When is automation bad?
AI is all very cool and fun to think about. But I view widespread use with growing unease because of what it does to our thinking. Why?
1. The labor theory of value — according to which, in at least one form, the value of an item produced is proportional to the toil in making it — is false. Some things are very costly to produce (imagine, for example, a statue made of your earwax), but have no value to anyone.
2. Some forms of automation are clearly good. I'm glad cars exist, and steam shovels, and dishwashers, and elevators. They multiply our efforts, reduce our misery, and enable meaningful work, play, creativity, and leisure.
3. Automation of creativity and cognition is not the good kind of automation. Indeed, it is dangerous. But why? It is not because effort alone is valuable (see point 1), nor is it because automation is bad (see point 2).
4. One way of phrasing the question at hand is this: where is the line? Where does the good kind of automation end, and the bad kind begin? Is there a line, after all? Does the line mark a difference of degree, of kind, or what?
5. Instead of asking whether work alone produces items of value, I'd like to shift our focus a bit: away from what is produced, and towards the person producing. My first hypothesis is this: whatever value there is in meaningful work alone manifests in developed human capacities, excellence, or virtue, not in the item produced. These are genuinely valuable, and their development and exercise can be hampered by automation.
6. Is automation as such bad, after all, then, since all automation displaces or leaves undeveloped at least some human capacities, excellence, or virtue? Perhaps so. To some degree. Philosophers might call this a 'pro tanto' bad: bad to some degree, and in some way. But something can be bad to some degree, and in some way, without being bad overall. Exercise, for example, is bad because it is temporarily unpleasant, but good overall because its effects make the effort worth it. One way of taking things here is to say that even the good kind of automation is pro tanto bad, but not bad overall.
7. My second hypothesis is this: what is displaced when automating creativity or cognition is far more precious than what is displaced in purely physical automation. Yes, there is excellence in shoveling well, which excellence is lost when using a steam shovel. But the loss of this excellence is far less tragic — indeed, not tragic at all — compared to the lost excellence of thinking well, as when someone has outsourced thinking to a machine.
8. So we have, then, one way to draw the line: look at what is displaced, and assess its value; when the human excellence displaced by automation is precious indeed, automation of its distinctive tasks is worse for us. We get the desired result, that automation of cognition or creativity is not so good compared to automation of physical labor, by way of the auxiliary hypothesis that the virtues of mind, as it were, are superior to those of the body.
9. I'm not totally satisfied with the proposal just yet. I don't like the implied hierarchy of human excellences, with the physical ones on the bottom, and the cognitive or creative ones closer to the top. And the whole setup, as though we were weighing things out, seems exsanguinating and false to my own experience in just the way cheap consequentialism does: the sort of thing Anscombe would rightly make fun of, and to great effect.
10. Perhaps automation of creativity and cognition as such is not so bad after all. Perhaps what matters is the context in which it occurs. To develop that idea, here is a third hypothesis. What matters here is not any one kind of hard toil, as such (for example: digging trenches, or writing long hand); what matters, instead, is that there be hard toil somewhere or other in a productive chain. It is the existence of such friction somewhere or other that enables the exercise of virtue, but not so important where the friction lies. So, the idea goes, automating some of your cognitive tasks — using a calculator, say — does you no harm, provided you are still doing something hard — thinking through a shopping list, say. You have not robbed yourself of opportunities to exemplify excellence in thinking, in using the calculator, the idea goes, since there are other opportunities to do so elsewhere in the chain of tasks.
11. Both practical and theoretical questions arise. I'll raise a few of the former.
- Why would anyone choose to make life harder by foregoing automation? Even if there is some rarified sense, visible perhaps to philosophers and other weirdos, in which meaningful toil is good for us, who has the fortitude to choose it? Of course some people will make this choice. We all know a guy who still uses fountain pens despite the mess. But that guy is an outlier. There are questions here about human psychology, but also institutional design: do we want to build situations that nudge us towards friction because it is valuable? That create friction?
- Where is the friction, where is the hard work to be done, in a world with increasingly sophisticated AI tooling? Prompt engineering? Hardly; a machine can do that for you just as well. In consumption? Not obvious. Customized slop built on the fly for just one pair of eyes is all too easy to slurp down; plug in and enjoy, no effort required.
- Must we despair? I actually think not. My own experience with using AI tooling is of vast new landscapes of meaningful effort opening up before me. Vibe coding up little games, as I do daily, gives me new and more things to do in total, not less. It feeds my mind new and delightful frictions to work through — though it does decrease my need to ever touch actual code. The world that gives me unease is not one where everyone does this; that's actually a pretty cool outcome, and for the best.
12. There's one more angle I want to try out here. Why is slop bad? One answer is that it took no effort to produce. Via the labor theory of value, we can immediately infer that it is of little value itself. Because the labor theory of value is incorrect, this can't be quite right, I think. But deeper reflection on what 'no effort' means can uncover an insight. When we put effort into producing something, we don't thereby make it valuable, but we do send a signal. The signal is this: "I found this valuable to produce, so much so that I gave of my time and life and energy, just to make it." This signal is absent when slop is slung into the void. Indeed, something like a counter-signal is sent: "I found this project of so little worth that I was unable to actually make something myself, so I outsourced it all; but I still think you should consume it. Enjoy!" This counter-signal is kind of annoying, and I think often disrespectful to one's audience. There may be more than one way in which slop is bad; but this is definitely one of them. Slop is disrespectful.
13. There's another way slop debases us; it fools its creators. When you copy and paste your slop from the chatbot into your Substack or social media feed, you feel as though you've done something. "I made this." A quick dopamine hit. It's like getting a reward in a video game — a new spell, a new power, an upgrade to your base. Feels good. But make no mistake. The feeling is no more true to reality than the one you get from a video game reward, and if you indulge in it much your grip on reality will diminish.
14. Bonus: Bitcoiners are persistently confused, incidentally, on some of these questions — about labor and value, that is. They think that because bitcoin rests on proofs of work — cryptographic demonstrations of energy and compute deployed in finding new blocks and adding them to the chain — that the result must itself be valuable. Incorrect! See point 1!
