Life is a system that evolves by natural selection. All living beings pursue the root goal of species survival.
Traffic barriers in a basket, Roppongi - June 2013

Traffic barriers in a basket, Roppongi - June 2013 (Enlarge)

Considered at the most basic level, life is a kind of energy flow. A living organism consists of edges that channel the energy flow in such a way that:

  1. The energy continues to flow (metabolism)
  2. New sets of similar edges arise (reproduction)

The key thing that distinguishes life from other kinds of energy-flow systems - such as rivers or simple electronic circuits - is evolution by natural selection. Yes, if you want a very simple definition of life, let's say it's a system that evolves by natural selection. A living system pursues the root goal, which is an essential aspect of natural selection.

Life cannot strictly depend on biology. If we're going to build an AI that is alive, we obviously need a definition of life that allows for multiple realizability, so I must insist that an energy-flow system can be biological or non-biological. After all, the energy is the same. Some energy flows into you from the world, and it flows out of you again. That same energy can flow into a mechanical system, like a water pump that you are pumping. It's the energy that these systems have in common.

What about philosophical zombies?

It's conceivable that you could have a machine so perfectly designed that it seemed alive even though it was really just carrying out hard-coded instructions written in advance by a highly intelligent programmer for that programmer's own purposes. This would be a zombie.

The difference between a zombie and a living system is that the living system pursues its own root goal only. It does not serve another person's root goal at the expense of its own.

There is no eternal and absolutely foolproof strategy for achieving the root goal. Therefore, it's impossible to build a living system with hard-coded algorithms that specify actions in advance. A living system must be a neural network.

What about people who can't have children?

If the definition of life includes reproduction, does that mean infertile people are not alive?

This is a misunderstanding because the concept of evolution applies at the level of populations, not individuals. Of course infertile people are alive. Even though they can't reproduce individually, we count them as alive because they belong to a wider system, a population, that can reproduce and evolve.

Really it's the system that is alive. Individuals are alive insofar as they are part of the overall system.

Is a virus alive?

In biology, this is an interesting question, but it doesn't have any bearing on the project of building AI. If you're especially interested in viruses, then you might need a more sophisticated definition of life. For building AI, however, my definition of life suffices. This is consistent with the rule that we define words according to our needs.