6: The Crystal Gateway Marriott

AI in the Battlespace: Chapter 6

6: The Crystal Gateway Marriott

Liu Kee sits in his civvies in the bar of the Crystal Gateway Marriott. He has an evening of dead time ahead. The one-day conference was a solid success and he has meetings in the Pentagon tomorrow before heading back to his office at Stanford.

The bar is lined with discreet alcoves - the hotel's clientele craves private encounters. The lighting is subdued and in a distant corner a pianist plays smooth, tinkling jazz. Kee nurses a scotch, rattling the ice, wonders whether he should bury his face in his phone. Instead he scans the room ... and notices a woman making a bee-line towards him.

It’s Yuan Lin, a girl he knew in school in California. Hell, he dated her a few times. What did she end up doing? Journalism, he thinks.

She’s wearing a bright red dress - her lucky dress, he quips to himself - and a big smile. She’s a bit wobbly in those heels and Kee wonders whether she’s already had a few drinks or whether she’s just not used to them. He’s a little bit on his guard because he’s read too many thrillers where the military hero, bored and alone in a bar, is approached by a femme fatale journalist interested only in his secrets.

It turns out that Yuan Lin is a reporter on the Chinese Daily News in LA. Why is she in Washington? Covering this and that in town. And what is he doing here on the East Coast, something super-secret for the DoD?

They sit together as he tells her a story, carefully staying in the public domain. She sits close and any observer would think they were a couple enjoying their stopover in Crystal City.

“Don't want to bore you, Lin,” Kee begins, “Please tell me if I'm getting tedious but there's something I'm thinking about at the moment which is absolutely fascinating. It goes back to the 1970s and a guy called Donald Michie. He was a Scottish AI researcher with a background in computer chess but he was more interested in the human experience of dealing with very smart AIs.”

Yuan Lin is very attentive.

“Computers at the time were powerful enough to play perfect endgames in a variety of board positions. The plays in their endgame tables could be more than twenty moves long.”

“Oh,” says Lin, whose preferred game is Go, “I’m sure they can do better now.”

“Sure they can,” Kee says, “But the important thing here is how the computer plays. While you, as a human, are desperately worrying about your next move, considering all your strategic options, the computer is simply looking up the board-configuration in its database, which was compiled previously using exhaustive search. The database is just a list of pairs: board-then-move. Given the board position, the program just makes the corresponding optimal move it finds there.”

She nods: makes sense.

“But usually the computer's move makes no logical sense to the human player. It kind of looks like the computer is playing at random, making irrelevant moves disconnected from any strategy. You get these weird machine-moves and the person ploughs on, only to find that for no apparent reason the noose is tightening around their neck. Slowly, inevitably and incomprehensibly, they lose.”

Lin looks fascinated.

“That's a great idea. It's like the Terminator. An implacable opponent.”

“Michie said that it was like playing against an alien. Interestingly, that’s just what Demis Hassabis of Google’s Deep Mind said about AlphaGo when they analysed the concepts in its neural-net. He said it was like the mind of an alien.”

Kee takes a sip of his whisky: “Donald Michie coined the phrase ‘The Human Window’ for AI behaviour which people could actually understand.”

Lin has a thought: “People say that deep learning systems are incomprehensible to people. Does that mean they’re operating outside the Human Window?”

“Exactly,” says Kee, “That’s exactly what they’re doing. And it’s a real problem. Except in one line of work - where it’s an advantage.”

---

Next     Previous.