An old man's grumblings

Warren Buffet once stated that he didn't invest in any business he didn't understand. I think that was in 1999 and he made that comment when a business journalist asked why he wasn't stuffing his portfolio with dot-com stocks.

I don't understand crypto-currency. From Eaton Rapids, Michigan it looks like somebody auctioning off "Prime Numbers" or selling stars in far-away galaxies.

I understand the value of a half-dozen, freshly caught bluegills, five-pounds of potatoes and a couple of freshly pulled onions. To me, those have tangible value although to many people they have less value than .jpgs of those same items. The difference is that I know how to turn those items into food while many people would find them repulsive and be bothered by the effort required to throw them into the rubbish.

Maybe that plays into the allure of crypto: The figurative has become more valuable than the literal due to ease of manipulation.

Artificial Intelligence

I must confess to having similar positions on Artificial Intelligence.

For one thing, even infinite knowledge about a finite event or item does NOT have infinite value. For example, even if I know to within one millionth of a millimeter the location of an eight-point buck...the maximum value of that information is the worth of that eight-point buck.

Is one-millionth of a millimeter meaningful when we are talking about an animal? Is a millimeter meaningful? Are we talking about the center-of-gravity of the heart? Maybe the brain? Perhaps the center-of-gravity of the animal? Does that include the content of its intestines (the deer shit)?

Another aspect of AI that isn't discussed are the fragilities that reliance of AI is baking into the system.

Dependence upon sensors

I worked in an truck plant (now closed) that struggled to "make rate". That is, it was unable to make the required number of truck cabs needed to make enough trucks to make the plant economically viable.

Every issue was addressed by adding more sensors and more "error proofing". The body shop ran slower and slower due to electronic "hardening of the arteries".

The chunk of the body shop I was given responsibility was from Zone-A to Zone-R. It didn't take long to figure out that the same issues were biting me many times a day.

For example, there was a magnetic position sensor in every station to verify that the skid had entered or left the station. That was to prevent the front of one skid from crashing into the skid in front of it.

The expulsion from the weld guns sprayed on top of one of the sensors* and in a few hours the welder had deposited enough magnetic weld dust on it that the sensor indicated that the skid had stalled on top of it and the line shut down.

The electrician had to hear the call sing-song, read the marque, walk over to the cell where the fault originated, lock-out the cell, wipe the dust off the top of the sensor, remove his locks, clear the faults and restart the cell. Guaranteed fifteen minutes of downtime every time it happened.

They had run like that for YEARS.

"Hey, Ron...when you getta minute can you meet me at cell K-40?" I called him on the radio.

"Sure, Joe. What's up?" Ron Muzurek asked (no harm in using his name. Great electrician, by the way).

"Easier to point it out than to try to tell ya" I responded.

 The cell was running when he ambled up.

"Do you see how the robot on the left side that is welding the rear roof-bow is spraying weld sparks on the sensor?" I asked.

"Yeah" Ron said. It was as clear as day.

"Do you suppose you could tip the orientation of the weld gun just a little bit so the spray went somewhere else?" I asked.

"I could try, but I gotta have your permission to do that because I will have to shut the line down" Ron replied.

"You got it. Just be a little bit smart about it so we don't strip the system dry" I told him.

Two hours later, Ron waited until K-zone was about to be blocked out due to down-time in the system it was feeding (L-zone) and stopped the cell. It took him five minutes to program the robot so it tilted the gun and shot the sparks inside of the cab instead of backwards-and-down.

That fault, in that station disappeared forever.

So...run that process in reverse. 

If you are running your enterprise on AI then you are totally dependent on sensors somewhere. Many sensors. Sensors struggle in dirty environments.

If a problem can be fixed in five minutes it can also be created in five minutes, either through malice or "noise" in the inputs.

*Some of the newer plants had the sensors mounted vertically and they looked at the side of the skid. The old plants had them mounted horizontally and they looked at the bottom of the skid and were vulnerable to weld slag accumulation.

One of the newer plants did have horizontally mounted sensors and they were vulnerable to "un-making" when a warped skid came into station and rocked when it stopped. The controls guys fixed it (five years after the plant went operational) by latching when the sensor made or only looking at the sensor's state at the back of the stack of inputs to give the skid time to stop rocking.

They also changed the decel from a step function to a ramp which helped with rocking and overshoot.

Little fiddly stuff that grows exponentially as sensors are added to a system.

Random fact

NVIDIA is 7% of the market-cap of the S&P 500 and it has a Price/Earnings ratio of 53. That high P/E indicates that the market believes that NVIDIA's profits will continue to grow exponentially.

To me, stock in NVIDIA looks like the last ticket on the Titanic.