Virtuous intellectual chores - now optional

AI Disclosure: Draft was written fully by a human. Heavy use of AI for editing. See On AI usage for more.

At this point no one doubts that AI is a tremendous capability boost for all human intellectual activities. There are interesting questions about what happens next - at what point this stops being a human operating an AI tool and becomes a fully autonomous AI. However, in this post I will stick to the current paradigm, and address critiques rising at the current level of capability.

There are numerous critiques of using AI for intellectual tasks. Roughly they say - AI is a trojan horse, we will slowly lose our own capabilities (e.g. The Machines Are Fine, backed by neuroscience research). The strongest versions go further: a Microsoft and Carnegie Mellon study warns that offloading routine work leaves our judgement “atrophied and unprepared” for the exceptions, and other research reports a strong negative correlation between AI use and critical thinking scores. While I think some worrying is important, I want to put forward a more optimistic version without ignoring the underlying concerns.

No matter how groundbreaking and novel this feels, technological innovation is not new. My daughter is almost six and learning arithmetic. She also has a tablet and I tried showing her the calculator app. In theory she can use it immediately to get answers to all her math homework exercises. But in practice she enjoys working out the simple exercises on paper in her workbook.

This story repeats for many innovations. After the initial disruption settles down, society starts to better understand the downsides and settles into best-practices and reasonable tradeoffs. Another example that comes to mind is after the invention and adoption of cars, we did lose some health benefits that basic daily movement provided. With some intentionality one can find a balance that involves using treadmills and going on walks for the joy of it, without demonizing cars when needed to carry groceries or go to the hospital in a rush.

I expect a similar path to be taken with AI tools. We haven’t had time to settle on best practices yet, but we will get there. Currently, we are in a chaotic absorption period, where the anti-patterns come a bit as a surprise, but in retrospect they will be very obvious and avoidable.

The main point of the naysayers is true: the grunt work that we offload to AI does build understanding. However, they are ignoring that the returns are diminishing and prior to AI many of us were doing grunt work way beyond the point its usefulness was gone. We just had no other option. With AI grunt work is now optional. We still get to do it whenever we need to build some of the core skill, but we can also offload it to AI, or find new hybrid modes where we collaborate with AI on more fine-grained tasks.

The right time to offload is murky and requires discipline and meta-expertise (expertise in thinking about the work rather than just doing the work). The incentives of industry and academia and the basic dopamine circuits in our head will lead us astray. But we will also learn from each other and processes and norms will emerge to make the right call for when to and when not to use AI. Paradoxically, the belief that we are collectively incapable of making the right calls is subtly dismissive of our human capabilities. It is saying - you don’t know what’s good for you and will make bad choices. I choose to be more optimistic than that. This is cautious (maybe foolish) optimism, not a solid prediction. But I would rather bet on us figuring it out than assume that we can’t.

Finally, I want to make sure all this worrying doesn’t let one forget how amazing all of this is too. Large swaths of intellectual work are just grunt work, and now no human needs to waste their limited precious time on earth doing it. Some of it was virtuous, but some of it was pure waste. No one can clearly separate the two, but we can approach a good balance. We can have our intellectual cars and airplanes, but also bicycles and treadmills. And sometimes we will turn off the computer and just go for an intellectual walk. The future is ours to choose.

On AI usage

As a meta-point, I am experimenting with different ways to use AI for writing and disclosing. What I am settling on is:

1. a running log of all possible blog posts - free-form throwing ideas to Claude, while it curates a running blog_ideas.md, deduping, making connections, keeping an outline, etc.
2. when I decide to write a post - I read the outline, but importantly don’t let Claude write the first draft. The draft is fully written by me. That way I don’t accidentally outsource the core spine of the post.
3. then I let Claude do heavy editing - typos, but also awkward phrasing, and discuss with me structural changes.
4. always disclose usage when publishing.