Cheaper than printing it out: buy the paperback book.

Out of Control

With these lessons firmly in mind, Farmer together with five other physicists (one of them a former Chaos Cabal member) engineered a start-up company to crack every gambler's dream: Wall Street. They would use high-powered computers. They would stuff them with experimental nonlinear dynamics and other esoteric rocket-scientist tricks. They would think laterally and let the technology do as much as possible without their control. They would create a thing, an organism if you will, that would on its own gamble millions of dollars. They would make it...(drum roll, please)... predict the future. With a bit of bravado, the old gang hung out their new shingle: the Prediction Company.

The guys in the Prediction Company figure that looking ahead a few days into the financial market future is all that is needed to make big bucks. Indeed, recent research done at the Santa Fe Institute, where Farmer and colleagues hang out, makes it clear that "seeing further is not seeing better." When immersed in real world complexity, where few choices are clear cut and every decision is clouded by incomplete information, evaluating choices too far ahead becomes counterproductive. Although this conclusion seems intuitive for humans, it has not been clear why it should pertain to computers and model worlds. The human brain is easily distracted. But let's say you have unlimited computing power specifically dedicated to the task of seeing ahead. Why wouldn't deeper, farther be better?

The short answer is that tiny errors (caused by limited information) compound into grievous errors when extended very far into the future. And the cost of dealing with exponentially increasing numbers of error-tainted possibilities just isn't worth the immense trouble, even if computation is free (which it never is). Santa Fe Institute investigators, Yale economist John Geanakoplos and Minnesota professor Larry Gray, used chess-playing computer programs as the test-bed for their forecasting work. (The best computer chess programs, such as the top-ranked Deep Thought, can beat all human players except for the very best grandmasters.)

Contrary to the expectations of computer scientists, neither Deep Thought nor human grandmasters need to look very far ahead to play excellent games. This limited look-ahead is called "positive myopia." Generally grandmasters survey the chess board and forecast the pieces only one move ahead. Then they select the most plausible play or two and investigate its consequences deeper. At every move ahead the number of choices to consider explodes exponentially, yet great human players will concentrate only on a few of the most probable countermoves at each rehearsed turn. Occasionally they search far ahead when they spot familiar situations they know from experience to be valuable or dangerous. But in general, grandmasters (and now Deep Thought) work from rules of thumb. For instance: Favor moves that increase options; shy from moves that end well but require cutting off choices; work from strong positions that have many adjoining strong positions. Balance looking ahead to really paying attention to what's happening now on the whole board.

Every day we confront similar tradeoffs. We must anticipate what lies around the corner in business, politics, technology, or life. However, we never have sufficient information to make a fully informed decision. We operate in the dark. To compensate we use rules of thumb or rough guidelines. Chess rules of thumb are actually pretty good rules to live by. (Notes to my daughters: Favor moves that increase options; shy away from moves that end well but require cutting off choices; work from strong positions that have many adjoining strong positions. Balance looking ahead to really paying attention to what's happening now on the whole board.)

Common sense embodies a "positive myopia." Rather then spend years developing a company employee manual that anticipates every situation that might arise -- yet be out of date the moment it is printed -- how much better to adopt positive myopia and not look so far ahead. Devise some general guidelines for the events that seem sure to arise "on the next move" and treat extreme cases if and when they come up. To navigate through rush-hour traffic in an unfamiliar city we can either plan detailed routes through the town on a map -- thinking far ahead -- or adopt a heuristic such as "Go west until we hit the river road, then turn left." Usually, we do a bit of both. We refrain from looking too far ahead, but we do look immediately in front. We meander west, or uphill, or downtown, while using the map to evaluate the next immediate turn ahead, wherever we are. We employ limited look-ahead guided by rules of thumb.

Prediction machinery need not see like a prophet to be of use. It needs only to detect limited patterns -- almost any pattern -- out of a background camouflage of randomness and complexity.