Complex, real-world tasks

I predict slow progress for computers

I am impressed by the chatbots and their relatives. I think that most of the people who write essays dismissing their potential are wrong. I believe that they can be creative. I believe that the issues with hallucination will be overcome. But I do not believe that we are only a few years away from machines being able to outperform humans in everything.

This morning, two of my grandchildren and I watched a man in a cherry-picker cut down a tree. This was a complex, real-world task. He had to maneuver himself into position and manipulate his chainsaw so that branches fell where he intended. He had to avoid having them fall on his own head or on cars parked nearby. Once the branches and the trunk were on the ground, they had to be trimmed and loaded into a dump truck.

I am sure that engineers, with some effort, could program a machine that could take the human out of the tree-cutting process. But my guess is that this project would take many months, even years. Some of the equipment required would have to be invented, tested under a variety of conditions, and re-worked, as errors are discovered and corrected. Now multiply that by every complex, real-world task that humans do.

We can program machines do to complex tasks in the physical world. Military drones and self-driving cars demonstrate that. But each complex task requires specific hardware and software, and lots and lots of training.

In modern society, there are many complex tasks to be performed. When these involve manipulating three-dimensional objects in the physical world, they pose a challenge for computers, no matter how skilled those computers are at manipulating information in the realm of bits. There are many computers that can outperform the average chess-playing human when it comes to strategy. But how many of them can pick up a piece on the board and move it?

Other complex tasks involve the social world, and problems persist because there are trade-offs involved in trying to solve them. For example, suppose that more surveillance would improve out ability to prevent crime but at a cost of reducing privacy and raising the specter of totalitarian control. What should we do? Putting computers to work on social problems is not going to eliminate complexity or get rid of trade-offs.

As humans, we are very proud of our mental abilities. And if you feel threatened that a purely mental task that you take pride in can be done by a computer, I can understand. But in terms of mental abilities for doing physical tasks, or the ability to solve social problems, the computers have a long way to go. In the world of bits and information, they are arguably ahead of us. But in the world of atoms and manipulation, and in the world of interpersonal interdependence, they are still far behind.