Keith asked:

The meaning of life…

Can an answer to that question then be used to give artificial intelligence developers a task for a computer program to execute? Such as robots build cars, computer programs create a better driving experience, is it that human life hasn’t been simulated by a computer program because the function of life/ the goal of life/ the meaning of life/ what life is out there doing/ life’s task is unknown? Hence if known, could then AI finally be able to be designed as that life like program.

Answer by Geoffrey Klempner

It’s a good question to ask: why should anyone want anything or do anything? Writers of science fiction books and movies assume without even a moment’s thought that an android or AI would want to do stuff — that it would have a reason to obey our orders, for example, or fulfil its design function.

But that’s the whole point: human beings have been designed by evolution to pursue a range of tasks all hinging around survival and reproduction. But then we got conscious and discovered that we could give tasks to ourselves and pursue them. The result is human culture, and the history of philosophy.

In sci-fi comedies robots can get depressed (Marvin the paranoid android in Hitchhiker’s Guide to the Galaxy) but that would be the true test of success, much more appropriate than Turing’s easy option. You can’t get depressed unless life is a question for you. How does that happen? How did it happen?

There is no task. Any time you want you can get off the train of life. Provided, of course, that you understand that there will be no ‘you’ to hop on again later.


Answer by Stuart Burns

First of all, life’s task is known. The function of life, the goal of life, the meaning of life, what life is out there doing, is simply the ongoing continuation of life. So the answer to your question has to take that fact into consideration. If we haven’t built ‘life-like’ computer programs, it has to be for reasons other than not knowing the proper task to give the program.

Life is a chemical reaction. Or, more accurately, a complex interdependent network of chemical reactions. In most cases, you do not ask what the purpose of a chemical reaction is – it just is, or is not. When a lump of iron rusts, you do not ask what the purpose of the rusting is. When a lump of sugar dissolves in a glass of water, you do not ask what the purpose of the dissolving is. You thus have to be careful when asking what the purpose is of the complex network of chemical reactions that is life. It just is. On the other hand, unless those particular chemical reactions continue to take place in the proper networked sequence, they soon terminate. So the only meaningful sense in which there is a purpose to those chemical reactions, is the sense in which their purpose is to continue to take place in the proper sequence.

Applying this lesson to the realm of AI does not provide the answer you seem to think. You can setup a program to operate a robot to continue to build cars. This has been done. The difficulty is that a robot designed to build cars has no capability to withstand, or correct for, things that might interfere with its operation. Such a program for a car-building robot has no provision for reacting to a changing environment – either by modifying its behavior, or reproducing itself with modifications (ie. genetically). If you want an AI program that is ‘life-like’ then you have to provide the program with a means of adapting to the environment – either immediately like an animal, or through some kind of process of evolutionary adaptation like a plant.

As you can see, the challenge is not strictly a constraint on the AI programming. The challenge is coming up with a process that allows adaptation – either genetically or behaviorally. Whatever kind of robot the program is designed to operate, the robot has to have some means of reacting to the environment. Writing the program is not the hard part. Building a robot with the capability of reacting to the environment is the hard part. Mostly because it is expensive, and not economical for anything other than a research program. What car manufacturing company is going to want to pay for a car-building robot that can react to the environment in ways other than those useful to building cars. It is cheaper to design and build a made-for-the-robot relatively unchanging environment (model year changes aside).