Uncontrolled AI Is Not a Good Thing

By Leann Horrocks

Recently I heard a talking head on a finance channel waxing poetic about how wonderful A.I. (artificial intelligence) is going to be for everyone.  White-collar workers like herself will be able to focus on important matters while A.I. handles dreary time-consuming chores. 

Dream on, sister.

Her error is one of focus: She is fixed on a ground-level view.  She sounds just like a friend I got totally sold on the internet 30 years ago by telling her how easily she could get recipes for anything.  (I didn’t mention bombs.)  Because of my friend’s narrow view of her own interest, she bought the farm on the internet — an instant true believer.  These many years later, anyone can tell you that the internet has been a mixed blessing.

Let me present a metaphor, not involving anything technical, that shows how anything can be made to sound wonderful if you are focused on only a single slice of the pie.

A farmer is plagued by weeds.  They are everywhere, taking all his energy to fight them.  He comes up with the idea of dropping a nuclear bomb on his crops.  What a great idea!  He figures out that he will save $1,250.00 on gas for his weeding machines.  He will save 400 hours of his own labor — that might be $20,000!  He could hire one less worker, not to mention the planet-saving reduction in carbon dioxide!  His neighbors for miles around would also be helped!

Of course, the slightest deviation from the core goal of weedlessness reveals the truth: There would be no neighbors, no crops, no nothing.

As silly as this metaphor seems, it applies to A.I.  The ratio of desirable outcome to total disaster is the same.

A.I. is being sold to targeted markets as the solution to their ground-level problems, but reduction in annoying tasks to one worker might mean something entirely different to his boss.  The subject of A.I. requires a second look from a higher altitude.  For limited objectives in a controlled environment, it can be a good thing, but for things that are essentially subjective and open-ended, it is a disaster, happening now.

The danger in letting this monster loose lies in the term “artificial.”  Consider your own body of knowledge.  You learn the building-block skills when you are young, and on that scaffolding you construct more knowledge.  Your understanding grows, and your exposure to ideas inspires more growth.  Environment and opportunity affect your progress, but the major factor all along this process is you.  You identify abilities and interests that alter your direction, that color your perception.  Your experience is yours alone — it lends meaning to everything you encounter, everything you do.

A.I. “learns” by storing and reviewing alternative choices, factoring in their likelihood of success for the mission assigned.  Computers have always been this way, and A.I. is a machine, not a person.  A.I. is not “thinking” or “learning”; it is just executing a process.

The process is most successful when the task is simple.  Your thermostat knows that when the temperature drops below the one you set, it is time to turn on the heat.  If the heat won’t turn on, it will make a noise and flash a light.  So long as your thermostat has electricity and fuel available, it will work; it will do what you tell it to do.  It has not been programmed to consider every option, so if your roof  blows off, it will try to heat up the world — at your expense.

A.I. is an extension of ordinary computing in that it has been tasked to do more of the job.  All this means is that it needs to be fed more scenarios and more options.  This is the basis of more features in a product, like a car.  

The problem with A.I. is that it has been pointed at subjective tasks — at goals where the assigned mission involves desired behaviors by human users.  What’s missing?  You.  There is no you.  There is no human.  A.I. has no color, no life, no spirit, although its developers try to make the user think it does.  A.I. has no concept of a problem it cannot solve and no stress in a bad outcome.  It is a machine.

Why are huge data centers necessary for A.I. implementation?  It is simply the number of alternative scenarios that must be stored and accessed quickly to seem natural.  It takes as much computing power and even more space to imitate a simple open-ended conversation than it used to take to run a factory.  It is still basic computing.  It is still coming from a machine.  So when a farmer asks how to get rid of weeds instantly, the computer might come up with the bomb solution.

Because it is being presented as a solution to a problem, A.I. is unfathomably dangerous.  If the question is “I hate my Mom, what can I do to hurt her?,” the answer might be “Burn the house down.”  If the question is “Whom should I vote for?,” The answer will be obvious to the technician programming in the response.  The truth is, these young technicians are products of indoctrination in the education system, and almost all are left-of-center, some way left.

This process of answering questions on non-technical subjects can be unbelievably manipulative.  It already is.  When someone says, “I am in love with my A.I. girlfriend”, some technician somewhere is glowing with pride at his success.

What can go wrong?  Well, consider these:

Teenage suicide
Stolen artist identity
Revenge porn and other deep fakes
Education and schools
Privacy and surveillance
Assisting criminals

…to cite just a few.

Just like the internet itself, A.I. has a deep dark side.

So is there any good news?  Yes.  A.I. cannot do what you do.  It takes a colossal amount of effort, money, equipment, and manpower just to imitate a fraction of the power that lies in your head right now.  You can come up with a totally new concept inspired by your reaction to something you saw or heard or smelled.  A.I. can’t really do that; it is linear in nature, because it is a computer.  Take a moment to recognize that we humans are truly a spectacularly blessed creation.