theuth's bargain: the economics of not knowing
In Plato’s Phaedrus, the god Theuth invents writing and brings it before King Thamus, expecting praise. Instead, Thamus warns that this new invention will not make people wiser, only more convinced of their own wisdom. Those who rely on it, he says, will carry the appearance of knowledge without ever doing the work of possessing it.
They will appear to be omniscient and will generally know nothing; they will be tiresome company, having the show of wisdom without the reality.
— Plato, Phaedrus
Most importantly here, Plato is not saying that reading is useless. The myth itself describes writing as useful as a reminder for someone who already knows something. The skepticism is aimed towards treating this written record as a substitute for understanding.
I think one of the more dangerous effects that LLMs will have on society probably isn’t the uprising of killer bots or the replacement of jobs but rather that the pursuit of knowledge will effectively be killed.
And I think what’s more disappointing is that there are many who are closer to my age, fresh out of college or just starting, that have a deep yearning to master a particular concept, field of work or study, that unfortunately give up incredibly early because they see the objectively miraculous things that LLMs can do, and think “it’s probably not conducive for me to learn this thing anymore.”
A huge factor in this public fervor is of course, the depraved and degenerate echo chamber that is X (and other social media forums and channels). Beyond that, others just prefer that they can keep a roof over their heads and live a life worth enjoying and passing that life on to their children. That’s always been the case with Computer Science in particular; many graduated from their degree and did close to nothing outside of work because their jobs were just that– a place to show up to 5 days out of the week to earn a paycheck in return.
But, it would be a disservice to chalk this up to laziness or instrumentalism. There is something deeper, that I think many lament, about the fact that since the rise of these tools, most notably for software engineers and others working in this space, you often end your week with a net-zero gain in knowledge. Despite getting your work done, or even having generated a considerable amount of income, you can’t really take inventory of the new things you’ve learned.
For the little you do learn, it makes negligible difference on your career prospects due to the competitive nature of the market right now. And here, it becomes hard for me to understand why someone might refuse to put time and effort into understanding their work.
Perhaps the useful distinction to land on is that of Plato, however many centuries ago. Writing was not worthless because it externalized memory and likewise, LLMs are not dangerous because they externalize parts of reasoning. But, when the output of the tool becomes indistinguishable, in our own minds, from something we understand, that is when we should begin to ruminate on this issue.
I know that an LLM can (and probably will) do my job better than me. We can even operate under the assumption that this is inevitable. But I don’t want to merely “do my job”. It does not make sense that because a machine can, that we shouldn’t.