Algorithmic reporting
Almost every story I write has some thread in it that the editor decides needs to be pulled out. In my recent piece looking at Boston’s Next tech goldmine, I decided to act on something I’d read just prior to starting work on the piece, an article about Nick Diakopolous’s thoughts on interviewing the algorithm. The story was about companies built around algorithms, so why not? But it didn’t make it into the piece, which was focused more on why Boston has lost much of its high-tech luster, and whether it can get some of that back.
You can decide whether that was a mistake. Here’s what got cut:
Watching algorithms get developed must be something like watching Thomas Pynchon write novels, sans the sex and paranoia. Someone stares at a screen, pecking at a keyboard, willing shape out of the bits that will respond to the world around it, a being without being. But listening to an algorithm being discussed is not like book group.
“Your equation assumes that the number of actions per flight follows a binomial distribution,” says one of DataXu’s computer science PhDs during a meeting to improve part of the company’s new product, Cruise Control. Poisson distributions and non-parametric equations, statistical methods used to determine probabilities, are routinely invoked.
But then, DataXu’s technology was designed to help NASA plan a Mars mission. Baker has a sly memento of this in his office, a photo of the founding team sitting at the Apollo capsule at the Museum of Science.
Fiksu’s algorithms weren’t developed for such lofty reasons, says Micah (pronounced Mee-kah) Adler, its founder and CEO. Fiksu is Finnish for “clever,” and Adler is both, though he was born in Australia and graduated from high school in Concord, New Hampshire. After getting his BS at MIT and a PhD at Berkeley, he started teaching computer science, specializing in algorithm development, at U-Mass Amherst, publishing papers with titles like “Towards Asymptotic Optimality in Probabilistic Packet Marking.”
Adler, 45, still has a bit of the professor in him, despite being on his fifth startup. He wears sensible brown shoes, slacks and a collared shirt, with Hugo Boss eyewear and a Pebble Smartwatch. He’ll take a moment to boil an algorithm down to its essence, the process of getting an answer to a mathematical equation. “Two plus two equals four is not an algorithm,” says Adler. “Taking two, adding two and saying it will equal four is an algorithm.” Algorithms, then, are meta math.
He says that at peak usage times, Fiksu’s algorithms are choosing between 150,000 potential ad placements a second. If its algorithms lag, somebody in Europe might not find the app they’re looking for.
Fiksu’s own algorithms draw 15 people, nearly 5 percent of its workforce, to a regular Thursday afternoon meeting. The group are mostly in their 20s, though 30 and 40-year-olds (including one physicist) are also in the room. The discussion topic of the day involves optimizing what sites Fiksu will use for ads. The leader gets up and starts writing mathematical formulas on the board, only it’s not a class; each coefficient relates to an actual part of Fiksu’s ad sales process. He uses so much of the board he’s eventually on his knees squeezing in a last bit of formula. He needs still more space, so he finally just says, “After a whole lot of math….”