On Automation
I had a few conversations this week that highlighted how (surprisingly) repulsed I am by the framing of AI as "automation." Been considering why and want to record it here so I can compare later:
Lately when I find someone talking emphatically about automation, asking them questions tends to lead to them motioning at things they/you/we won't be doing anymore. That's disentanglement or disempowerment, depending on whether they're for or against.
Either way, they also tend to share an assumption that the point of collapsing steps in any process is to decrease effort. When they resent the work they're doing, that's a win. When they derive their meaning from the work they're doing, it's a loss. This is AI as a tool for deprecating involvement. Less and less participation until there's none at all.
In these convos I find myself having to pitch that:
- Effort is a fixed budget,
- Every process is a stepping stone between a worse way of thinking and a better way of thinking, and
- The good in our lives is directly proportional to our ability to use (2) to get more out of (1).
This is AI as a tool for amplifying involvement. Not more or less participation, but greater impact from what we already know we can give.
Right now I think my framing is slightly less common than the automators', but not by much. And I suspect it'd be the standard view if markets and true believers didn't reward visions of mass labor displacement; it's much closer to the typical perspective on new technology as a ratchet for reach, and the last few years have borne out that pattern.
Speaking of markets, I think the aspect of this technology that least maps to recent, widely-available tech jumps is that AI is dynamical. Very few people have experience applying dynamical systems, which means the learning curve is greater (and can't even begin until users realize that's what they're working with).
The closest recent example would be "virality" as a learnable feature of the internet, where it's less about telling the computer to do something directly and more about directing the attention of the network the computer is connected to.
Some inputs don't do much, some kick off massive waves of high quality response, and both sides of the equation evolve reciprocally (sound familiar?).
The other difference is AI's much-discussed trajectory. It's unavoidably more difficult to reason about the future of a technology that ratchets our reach AND its own reach AND then by extension our reach and so on ad infinitum/nauseam.
When I tease that out, it starts to feel more like a new kinetic force than a tool (or person, per se).