There’s a scene in The Simpsons in which Homer’s half-brother Herb unveils his new invention – a machine for translating baby talk – and Homer tells him: “People are afraid of new things. You should have taken an existing product and put a clock in it.” Thus, in one offhand remark, did America’s favourite animated buffoon distil the message of an expanding body of scientific research: human beings do not like too much novelty. The innovations that take off combine a lot of familiarity with a little bit of new. The Simpsons itself illustrates this. When asked why the TV series was so popular, industry insiders said that it’s because it took an existing format – the sitcom – and injected animation and irreverent humour. But the rule holds across many other areas. In 2016, Kevin Boudreau, then of Harvard Business School, published an analysis of the way medical research proposals are evaluated for funding. Those that were highly innovative tended to garner lower marks than less innovative proposals. Highly conventional projects were marked down too, but there was a sweet spot in the middle: the highest marks went to projects that brought together received wisdom with some fresh thinking. A year earlier, Hyejin Youn, who studies complex systems at Northwestern University and the Santa Fe Institute, scrutinised all the patents filed in the US between 1790 and 2010 – using them as a proxy for innovation. She found that many of the patents until about 1870 represented new technologies, or genuine discoveries. From then on, however, innovation became more about combining existing technologies in new ways. It became modular, like Ikea furniture. Scientists have long attempted to understand the secret of successful innovation, with a view to guiding it and predicting the next big thing. Predicting innovation has a whiff of the oxymoron about it: if you can predict it, is it really new? Nevertheless, it has been done – sort of. An example is Moore’s law. In 1965, American engineer Gordon Moore predicted that the number of transistors that could be packed into a silicon chip would double about every two years. Though it’s a matter of debate whether his law still holds, most experts agree that it did so well into the present century. So the general trajectory of technological progress, if not individual innovations, does seem to lend itself to forecasting. Is it possible to be more precise? Two years ago, Andrea Tacchella and colleagues at the Institute for Complex Systems in Rome suggested that it might be. They noticed that the language used to describe innovation is, like innovation itself, combinatorial. The Wright brothers called their 1906 invention a “flying-machine”, for example, having no other word to describe it, and it was only a decade later that this hastily glued-together name was replaced by a shiny new one: “aircraft”. The codes used to classify patents are also modular. Under the International Patent Classification (IPC) system agreed in 1971, every patent filed is assigned a combination of letters and numbers depending on which of eight sections it falls under – examples include “Electricity” and “Fixed Constructions” – with additional letters and numbers adding detail. When technologies combine in new inventions, so do these codes. Tacchella’s group used this fact to try to predict future combinations of codes – and hence, future innovation. The first step was to feed about 7,000 patent codes into a neural network and let the network arrange them in space according to the frequency with which they appeared in a global patent database. The space in question was not physical space, obviously, but something more abstract: the space available for innovation. Once they had done this, they could identify zones in that space that had yet to be invaded by existing technologies. Such areas, which biologist Stuart Kauffman referred to in the context of evolution as the “adjacent possible”, are ripe for innovation. As the patent database evolved over time, the researchers could see pairs of codes moving towards each other as they cropped up in ever closer technological “neighbourhoods” – the same IPC sections, then subsections, and so on. This happened, for example, when codes for avoidance features in road vehicles and for obstacle detection converged in patents for selfdriving cars. Using this approach, they could predict an innovation up to five years before it happened. Tacchella, who is now employed by the European Commission, is adapting this method to try to guide innovation in the environmental sector. The idea is to analyse the language of regulations to pinpoint unmet needs – in reducing pollution, say – and then to direct people working on new technologies towards them. Meanwhile, two researchers from the Massachusetts Institute of Technology, James Weis and Joseph Jacobson, have used a machine-learning algorithm to identify past innovations in the field of biotechnology. Last year, they were able to retrospectively predict 19 out of 20 of the most significant developments made between 1980 and 2014. The next step will be to predict the future. There’s an enduring paradox for those who think about innovation: if technology is self-organising and progress predictable, what is the role of the inventor? Youn thinks of them as the people who, by some mixture of experience, curiosity and luck, find themselves at the edge of the adjacent possible. So it’s not surprising that, throughout history, several minds have converged on the same novel idea at roughly the same time. Witness Bell, Gray and Meucci, who came up with the telephone; Newton and Leibniz’s near-simultaneous development of calculus; and evolutionary theory as described by Darwin and Wallace. All of them transformed the human experience in incalculable but profound ways, which is why we remember them. Homer was right that people shy away from novelty, but they often appreciate it in retrospect. Herb’s talent was to spot the unmet need, and then meet it. Happily, his baby-talk translator made him rich, and he shared the proceeds with his half-brother. Further reading The Nature of Technology by W Brian Arthur (Penguin, £9.99) Where Good Ideas Come From by Steven Johnson (Penguin, £10.99) What Technology Wants by Kevin Kelly (Penguin, £15) This article was amended on 9 January 2023. Gordon Moore’s prediction was that the number of transistors that could fit on a silicon chip would double about every two years, not each year as an earlier version said.