Stay Sharp

I bike to the train station to catch a train to work. Biking requires a lot more work than taking the train. If I were dedicated to maximizing my personal fitness, I would bike everywhere I go. I choose to prioritize getting to where I need to go efficiently, using only as much power as I need to arrive in a timely manner. Sometimes I need the power of a train, but even when I do I have always kept biking just a little so I stay fit.

Google started getting worse when it tried to add knowledge cards rather than direct us to search results. A summary is not a source. The web pages Google linked to were often confusing. They required wrestling with the material to understand the answers. The knowledge cards and now the AI summaries are pat. They pose no obstacles. They inspire no serendipity.

If you take the AI summary and only that, serendipity never reaches out and teaches you something you need to learn. The modern equivalent of Google knowledge cards is AI output, trying to tell me the answer without teaching me anything at all. When I need the power of AI output, my mind does not stay fit because I am not engaging with a primary source to learn something new. I am just repeating a fact I am told.

The only way I found to keep mentally sharp while staying on the train ride of AI and agentic compute is to treat AI like a list of search results I have not read. I do not treat AI output as the answers themselves. Doing my own research means wrestling with the evidence.

If a computer is like a bicycle for the mind, freeing us from drudgery to do creative work, then what is AI? AI can seem more like a train, moving at a high speed without much work from us, but it can be a bicycle. AI is still just a computer program, a mathematical model of the probabilities of words being next to each other. An AI can give an answer that is probably correct a lot of the time, which has real value especially at tasks where it can quickly and cheaply test if it is wrong. An AI does not learn from us, which means it can only take us so far. We will need to bike the rest of the way.

It’s in the wrestling to incorporate new knowledge that we learn. And it is learning that gives us the ability to imagine what comes next, even when the probability of that is very small.

An AI is trained on probable outcomes, not correct outcomes. It can never output the most unlikely but still possible outcome.

That is where human ingenuity is unmatched.

That’s how people who treat all computer programs like lists of links to follow, and discuss, and repeat back, who learn knowledge from primary sources and synthesize new ideas, can teach themselves to accelerate the process of human ingenuity. We will take the train when we need to, but ride a bike when we can. We put our hopes for the future into accelerated human ingenuity. We need to stay sharp.