With the new world of cryptocurrency starting to make headlines and enter mainstream knowledge, it’s important to look at the relationship between technology and finance. It’s becoming more and more important to recognize how new things like “bot” can manipulate markets and cause the tides of certain stocks and currencies to change rapidly.
Even back in 1984 it was a huge issue in terms of fear of how technology and computers would affect industries. There was a lot of fear back then, now there’s not so much fear however we are closer and closer to advanced artificial intelligence.
For an outsider, probably the most striking thing about AI is the way it violates the common notion of what a computer is. Instead of crunching numbers, an AI program uses the computer as a machine to manipulate symbols; instead of following a rigid and precisely defined algorithm, it picks its own way through a problem according to a store of data, facts, and heuristic rules of thumb about the world.
Indeed, it is arguably the most important insight of AI’s first two and a half decades that machines can behave intelligently using just two basic ingredients: search and knowledge. The paradigm is a chess program. At each step the program has to search through all the moves available to it to find a satisfactory one; but because there are some 10.sup.120 possible sequences in a chess game, the program would be paralyzed unless it had a few rules of thumb to narrow that search to manageable proportions.
Broadly speaking, AI deals with two kinds of knowledge. Factual knowledge, or “book learning,” might be represented in the computer as a network of associations: TWEETY is a BIRD is a VERTEBRATE is an ANIMAL, and so forth. Heuristic knowledge, the intuitive rules of thumb derived from experience or passed down from master to apprentice, might be encoded as a maze of logical propositions: IF this condition holds, THEN do that.
To get a fel for the scope and limitations of current AI programs, consider that a human expert–say a chess master–has at his command the equivalent of 50,000 IF-THEN statements. A modern expert system contains at most a few thousand; even the best is still an idiot savant.
INTERNIST-1, for example, knew a lot of internal medicine. It understood nothing about physiology or anatomy. Programs such as its successor CADEUCEUS, which have deeper knowledge and which can begin to reason from first principles, are still very much on the forefront of research.
By the same token, existing systems are very narrow, in part because of hardware constraints on computer memory and processing power. So far the programs have been successful only in well-defined and self-contained domains. (In fairness, of course, the same thing could be said of human experts: a lawyer may well be a klutz at auto repair.)
More important still is the fact that none of the existing expert systems can learn in any real sense. The biggest bottleneck in the creation of a new system is the laborious back and forth between the human experts and the programmer as they discover new rules and refine the old ones. Programs that can learn are again on the forefront of research.
Waldrop, M. Mitchell. “Artificial intelligence (I): into the world; AI has become a hot property in financial circles: but do the promises have anything to do with reality?” Science, vol. 223, 1984, p. 802+.
It’s very important to keep track of the trajectory of technology so as to discern its true impact on industries, especially the financial industry.
There are already AI computer programs that attempt to trade stocks by looking at patterns, however because of the “human” element these systems are still not quite perfect. Everything is predictable until it is not.