Tacit Knowledge

Following up on yesterday’s post, we can’t digitize empathy, feeling, etc. I have a certain kind of person that starts looking at me funny, because who cares about feelings. Everything can be rationally explained, right? Isn’t GenAI the most rational of rationalists? \<sarcasm>Isn’t empathy the downfall of western civilization?\</sarcasm>

OK, now stop anthropomorphizing that poor bucket of weights and biases. It doesn’t think, it can’t reason, none of these qualities I see people try and ascribe to it. It is an amazing pattern matching engine that can correlate a lot of ideas at the same time though, putting our limited human working memory to shame. Let’s give it credit where it’s due.

I want to channel two things here - first, and this gets lots of air time so I won’t dwell, is humans aren’t rational. If you need support for that pick up Dan Ariely’s Predictably Irrational, or better yet, Daniel Kahneman’s Thinking Fast and Slow, or anything in the field of behavioural economics.

Second, tacit knowledge is incommunicable by words. You can’t describe in words how to ride a bicycle to a human, nor to an “AI”, and have it pick up a bike and start riding. They have to learn for themselves. They have to translate through experience into tacit knowledge. (OK, calm down you robotics folks, each human’s sensors and actuators are unique)

I’ve talked to many experienced developers about why they took a certain path, heck I’ve tried to explain it to others myself sometimes, and so much of the time it’s just :shrug: I dunno, it felt right. Trust me. I’ve done it a hundred times.

Occasionally someone who is very deep in their field attempts to write out something important that they’ve learned about it, and it comes out very philosophical. They may even have to ask the reader to interpret words in a very specific way in order to find the precision they need. The best way to cross this bridge, to learn how you might communicate your tacit knowledge, is to begin coaching, teaching, mentoring others in the craft you know so well.

LLMs have ingested enough to answer questions with a regression to the mean. Sure highly guided by distinguished patterns, but the mean nonetheless. They’re generalists, not experts.

What we get when “vibe-coding/designing/writing” etc is iterative interplay between an LLM, which is very good at rendering ideas into code and words, and the tacit knowledge buried in the expert’s mind, which I called feelings in my last post, because that’s how I see and experience it.

Now, one last thing that’s on my mind - just because an LLM can’t explain to us what its weights and biases mean, that doesn’t imply that it has tacit knowledge. For that you’ll have to go back to my last post, because The Map Is Not The Territory. We can’t digitize most of the territory (yet?).

If you want to explore more about tacit knowledge, to understand the background of why it’s so important in this new GenAI world, check out this series of posts.