It’s called FatKat, and apparently uses AI to make stock buy and sell decisions, according to the NY Times. The article goes on to discuss hedge funds that use AI and genetic algorithms to find arbitrage opportunities that less sophisticated software might miss. An emerging problem is the ‘black box’ issue: machines evaluate completely random rules that humans would never consider and combine the most successful ones. The evolved machine concepts can be completely outside human understanding or intuition.
As a result, the decisions that come out are such that a human has no idea if the black box is broken or operating at peak efficiency. Says MIT’s Andrew Lo “As with any black box, if you don’t know why it works, you won’t realize when it’s stopped working. Even a broken watch is right twice a day.†Scary…
{ 1 trackback }
{ 3 comments… read them below or add one }
Kurzweil would seem to know something, but has he filed for patent, or is this all ‘gosh and by golly’ trade secrets?
The methods of AI need some big failures or hacks to really make the managers go back and parse the record, assuming there are lines of code that traces all branches for a particular incident. How much are they relied on (market penetration, inc’l. outside stocks)? What’s the record under stress? ‘Still can’t read a book as well as a 4-year old’?
And then there is public opinion, if informed enough . . .
A couple of my pension funds are probably being traded by computer, but it’s unlikely thy’re using anything but well understood algorithms (’buy low, sell high’) and big trades undergo human scrutiny… I think. I should look into it.
Maybe those disclosures they are forever sending me should include a discussion of how they use technology…
. . . seems an area that may not yet have been exploited . . .