"computer" refers even to an analog feedback amplifier arrangement.
(Not sure why that ended up posted as a followup to the article but I'll repost it here where it belongs.)
Adjustment of a mechanical yaw contorl is actually simple: Design it to furl too soon, then add weight to the tail to bring the furling wind speed up to where you want it. It's not hard to do. But it does require waiting for a wind high enough to make the mill furl - which requires patience to avoid overcompensating into extremely late furling. The risk of destruction due to late furling occurs early in the mill's life, before this adjustment is made. Once it's done, it's difficult to get anything to break.
Part of the drill for designing a mill is to keep it inexpensive enough that it makes sense to bother. Yaw control is fundamental to the operation. A passive mechanical system involves two bearings, a tail, stops to keep the tail out of the blades, and positioning the axes of the three shafts (turbine, yaw, furling) properly when constructing the head. Adjustment involves adding weight to (or removing weight from) the tail assembly or (if they weren't in the right ballpark in the first place) changing the length of the tail boom or the size of the tail. Once set up properly points of failure are few.
Even a non-redundant active system involves wind direction and speed detectors (with bearings and electronics that can fail), a computer, a servo motor, a yaw position sensor, interconnecting cables, and a power supply. They must run 24/7/365 (even when the mill is shut down for maintainence - at least until the rotor is tied down) for the life of the mill. At least some of these parts are located on a tower well above surrounding objects, making it a likely target for lightning. (Putting part of the electronics elsewhere increases the vulnerability.)
That adds up to a lot more points of failure - many of them far more fragile than bearings. So it implies a much higher failure rate to go with the higher cost. It also requires a lot more things be gotten RIGHT to make it work at all - and a broader range of engineering skills to design it.
But redundancy as a solution? Now you've increased the costs and the engineering needs a second time. Are you going to mount multiple wind sensors? Have multiple computers? Multiple yaw motors? How do you orchestrate failure detection, so an insane machine doesn't take over the show? Do you have separate batteries for the computers to keep them alive during calm periods (or when the house batteries fail)? Solar panels to keep those batteries charged?
As for simplicity and abacaii: "A system should be as simple as necessary but no simpler."
I've done software for the auto industry - where nearly any program failure can be life critical: Your car engine control goofs idle speed control and you stall a few feet after going through a stop sign. Airbag tester screws up and it fires an airbag with a worker standing over it. Energy management system screws up and the lights go out in the plant with the workers surrounded by moving machinery. (To name hazards of just three projects I've worked on.) One of my former colleagues once said I'm the only guy he'd trust to program his pacemaker. B-)
For a homebrew mill I would NEVER pick an active yaw control. As someone who's engineered active systems I'm all too aware of how easy it is to get them wrong. Meanwhile, a well-debugged and reliable passive design is easily within the reach of even a newbie garage mechanic. (For a megawatt commercial mill, with a team of engineers, a major construction, modeling, and testing budget, a periodic maintainence schedule religiously adhered to, and failure alarms bringing out the on-call emergency maintainence team, it might be a good idea.)