Lift11: Kevin Slavin, Those algorithms that govern our lives [en]

[fr] Notes de la conférence Lift11 à Genève.

Live and India-lagged notes from the Lift11 Conference in Geneva. Might contain errors and personal opinions. Use the comments if you spot nasty errors.

Presentation he gave four years ago, about Manhattan. Cities learned how to listen: radar. => invented stealth. Ended badly: in 99, crash of stealth plane in Serbia.

*steph-note: bird flocks on screen are distracting… hypnotizing*

You can’t make the plane disappear, but you make the big thing look like a lot of little things. Radars can’t be tuned to see birds, or they would be useless.

What is the black box in the financial services? Black box trading. You can’t move a million shares of something without showing all your hand. So you have algorithms that break up those shares into lots of packs of little ones so that people don’t notice. But other algorithms are tracking small transactions to try and figure out how the market is going to move.

Algorithms to try and look invisible, or to try and track hidden stuff.

High-frequency trading: three things can help — better algorithms, better computer, but mainly, better network speed. So, you get closer to the internet. The internet is not that distributed: there is a space in NYC where all the pipes come up, and you can effectively have 0-sec delay. => weird real-estate specs. Being close to the carrier hotel is important!

Building and cities are structurally changing around the needs of a bunch of algorithms. Remove all furniture and put in steel floors: because of where the building is we’re going to put tons of servers in and we chose this place because it gives us a strategic advantage by being close to the internet.

Algorithms actually determine a huge amount of stuff in our lives — not just financial trading. What we see on TV, what something costs, who gets arrested, what happens next, what it looks like, how it’s made, what we eat…


  1. opacity
  2. inscrutability
  3. something darker and a little hard to describe

Opacity: elevators. Algorithm to rationalize the use of the elevators. New elevator with no buttons in it except a stop button. Floor buttons outside the elevator.

Roomba vs. other type of robot cleaner: the Roomba is unsettling, because it doesn’t clean like us. It’s algorithm is pretty alien. The other one cleans more like us.

Algorithm to design car evolution. By trial an error, the computer actually manages to figure out how to make a car — how to design it (wheels on the bottom, etc). But it doesn’t at all get there the way that we think. Unnerving. Get from A to B with available solid structures and a bunch of wheels.

Corewar. Game where programmers pit algorithms against one another. Abstract? No more than your pension in the stock market.

Frances Galton and sweet peas.

*steph-note: interesting to confront this thinking to Fooled by Randomness. We do not live in Mediocristan… right?*

Cinematch. Movie rating, rmse = 0.9525 (means: “it’s four stars, but it could be 3 or 5”).

Other rating method: takes into account the crappiness of the human brain. Really bad database (e.g. movie recommendations one month apart… rmse = 1!)

Further: what kind of movie should be made? *steph-note: scary*

Algorithms determining what movies should be made and whether they were any good. The user in this scenario is not us… the public. Long-term effect? mean regression, homeostatic monoculture.

When it goes wrong: Flash Crash. Trading P&G at a penny and Accenture at 100K.

What does a Flash Crash look like in Hollywood? On a dating site? In the criminal justice system? In the wine market?

Third: astrology. *steph-note: glps, didn’t understand that bit*

We can outdo any algorithm with the kind of willful distortion we engage in every day.

Dark pools. Huge masses of liquidity moving around outside the algorithmic pools we’ve created to trade them. So, taking things out of the algorithmic areas. What does the dark pool of Hollywood, real estate, music look like? That’s what’s really interesting, and where there lies hope.

Transparency of algorithms? Very hard to understand. What we need though is some kind of systems literacy, which games are by the way pretty good at providing.

“I wouldn’t date an algorithm but I would hang out”