The recent foul-up of exam-grading computer systems is no surprise. Catastrophe-watchers in software will say: "You ain't seen nothing yet."A normal brain suffers system malfunctions with no serious ill-effects. We periodically get the wrong end of the stick, or mis-hear our partners. They don't call for a brain scan. They just glide along with "Were you just thinking?"
Now neuroscientists are beginning to map the brain's operating system, we can spot the saving trick. The brain has a two-layered structure. The top layer, associated with consciousness and self-report, updates its own rough model of the myriad routines below the threshold of awareness.
In neuroslang, the upper layer is the "person", or "you yourself". The unseen "zombie" is a lower layer, largely unconscious, teeming with specialised agents dedicated to the neural circuits regulating the body's internal functions and external reactions.
When I drive to the station to meet a friend, we may have an animated chat. My brain's "person" conducts the conversation. The zombie drives. I may not even recall our return route unless a rare interrupt occurred in which the two layers briefly communicated.
"Look out for that cyclist!" my passenger may have cried, causing my zombie to swerve, and imprinting the landmark in my memory. No operating system has such a top layer. But even with only a sketchy self-model, such a layer could be extraordinarily useful. Intelligible self-report (not to be confused with error messages) could render harmless most runtime bugs. With a top user-interactive layer, diagnosis becomes a snap.
Highly illustrative of the separation of layers inhabited by person and zombie is a patient called Bill, studied by VS Ramachandran, director of the Center for the Brain and Cognition, University of California, and author of Phantoms In The Brain. A small, localised stroke has taken out Bill's left angular gyrus, associated with calculation. Here is a conversation.
"Okay, Bill, can you subtract seven from 100? What's 100 minus seven?"
"Oh, 100 minus seven?"
"Yeah."
"Hmmm, 100 minus seven."
"Yes, 100 minus seven."
"Ninety-six?"
"No."
"Oh."
"Let's try something else. What's 17 minus three?"
"Seventeen minus three? You know, I am not very good at this kind of thing."
"Bill, is the answer going to be a smaller number or a bigger number?"
"Oh, a smaller number," he said, showing that he knew what subtraction is.
"So what's 17 minus three?"
"Is it 12?" he said at last.
Further questioning confirmed that Bill has a sophisticated conscious model of the number system, including infinity, approximation and relative magnitudes. But the calculation agent had dropped out of the library, depriving the library manager (Bill) of this specific function.
One question ("Is the answer going to be a smaller number or a bigger number?") was enough to reveal that the bug was nothing to do with Bill's knowledge of what subtraction is. It was confined to the operation. With computers, we still address these questions to some outside guru. This is from the June 2000 What Palmtop And Handheld PC.
Ailment: "But whenever I plug the Palm into the keyboard it attempts to HotSync via the direct serial connection.... What am I doing wrong?"
Treatment: "The most logical solution is that your Palm V is not aware that the keyboard is present. You will need to reinstall the drivers."
Now imagine a woman taking her husband to the doctor. Ailment: "But when I put his crossword before him he tries to activate the CD player."
Treatment: "The most logical solution is that your husband is unaware the crossword isthere. You need to replace his spectacles."
In practice, husbands are usually fairly well trained in spectacle management in the first place, and even in answering simple questions for themselves.
What is stopping the importation into software of such question-answering capabilities - well within the state of today's software research and development. Is it just the fear that transparency and simplification might spoil the market for the trouble-shooters? Or have I missed something?
