the joint actions and the joint distribution of the utility quite hard.
This model is (if it works well) an approximation of the utility function (aka objective)...
This all reminds me of Isaav Asimov's "The Machine That Won The War". :)
It's a story about a military computer that becomes unreliable as the battlefield data becomes less complete.
The operator finally decides to fiddle with the data by making decisions on a coin toss.
The soldiers still believe the machine is infallible, so follow the orders and win the war.
Ive always secretly believed that the NSA had this figured out decades ago, and has been quietly gaming the stock markets to fund its expansion. Thats why market volatility has increased dramatically. *Sure* those acres of supercomputer arrays are all doing code-breaking...
Journal of Applied Behavior Analysis
Journal of the Experimental Analysis of Behavior