Watson's advantage is clear, significant, and game-deciding. If you were familiar with the two human contestants you would know this. Barring an unlikely malfunction, Watson can't possibly incur the delay penalty from buzzing in early. Watson has no real-time audio or visual input, so it is certainly not anticipating the "clue finished" signal.
The instant before the "clue finished" signal is sent, Watson already knows whether or not it will attempt to buzz in. When the signal is sent, you've got a few dozen nanoseconds of wire propagation delay, plus the propagation delay Watson's "clue finished" signal-to-solenoid logic incurs, plus the delay of the solenoid itself. The final two steps almost certainly take less than 10 ms each.
Even with a very forgiving estimate of a 50 ms "clue finished" signal-to-buzz latency, it would be extremely unlikely for even the most anticipatory human to beat that. Granted, it does happen. In day 2 of the IBM challenge I counted 3 times when a confident and correct Watson was beaten to the buzzer by a human contestant.
The instant before the "clue finished" signal is sent, Watson already knows whether or not it will attempt to buzz in. When the signal is sent, you've got a few dozen nanoseconds of wire propagation delay, plus the propagation delay Watson's "clue finished" signal-to-solenoid logic incurs, plus the delay of the solenoid itself. The final two steps almost certainly take less than 10 ms each.
Even with a very forgiving estimate of a 50 ms "clue finished" signal-to-buzz latency, it would be extremely unlikely for even the most anticipatory human to beat that. Granted, it does happen. In day 2 of the IBM challenge I counted 3 times when a confident and correct Watson was beaten to the buzzer by a human contestant.