Are other people experiencing this? Or just me?
I admit that I may be a bit wound up about the latest round of AI demos. But I’ve spent the last few months with the bizarre feeling of vertigo, standing on the precipice of history.
In the early days of the industrial revolution, there must have been some people who saw, not simply the neat new steam machine for pumping water out of the mines or the additional decimal point of accuracy in measurement, but the deeper truth behind what was happening: Humanity was escaping a hundred thousand years of limitations on our ability to harness and use energy to solve our problems.
A very smart person, in around 1760, could have potentially guessed many of the ways this would improve people’s lives over time, in terms of material abundance. But it’s unlikely they could’ve guessed the social changes, the governmental changes, the systemic changes in human interaction and class relations that would result over the next hundred, let alone 200, years.
Those 200 after 1760 saw the biggest changes for humanity as a whole since the advent of agriculture, and over a much shorter period. Imagine if they could’ve known what was coming, not just about the railroads, but also about the labor unions. Not just about the World Wars, but about the Holocaust. And not just the bad things either, what about the availability of food, the number of children who survive to adulthood? Vaccines.
They’re not just building chatbots. They’re decoupling cognition – or a near enough similacrum to it to be useful – from biology, like we did with energy in the 1760s.
When I think about the medium to long-term implications of the research that is currently generating large language models, I feel a bit like the hypothetical man in the 1760s, looking at the neat steam-powered gadget being used to pump water out of an English mine, and feeling the shadows of history loom ahead of him as he ponders the implications. Shadows loom for us
In the industrial revolution, we decoupled our ability to harness energy from the limitations of biology. *Now, we’re decoupling cognition from the same biological limitations. The end results, if progress continues, cannot fail to be as monumentous – and as unpredictable – as the last revolution.
Large language models are just a minor foreshock. True self-replicating machines, automated scientific research, and other exponential forces loom.
How will we distribute the benefits? Who will face the bulk of the harm? What unexpected delights and horrors will we see?