Robert Shelton
Robert Shelton
Ideas and knowledge, like seeds, are only valuable when people are able to grow them in their own garden.
Last year I flew all around the country advising companies and speaking at conferences about AI. The experience taught me that companies and individuals often overestimate the impact of providing more information to a system when the soil where that information to be planted is itself undeveloped.
Specifically in a business context, this overestimation may manifest as using AI to summarize and regurgitate your internal company documents faster, when the documents you are feeding the system are themselves wrong or poorly constructed. Here it doesn’t matter how fast AI can recreate the pattern if the pattern itself is wrong or inconsistent. Scaling a bad process is scaling a bad process.
This is not a localized business phenomenon however, it is unfortunately the nebulous difference between knowledge and understanding. In my own college experience for example, my classmates and I would manifest this same core gap, memorizing the shape of a solution from Chegg (a website where you’d pay to see old tests from similar professors) to get a good grade on a test, and then having absolutely no idea of any of the implications of a given set of solutions for a problem. We had knowledge but we didn’t understand. Many of the current existential fears about LLMs (aka ChatGPT etc.) are related to this gap because we have essentially trained models to accomplish exactly what me and my classmates were doing: function approximation.
The beauty and horror of function approximation as a method of understanding is it implies the user hardly needs to understand anything about what produces a phenomenon, as observed in data, as long as the pattern can be identified in a task-specific way. For example, to this day, I have no clue how to apply the information I learned about forces exerted on a spring within any sort of system. But did I effectively memorize how to solve the word problems in my physics class to come out with an A? You bet I did. LLMs can now accomplish many of these sorts of physics class type problems which has us collectively pondering about the future of work, education, and the difference between probability based model thinking and human cognition.
The difficult truth is the sort of pattern recognition that many of us did in school to gain our various titles (and that now LLMs can do) is often all you need to get by and is a valuable skill. There are real reasons investors are so hyped up about the technology because certain job functions will dramatically change. Here I think it’s important that we are able to hold two truths: AI is both valuable and limited.
I don’t know how things will change exactly or on what timeline but I have observed increasingly that the difference between good and great is not because of technical barriers but because the intention and care of the thing being worked on is lacking. The seed may be promising but the soil is not ready (Ask the Jets about Sam Darnold).
As I look forward then, I think what will become an increasingly important word to keep at the top of our vocabulary will be friction: the ability to wrestle with ideas and information in context over time to create that feeling of intuitive understanding and wisdom that often surpass the limits of description (read Mathematics as Metaphor by Yuri Manin if you want more mathematical rigor about the limits of natural language to describe phenomenon).
The gap between knowledge and understanding is the same difference between joy and happiness. It’s the grinding friction of attention and intention which is the price of quality. It is not something I know how to formulaically achieve – so that I might be more lazy – but is an aspirational state that requires designing my life and my future in such a way that I create meaningful intentional friction.
In closing I’ll point to the ritual of Shikinen Sengu, which I learned about from Phillip Seargeant’s book The Future of Language,. On the main island of Japan, there is a shrine dedicated to the sun goddess Amaterasu. It dates back to the 670s. However, the current buildings were only erected in 2013. This is because every twenty years they perform Shikinen Sengu, wherein the entire structure is rebuilt brick for brick. This ritual passes on traditional knowledge and, as Seargeant puts it, “Permeance is thus achieved through a process of repeated regeneration (The Future of Language, pg. 36).” This ritual is an inspiring example of designing a process around meaningful intentional friction and I wager that as we shave off more and more edges to our lives in the name of short sided "efficiency" we will be better served to think more along these lines.