Nervously. He’s going to be late, god damn it. He’s in love, and he’s going to be late. Everyone in the AI department is going crazy. Nerds and professors with frantic red splotches on their faces. What happened? A new Star Trek adaptation announced? Did the pizza delivery not arrive? Have the nerds gotten into some hilarious trouble again, positioned drones in front of a gynecology practice or steered delivery robots into the Thames?

Nothing there—

After the unbelievably cute, self-teaching neural networks had for a long time been a source of joy for developers, things had at this point gotten a little out of control. Thome has no idea what exactly has happened, as neither his intelligence nor his technical expertise are sufficient for that. AI is like a black mystery box, you specify algorithms and feed it data and never know what will come out. That sentence has always stayed with Thome. Reinforcement learning had already produced some strange errors. Like, for instance, the handling of cancer patients a few months prior. Who had been designated as cured. Now they’re all dead. Wasn’t true. Premium calculations, crime statistics, computer profiling and the resultant raids, which disproportionally affect Blacks or—let’s say—Arabs and most recently the poor. The number of accidents involving self-driving cars, the sexism of the algorithms … But that’s just anecdotal evidence on the fringes of a major restructuring. Now there’s Blackbox Watch. A group of scientists who examine the black box and readjust and—what exactly they do, apply templates of higher-output neural networks to lower-output layers, um, yeah, fuck—no clue.