Monday, March 5, 2012

Science Has Failed Us Before

Science Has Failed Us Before

I remember when I was a kid there was a ton of news about building bio-domes on mars and shit. What the fuck ever happened to that? Or really, the lack of anything useful on Mars is a bit of a letdown. Maybe it has to do with the Soviets collapsed, thus we didn't feel the need to impress anyone anymore.

I used to watch a ton of those "future technology" shows during the late 90's. I guess they were popular because the millennium was coming up and everyone wanted to know about the future and the whole concern about Y2K was all around.

The problem with all this is that I don't remember a single fucking technology that they showed that actually went anywhere.

And you know what, recently I've been thinking a lot about times when tons of scientists, extrapolating current trends in a given discipline, made all sorts of exaggerated promises about what "the next 10 years" would bring. Obviously individual researchers or labs do this all the time but there have been a couple of moments where disciplines as a whole have done this for example.

1) AI research in the second half of the 20th century. The modern field of artificial intelligence is usually said to have begun at an extended conference at Dartmouth College in the summer of 1956. It was an exuberant field from the get go - researchers were incredibly optimistic about the speed with which breakthroughs would occur, particularly in the period between 1960 and 1975. It was generally assumed that a representationalist approach to AI programming would easily produce intelligent, flexible and general purpose algorithms that could assist humans in various tasks (debates as to the "strongness" or "creative potential" of AI were something of a sideshow compared to the overwhelming optimism of the field).

Unfortunately, the representationalist paradigm fell flat on its face. While computers proved adept at specific tasks, such as playing chess, they proved hopelessly inept at assimilating wide-ranging data into a general model. Moreover, they were horrible at understanding information that was not easily representable in a machine-intelligible format. While AI as a discipline has matured enormously since the heady days of Shakey the Robot and DARPA's attempts to build autonomous tanks in the 1960s, its history is still best described by an (apocryphal) 1995 quote from Marvin Minsky, one of the co-founders of MIT's AI lab:

"I assigned a graduate student the task of getting a machine to recognize simple shapes from visual input from, say, a video camera. I figured it would take him a few months -- a good summer project. That was in 1966. We're still looking for good methods."

So yeah... Science.. thanks for fucking failing us before. Hopefully you won't make these same mistakes in the future.

No comments: