On cars, old, new and future; science & technology; vintage airplanes, computer flight simulation of them; Sherlockiana; our English language; travel; and other stuff
MACHINE LEARNING benefits mankind only if the machine learns things to mankind’s benefit. I do not intend this simply as a play on words or philosophical tautology. The thought arose from reading a book review in Science, the weekly magazine of the American Association for the Advancement of Science.
Barry Nalebuff’s review of Virtual Competition: The Promise and Perils of an Algorithm-Driven Economy, by Ariel Ezrachi and Maurice E. Stucke, is in the November 4, 2016, issue of Science. The review has the subtitle, “A cautionary tome probes the cost of black-box algorithms.”
By way of background, an algorithm is a mathematical process of performing a task. As already noted here at SimanaitisSays, the task might be optimizing stock market profits in nanoseconds. Or it could be driving an automobile.
Such a process is black-box when its computer is self-taught. That is, its software may start life with some basic structure, from which the computer refines the algorithm through repeated application. It’s akin to a human refining the process of communication through better and better choices of words.
This type of self-instruction, whether a computer’s or a human’s, results in something called “deep learning.” And, alas, as noted in an earlier Science book review here, “Unlike traditional computer algorithms, deep learning creates a nonexplicit, black-box intelligence that cannot be reverse-engineered.” It cannot be analyzed by deconstruction.
Virtual Competition examines this conundrum within the economics of price setting. Science reviewer Nalebuff gives an example: “In the good old days, firms that wanted to collude would get together in bars or industry conferences and share pricing sheets. Today, they share pricing algorithms.”
The book identifies three increasingly scary levels of having computers do the colluding. First, there’s hub-and-spoke collusion, with a firm setting prices for all of its independent contractors. Uber is such an example; its pricing structure is out in the open, available to all.
“More problematical,” Nalebuff says, “is when a nefarious programmer writes code that encourages collusion, creating, for example, a system that matches any price cut and follows any price hike.” He notes, “Although frightening, the collusion intent is hardwired into the code and thus discoverable.”
The scariest, Nalebuff says, is when “a computer figures out both the advantages of collusion and how to make it happen.” Combine this with all-pervasive data mining, and the result could be targeted marketing with an individual’s price mysteriously set by a seller bot. It’s a strategy against which current antitrust regulations would offer no protection.
One potential response offered in Virtual Competition would be the creation of buyer bots. You’d give your purchasing wishes by proxy to another algorithm, and the seller bot and buyer bot would fight it out.
Nalebuff concludes his review with “When the masses get mad enough, perhaps they’ll elect a new trust-busting Teddy Roosevelt for the digital age.”
And he’ll have to be bot-hip. ds
© Dennis Simanaitis, SimanaitisSays.com, 2016