Prairie Trail Logo

Views from the Prairie

March 17

The Stupidity of Artificial Intelligence

From the stock market "flash crash" to "bots" running up huge text messaging fees to self-driving cars crashing into trees they can't see; Artificial Intelligence is showing just how stupid it can be. Much science fiction has been written about our fears of the super intelligent computer. But the real danger is giving too much power to these very stupid devices. There are two very important skills that computers lack; the ability to take a lesson learned in one field and use it in another, and the ability to see when a proposed answer is not just wrong, but way off base. Several AI machines interacting with each other can drive each other into absurd situations.

Many people fear Artificial Intelligence. There is talk about the "singularity" where the artificial machines are "smarter" than humans and can go running off into the future leapfrogging over anything we can do. In TV shows, computers know everything and the holographic doctor is able to cure any problem.

The reality is far less threatening in that sense.

When humans master a skill that usually indicates a general intelligence level but not when computers do. Computers lack important basic capabilities that humans have.

Humans are able to take a skill learned in doing one task and use that skill in many other tasks. We learn both sequences and risks in chess and use those skills in business and war. We learn analytic reasoning in math class and use that skill in business. We learn to go behind the words in English class and then use that to see the hollowness of politician's promises, claims, and speeches.

Computers are not able to do that.

The other human skill that computers lack is the ability to see that a proposed answer is totally wrong. The computer program Watson that won Jeopardy won only because the wrong answers didn't have that high of a cost. It was able to come up with enough right answers to win. But when it had a wrong answer, those wrong answers were way off base. It would be like walking into the emergency room with the holographic doctor complaining about a stomach ache and being told that you needed a cast on your arm. Even a kid could see that the answer was wrong.

When several AI machines are interacting with each other, they can quickly go into absurd places. The stock market "Flash Crash" is one example where for a few seconds, prices were crazily low. Other cases have been where computers set prices and where two systems ran up a $50,000 texting bill sending messages to each other.

These artificial intelligence systems need humans to review the proposed results to prevent the wrong answers from causing problems. Automatic pilots need humans to take over in situations beyond the programming. Automatic driving systems need humans to take over. Automation in factories need humans to handle problems.

Run a Tight Ship

Every business wants to run without wasted effort. There are two main ways that can be done. They are very opposite in style and results and in the amount of effort needed to accomplish them. One is control, the other is through persuasion.

The first style that people think of when they want to "tighten up" is to be very controlling. In history, Henry Ford ran the company that way. All departments reported to him, there wasn't much management between him and a worker, and after the profit margin dropped, the only way that he could keep that kind of control was through using violence. One manager, Charles Sorensen, was known for his insensitivity to others and an explosive temper. Harry Bennett was a thug who ran Ford's security office and when fired, drew a pistol on the person giving him the message.

The other style is to work to convert everyone to working on the same page and all believing the same message. Toyota was famous for doing that where even the janitor might know why certain parts of the factory had been designed that way.

These two methods fit different environments. Being very controlling works when both the goals and how to reach those goals are well known. Thus, brilliant engineers can fall into using that method. However, when you have the goals well defined but not how to reach them, the second method out performs the first. By convincing people of the goals, we enlist them in actively searching for how to reach them. The combined brain power of many people has outperformed heroic brilliance many times.

Choose wisely on how to tighten the ship.

Risky World

The recent failures of both Amazon AWS and Microsoft's Azure point out the risks of these services. Even though the systems are designed to have fantastic uptime, they can't be designed to avoid human errors such as the one that brought Amazon down. It is best to have redundancy between providers.


This newsletter is posted here as well as sent via mail and email. If you wish to receive updates, please sign up above.

Recent Issues

  1. Jan17.html
  2. Feb17.html