Two missed AI points

Not hugely important, but I’ve started this tradition and might as well maintain it.

= = = = =

Marks at MindMatters has a long podcast on the latest proposals and projects using AI. There are two notable missed points:

= = = = =

National Football League is holding a competition to develop AI to track all the helmets in a game, in order to monitor collisions. AI is blatantly unnecessary. If you have a simple accelerometer IN each helmet with a transmitter on its own frequency, you don’t need AI. Just record all the signals and graph them. When player 4 and player 17 register a collision at the same millisecond, you know what happened. This could have been done in 1957 with discrete transistors.** The correct solution is so ancient and obvious that I’m pretty sure NFL is really aiming for some secondary purpose.

= = = = =

SOME influencers have an eternal utopian faith that true human intelligence is possible with computers. Marks thinks these are “keyboard engineers” who are good at coding but not good at theory. Nope, EXACTLY BACKWARDS. Long experience with actual programming teaches you the weakness of the program and the strength of your initial assumptions.

Faith in utopia is THEORETICAL THINKING BY FUCKING DEFINITION. In any field of endeavor, real butchers and bakers and candlestickmakers know the limits of their skills and the limits of their materials. Bureaucrats and venture capitalists DON’T know the limits of the material world. VCs only know pure math, which has no limits. When a VC takes over a bakery, the VC demands a 900-mile-high cake to celebrate the infinite godlike perfection of the VC. The baker tries in vain to persuade the VC that this is physically impossible. When he goes ahead on the project and fails dismally, the VC fires the baker and starts over without learning anything. VCs and theorists and bureaucrats and grantors are delusional wackos.

EXPERIENCE SURVIVES. THEORY KILLS.

= = = = =

**The receiving part of the system could have been tubes, but the sending parts wouldn’t have worked with tubes. Partly because of size and batteries, mostly because tubes have delicate internal parts that alter the tube’s amplifying characteristics when banged. With clever design these microphonics could have been made to serve as the acceleration sensor itself, but the whole setup would be much more reliable with transistors.

%d bloggers like this: