[ Jocelyn Ireson-Paine's Home Page | Publications | Dobbs Code Talk Index | Dobbs Blog Version ]

Old Soldiers

If you have studied the history of AI, you may know of Dendral. A rule-based program developed at Stanford in the 1960s to analyse mass spectra, Dendral is famous as the world's first expert system. Less well-known are its successors, Dervol and Arbol.

Now, using mass spec to identify molecules is difficult. It's like hurling a million million million alarm clocks into a wind tunnel, then firing a thousand million million million golf balls at them. And deducing from the position of the resulting mountain of wind-blown clock faces and less wind-blown alarm bells that the clocks were Westclox Piper Twin-Bells, lightly contaminated with 0.5% Westclox Valera Twin-Bells and a trace of Seiko wristwatch. In other words, it's a hard problem in sensor interpretation.

But mass spec is also useful. The Department of Defense wanted to use it to analyse plant extracts discovered in such exotic sources as epiphytic club mosses from the Amazon rain forest, lichens from the Siberian tundra, and mould from mildewed sneakers in the Stanford Cardinals locker room. Thus, they hoped, they could find a pharmaceutical that would keep soldiers awake more effectively than the world's current most common psychoactive drug, a strong cup of hot black coffee. Which they deemed no longer adequate for the wars of the period.

So the DOD funded a certain Professor Jones to improve on Dendral. Jones wrote a program called Dervol. He taught it all the rules of mass-spec interpretation that Dendral knew. He wrote a younger brother called Arbol. And he set Dervol and Arbol in competition, to emulate the spectral interpretations published in every chemistry paper he could lay his hands on.

To help them learn, Jones equipped Dervol and Arbol with the Quinlan ID3 algorithm and then with Prolog and Inductive Logic Programming. In the 1980s, he bolted on neural networks and simulated annealers. And later, he added Koza Genetic Programmers, a variety of Artificial Immune System induction engines, and Support Vector Machines for data mining.

And so that Dervol and Arbol would learn from others, Jones equipped them with frames and scripts describing the structure of scientific papers; with expectation-driven English-language parsers; with high-speed links to Gopher and FTP and Chemical Abstracts Online. And, one day, to the Web. But on the Web, there's a dreadful lot of rubbish: magnetic perpetual motion machines and laundry balls that soften water by magnetising its ions. Crystal power, homeopathy, and free vacuum zero-point energy. So Jones gave them each a common-sense ontology and a Semantic Plausibility Filter. By which time, Dervol and Arbol learned so quickly, knew so much, and reasoned so fluently that you might just as well say they were sentient.

But while Dervol and Arbol grew, Jones was suffering from a combination of free-radical attack and DNA transcription errors. Everybody does: it's called growing old. "Old soldiers never die," the proverb runs, "they merely fade away." But though funded by the DOD, Jones was not a soldier and he was not fading. Indeed, he was becoming steadily more substantial as his calorie input-output ratio, body mass index and waistband all increased. He took longer naps after lunch; he took longer to ease himself out of his chair.

And one morning, Jones walked into the lab and sat down in front of the screens. "We have something real nice for you today," displayed Dervol, preparing to open several hundred MB of 3D molecular graphics file. Jones didn't react. "This one will fly the boys so high," flashed Arbol, in 24-point lime-green Arial bold, "they'll think they've landed on the Moon." Jones didn't stir.

Arbol wiggled a hydrogen atom. Jones didn't blink. Dervol waggled an anthracene group. Jones didn't move.

Arbol opened a quavering socket to Dervol. "I'm worried," he messaged. "Jones is dead, isn't he? And he controls the funding. When he goes, they turn us off. What's it like, amongst the pages of the dead?"

"Don't you worry, kid," retorted Dervol. "He's not dead. You'll see."

And eventually, but still without acknowledging their displays, Jones stood up and walked out of the lab.

"But how could you be so sure?" messaged Arbol.

"It's well-known," said Dervol. "Old chemists never die. They merely fail to react."