Learning from fruit-flies to solve cellular problems

Slashdot's on fire this morning — tons of great stuff there. Let's start with this story, on researchers who are solving the cobinatorial explosion you get when you try to optimize which cell-tower will transmit on what spectrum, using autonomous computing techniques modeled on the way that embryonic fruit-fly cells manage the relative density of exoskeleton to sensory bristle.

My friend Geoff Cohen turned me on to a book called "Cat's Paws and Catapaults," which examines the historical failure of mechanical designs based on nature — for example, the ornithopter. There are rare exceptions, like Velcro, but for the most part, nature's designs suggest mechanical dead-ends when applied to human engineering.

It seems that this is not the case in networking. Eric Bonabeau's Ant Colony Optimization research (which I've written about here on several occassions) has been used to solve real-life networking problems and to approach optimal solutions to the Travelling Salesman problem inherent in Southwest Airlines' routing. The Santa Fe Institute has also used cellular automata research to solve complex traffic and urban-planning problems.

The amazing thing about evolved solutions is that they're typically counter-intuitive. The Santa Fe institute will recommend that town planners reduce the number of lanes on certain roads (rather than building alternate routes) in order to reduce traffic congestion. Southwest Airlines' jets fly seemingly nonsensical routes ("Announcing the arrival of Southwest Airlines flight 432 from Denver, continuing on to Los Angeles, Philadelphia, Phoenix and Orlando"). Autonomous cellular towers will choose spectrum via a complex negotiation that will not only be non-deterministic, but also utterly unpredictable.

I think this points to a world that is not human-readable. We will be surrounded by autonomous systems that pursue optimization by zigging and zagging in ways that we can't make any sense of, at least not without serious and determined study (just as now, a compiled binary is nearly opaque to human comprehension). What a strange world that will be — our virus and anti-virus software will collaborate across networks to modify themselves and their behavior; our spamfilters will collaborate in much the same way; search-engine results based on network analysis (like Google) will grow even more magical and defy comprehension even further.

A time traveller brought here from 100 years back would be mystified, I think, by our systems and social conventions, especially those created by the ability to communicate at a distance (why make plans in advance to meet friends after work when you can just call around and see who's available once you shut down your computer, and then zip over to their location at taxi-speed?). At root, though, these interactions will be designed by humans, for humans. They are, at root, comprehensible. Our time-traveller will slowly but surely adapt. What about a world where all of our interactions, lcoally and at a distance, are governed by unknowable, evolved and adaptive mechanisms that serve us, but are not of us?

When a fruit fly is developing, its back needs some cells to develop into its exoskeleton and some to develop into sensory bristles. Too many bristles, and the skeleton isn't strong enough; too much exoskeleton and the fly is ill-equipped to sense its environment.

But there's not a central system that dictates which cells will develop into exoskeleton and which into bristles. Each cell holds the potential for both, and when a cell starts developing one way, it sends a chemical message to its neighbors. A bristle developing in one cell will tend to suppress bristle development in its neighbors, so equilibrium is established, Shackleton said.

This decentralized model, in which each cell or base station settles with its partners, works in Shackleton's tests. "It will come up with a useful plan which minimizes interference" and can better adjust to changing usage patterns, he said. Which antennas use which frequencies would no longer be BT's problem. The antennas could simply work it out among themselves.



(via /.)