Consider these chains of causation:
The invention of the stirrup required bigger saddles to keep riders on their mounts, which meant horses had to be bred larger with more land to feed them, so lords employed peasants to manage their growing estates, giving us feudalism and eventually the nation-state.
The plough industrialised agriculture, meaning farmers had to irrigate, leading to expanded government to manage the floodgates; the taxes to pay the bureaucrats who governed had to be recorded, so we got written numbers and eventually mathematics.
The sudden illness of the Caliph of Baghdad led to a flowering of Arab astrology and astronomy, which required accurate timekeeping for observations and inspired Christian monks to come up with new mechanical clocks, which were then improved for use on ships and led to the exploration and conquest of the rest of the world by Europeans.
These are just three of the fascinating historical threads traced by James Burke in his seminal 1970s television series Connections. In addition to entertaining millions with his erudite and peripatetic approach to history, and filming the most carefully timed shot in the history of television, Burke made a strong case for scientific gradualism, citing example after example of technological breakthroughs that, like the examples above, came about through the combination of existing ideas rather than sui generis via a sudden Eureka moment.
The entire Internet and all the profitable innovations itâs brought us are products of Burkeian recombination, but thatâs not why I bring him up here. Instead, consider Googleâs recently announced robo-scientist, which comes complete with breathless praise for its potential to accelerate research using âadvanced reasoningâ to produce ânew hypothesesâ. Does this mean an end to the stumbling, haphazard progress that Connections charted? Perhaps we will no longer need to wait for, say, a new method of boring cannons to collide with the brew kettle used to make beer; we can ask our shiny language model to invent the steam engine and then just wait for it to churn out a patent application.
Not so fast. As the New Scientist points out in a recent article, Googleâs model is following my First Law of AI: itâs a super-speedy research assistant, sprinting through the library and bringing together loads of existing ideas to find potential connections among them, rather than generating totally new avenues of investigation and writing them up for a journal. One researcher using the system found it very useful, but noted that âEverything was already published, but in different bits. The system was able to put everything together.â I keep begging you not to use ChatGPT to serve up slop in the name of âcontentâ, and similarly we shouldnât expect these machines to replace grad students in research labs cranking out scientific papers, at least not if we want genuine innovation.
Instead, ask yourself what sources of data you already have that are âpublished but in different bitsâ, and how you could clean and combine those elements into something as valuable as, say, street lights powered by coal gas, a discovery made possible by a candle, a leaky oven, and the need to cover ships with tar. Harnessing serendipity is the future of AI, including in your business and that of your competitorsâwhat steps are you taking to accelerate it?
This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 3rd March 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/
