digital life

This century started as a time when life very easy to lose; it is finishing as a time when life seems very easy to create. Whether by genetic manipulation of biological things or the construction of computer viruses, we can design and make things which have never existed before, and which might spread themselves throughout the world without our deliberate help. The two fields are intimately related and not just because computers are central to both. They are both founded on a belief that there is nothing special about the way in which life arises. The two central tenets of this faith are that spirits are just very complicated bodies and that the only way to get bodies sufficiently complicated to have spirits, or life is by evolution.

Last week thirty or so of the smartest people interested in these things gathered in a Cambridge College to puzzle over where life might be going in computers. There were Douglas Adams and Richard Dawkins; Chris Langton, the man who coined the term "artificial life" to describe what might happen inside computers; and Tom Ray, who wrote Tierra, the first program inside which parasites spontaneously evolved.

Circling with an urgent whine around the floor in coffee breaks there was Grey Walter's tortoise, the first apparently intelligent robot ever built, along with two modern copies.  The original had been built in Britain by a pioneering neuroscientist, in 1951, to show how a very simple feedback loop might lead to complicated behaviour. Its parts had been rescued from obscurity by Owen Holland, an engineering professor at the University of the West of England: one of the original clear plastic shells still has holes drilled in the top where it had been used as a cloche to grow vegetables.

One might argue that recycling a robot into a device to grow vegetables is a really memorable example of human ingenuity. But in the context of digital life, it was something like blasphemy. The puzzle for most of these people was not when life might begin to spread through computers, but why it was taking so long to do so.  Tom Ray's Tierra, a digital simulation of evolution, had produced its first parasites in 1991. But nothing more complex had evolved since then, despite millions of millions of runs of the program.

It may be, he said, that this is how life works, on computers as on earth: most of the history of life on earth is the history of single-celled creatures. It took billions of years to evolve multi-cellular life — the only sort you can see without a microscope — which has taken only about 650 million years to get out of the sea and cover the planet. But the only way to do this is to find out. So he has modified Tierra so that it simulates multi-celled creatures, and these programs are moving around a network of 200 computers around the world. The new creatures, like the old, have parts that represent genes strung out along a chromosome, and already some forms of gene duplication that are seen in the natural world have appeared in his.

Th point of this is not to produce, ultimately, intelligences that would be like human beings. "It's easy to get human intelligence the fun and natural way" He said. "Why should make more with machines. We should have things that complement us rather than imitate us. We should let it develop its intelligence in a completely alien way."

Of course, there is no guarantee that this will happen. But then, Ray said, there is no way to find out but to try it. He has a striking analogy to prove this point: Imagine that we are robots, all of us, and our brains are built out of silicon chips. We don't have any experience with carbon-based life-forms. Some robot scientist comes in with a flask of a ammonia, hydrogen, methane, and all the other gases in the atmosphere of the primitive earth, and says "Do you think we could build a computer out of this stuff?". The only way to find out would be to build a world and let it run, until finally humans appear. But that might take billions of years.

In the meantime, there is a safe general  rule that anything which looks alive on a computer is done with smoke and mirror and relies on human expectations that whatever looks alive and intelligent actually is.  There is a huge market for simulations of living, cute or cuddly things on personal computers. A million and a half copies of one program representing digital pets has been sold.  Half a million Japanese have a kind of furry penguin called Finfin living on their screens. The conference sponsored by Cyberlife, a company which makes possible the most advanced of these creatures: the player, or owner (or perhaps their personal God), has to teach them to talk and to explore the world after they emerge from the incubator. They need affection as well as food; and their inner states can be read off on screen. They can fall ill, and die or be restored to health. These "norns" are so lifelike that a couple of web sites appeared  on which they were being tortured. The man responsible received death threats. One wonders whether a silicon-based life-form would regard that as evidence of intelligence.

Fixing digital life into the outside world is even harder. Owen Holland, the man who restored Grey Walter's tortoises, is planning a robot that would be self-sufficient. Instead of recharging its batteries at the mains, it would run on slugs, caught in the wild and fermented to make methane. As invertebrate agricultural pests, slugs have no defenders. A sort of fishing rod tipped with tiny cameras which can recognise slugs by their distinctive signature in the infra-red is easy to build. So is a robot body that can trundle slowly round fields newly sown with wheat. Combine them, and teach the resulting beast to hoik up slugs with a grabber like the ones used to scrape up teddy bears in amusement arcades, and you have the beginnings of a machine that could make its own way in the wild. But there is a huge amoutn of engineering to get through before it is actually built.

The differences between artificial intelligence and artificial life are subtle. Lots of people don't see them at all. Artificial life complex enough will acquire intelligence; artificial intelligence bright enough will surely be alive. But they are clear enough in the beginning and fumbling stages of the enterprise, when artificial life is modelling aspects of biological life, and artificial intelligence is modelling aspects of the real thing, whatever that may be.  

Tom Ray's analogy of the robot scientist with his flask of ammoniated goo shows how very much larger the Internet is going to have to get before it can support things that seem robustly and interestingly alive. It is possible to produce the building blocks of life in a a laboratory flask by passing simulated lightning through it. But if you want intelligent life to evolve from these scraps of amino acids, you need a whole world. Similarly, if we want an independent silicon-based life form, it will need a lot of cyberspace to play in: hundreds of thousands as much computational power as there is the in the world at the moment.

Bruce Damer, the conference organiser, compared the present state of cyberspace to the places where life may first have evolved, on the edges of volcanic vents under the sea, in small colonies very tenously connected with each other. Since then, however, life has transformed the planet, filling the atmosphere with oxygen — which was a deadly poison to the first bacteria to evolve.

I am not absolutely certain that I want similar things to happen in cyberspace. If genuinely autonomous life evolves there, we might find it changes its environment as drastically as carbon-based life has changed the earth. Since human culture and most of human life will be then be utterly dependent on computer networks, this is not a wholly enticing prospect. But it may not be one that humans have much choice over.

 
Front Cuts Book Back