Life 3.0 is a mind that can redesign its own hardware

Tegmark defines life very broadly, "simply as a process that can retain its complexity and replicate" 1. What gets copied is information, the instructions for how to behave and for how to build the body that does the behaving, and in his terms those are the software and the hardware.

That gives him a clean way to sort every living thing, by asking which of the two it can change by design. Life 1.0, the biological stage, "evolves its hardware and software." Life 2.0, the cultural stage, "evolves its hardware, designs much of its software." Life 3.0, the technological stage, "designs its hardware and software" 1.

A bacterium is Life 1.0. Everything it knows how to do is written in its DNA, and it can't learn anything new during its lifetime. When its surroundings change, the species adapts only because a few individuals happen to carry a helpful mutation and out-reproduce the rest, so learning does happen, but it happens to the species over many generations rather than to any one cell. That system ran the planet for most of its history, starting roughly four billion years ago, and it's painfully slow.

Humans are Life 2.0. Our hardware, the body and brain, still comes from evolution, but most of our software gets installed after birth through learning and culture.

Here's the comparison that surprised me most, and I've never forgotten it. Tegmark writes that your DNA "stores merely about a gigabyte, barely enough to store a single movie download," while your synapses "store all your knowledge and skills as roughly 100 terabytes' worth of information" 1. The arithmetic backs him up, since the human genome runs to about 3.055 billion base pairs 2, and at two bits per base pair that comes to roughly 764 megabytes. A hundred terabytes is a hundred thousand gigabytes, so something like 99.999% of what makes you you arrived after you were born.

Life 3.0 can redesign both halves. An AI system is software we can rewrite, running on hardware we can rebuild, and in principle it could do both to itself.

It's starting to happen in small ways. Google DeepMind reports that its AlphaChip system, which lays out parts of computer chips, "has generated superhuman chip layouts used in every generation of Google's TPU since its publication in 2020" 3. TPUs (Tensor Processing Units) are the chips Google trains its own AI models on, so this is AI helping design the hardware that runs AI.

To be fair, some chip-design researchers have publicly disputed how much better AlphaChip really is than conventional tools 4, and it only handles one step of the design, so for now the loop has closed only partway.

So how long does life on Earth have left? NASA calls the Sun a main sequence star, the steady kind it is today, and expects it to "remain one for another 5 billion years before becoming a red giant" 5. Earth's deadline comes much sooner, because the Sun is slowly brightening: "Earth will become uninhabitable for higher forms of life in a little over 1 billion years, as the Sun grows warmer and dries our planet" 6.

Even so, a billion years is ten thousand times longer than human culture has existed. Squeeze life's four billion years into a single day and culture fills only the last two seconds before midnight, while Earth still has a little over six hours of livable time ahead. My guess is that whatever intelligence is around at the end of the Sun's life will be as different from us as we are from bacteria, whether or not it descends from us.