djmurphy.net

E-bike for the mind

I really like Steve Jobs’ “bicycle for the mind” analogy. His observation was that when we rank the efficiency of animal locomotion humans are mid-tier, but humans riding a bicycle are far and away more efficient than other animals. We made ourselves best with a tool, and Steve Jobs imagined computers could be that tool for our minds - aids that we build for ourselves to exceed our natural abilities. The full clip is here:

Play

My take is, what he’s saying is computers enable us to go further - think bigger, ask larger questions, make connections we can’t make by ourselves, invent and make things that are not possible without their assistance - it’s not just more of the same thing.

I like bikes.1 I used to fantasize about quitting my day job, becoming a bike courier and just riding around all day getting seriously fast and strong. It seemed like a great job (in my 20s). I stopped fantasizing once bike couriers switched from fixies to those massive e-bikes you don’t need to pedal: it doesn’t sound fun to my inner cyclist to sit there and let the motor do the work all day.

I find myself worried AI will do to software engineers what e-bikes did to bike couriers. LLMs aren’t just better bicycles, they can put out more than you put in - they’re e-bikes. Everyone has been given motors for their mind-bicycles and the expectation is we will do more of the same. Right now it’s the veterans that are benefitting - and this makes intuitive sense - if you give every rider an e-bike, the ones who already have skills and fitness are going to go faster.

I haven’t yet heard anyone from leadership reflect on how AI adoption will or could change us. I have heard engineers concerned their code-muscles will atrophy and they’ll lose the skills they’ve built their careers and often their identities on. I haven’t yet heard anyone challenge us to explore ways to use AI as a training tool to improve and grow. What is our equivalent of using the motor the way mountain bikers do, to session challenging sections (using it to ride up hill to practice a tricky descent) and refine their technique? How do we make it an e-bike for the mind?

I have had a couple of interactions with LLMs that convince me they have broad training potential if we have the discipline and the capacity to use them. I know I’m not the best in meetings, so I tried prompting one to critique my meeting performance based on the transcripts.2 The very first time it ran, it pointed out that I’m great at providing technical solutions, but don’t commit to actually delivering those, or allocating resources, or determining we just don’t have capacity for that. I was wishy-washy on action. I relayed this to my wife who gave me a look she seems to reserve for when I’m being particularly clueless and said “you don’t know this about yourself?”. That was obvious in hindsight (and to my wife), and now my prompt asks to critique my meetings with respect to this behavior I want to train.

One of my new work habits has become to chat with AI about the state of the art in an area I’m about to work on for research and ideation. This has been very educational. I’ve worked through an idea to some concrete representation like a schema or API, and then really worked it with the help of AI - it’s like having a senior engineer on call to do technical reviews and critique on demand 24/7. This is leaning into a culture at Google where engineers would spend more time on design docs than actually writing code, in the acknowledgement that often once you had the idea down and agreed upon in sufficient detail, the rest is just typing. We can use AI to flex our design muscle (and when we’re done it can type way faster than me).

At home I’ve been using LLMs to generate conversations in Irish that are in topics that I’m actually interested in “talking” about - which is a lot more interesting than using Duolingo. (People my age learned Irish in the context of a very Catholic school system - everyone in Ireland knows how to say “can I have permission to go to the bathroom” - if there were ever an invasion this would be the shibboleth - but I realized I had no idea how to say “I need to pee”. There are some curiosities best satisfied by an entity incapable of embarrassment recognition.)

My explorations are pretty rudimentary, but ideally, I’d like for us to be working towards using this technology to build tools for training critical thinkers, like the children’s book in Neal Stephenson’s “The Young Lady’s Illustrated Primer”3 that gradually ramped up the challenge until a memorable scene where the protagonist derived the Turing Machine for herself while solving a puzzle - perhaps the very opposite of the Torment Nexus. While AI-accelerationists anticipate AI will solve all our problems by itself, I think using it to improve ourselves is a potential societal AI acceleration that is being overlooked - society getting smarter would be a benevolent outcome of this technology.4

We are in charge of our own skills, and many of us have the luxury of some autonomy in which tools and processes we bring to bear on our objectives. My challenge would be: let’s figure out ways to use the motor to better ourselves. Let’s shift the conversation back to skills and judgement and growth. Let’s remind each other that our industry is supposed to be making bicycles for the mind.

Footnotes

  1. Because it’s for a good cause, I’ll mention I’ll be riding America’s Most Beautiful Bike Ride in June to raise money for Blood Cancer United on behalf of my daughter who rang the bell this month. ↩

  2. Tip: an incantation to stop the AI from being sycophantic is “radical candor”. ↩

  3. https://en.wikipedia.org/wiki/The_Diamond_Age ↩

  4. I generally ignore recruiter emails, but if you’re working on anything like The Young Lady’s Illustrated Primer, please call me. ↩

No webmentions yet. Link to this post to see your mention here!