{"text":[[{"start":5.55,"text":"Regular readers of the FT might assume the world is becoming more unpredictable. Wars erupt, pandemics spread, markets gyrate, technology accelerates and politics fragments to an alarming degree. History appears to be on fast forward and we are all giddy with motion sickness. News is booming even if, sadly, the traditional news industry is not. "}],[{"start":28.3,"text":"But in some areas, at least, the opposite proposition may be true: the world is becoming more predictable, no matter how counterintuitive that may seem. The emergence of bigger and better data sets and the use of powerful AI models to spot previously undetectable patterns is in some cases demonstrably improving our predictive capabilities. Take short-term weather forecasting, for example. In spite of the chaotic effects of climate change, the best AI-enabled seven-day weather forecasts now match the quality of three-day forecasts made in 1980 thanks to better sensors, more data and vast computing power."}],[{"start":66.35,"text":"Might the same be true for predicting human-generated political and economic squalls?"}],[{"start":72.14999999999999,"text":"We certainly have better tools than ever to try to extract the signal from the noise, even as the geopolitical climate becomes more volatile. “AI is truly a breakthrough when it comes to forecasting events,” says Anthony Vinci, a former US intelligence officer, international affairs expert and author of The Fourth Intelligence Revolution. “I look at the world and no longer trust my personal opinion of what will happen. I need an AI tool to help me.” "}],[{"start":99.79999999999998,"text":"According to Vinci, there are four ways of trying to assess the probability of future events. A few individual human superforecasters are superb at such analysis. The collective intelligence of an organisation, such as the CIA or a political risk consultancy, can also be applied. The wisdom of crowds can be harnessed through prediction betting markets, such as Polymarket or Kalshi, although these can be manipulated. And a trained AI model can parse these three sources and crunch the data for additional insights. "}],[{"start":130.95,"text":"Vinci’s company Vico Technologies, an AI-powered forecasting, simulation and scenario-assessment engine, is one of several businesses attempting to do just that. Vico combines the power of generative AI models, operating within a human “harness”, to sift swaths of data to assess the probability of future events and their second- and third-order effects. Such forecasts can be useful for investors, companies and governments trying to anticipate when the Strait of Hormuz might reopen, for instance."}],[{"start":161,"text":"One recent highlight for Vico was its forecasting of the US attack on Iran. By February 19, Vico’s model ascribed an 89 per cent probability of this happening by March 31, compared with a 63.5 per cent probability on Polymarket. The attack was launched on February 28. Vico claims its overall “Brier score”, which measures the accuracy and confidence of predictions, is below 0.15, better than all but the best human superforecasters."}],[{"start":193.9,"text":"An increasing challenge for such predictive modelling is that the data fed into generative AI systems can be polluted, often by design. So-called LLM grooming, in which powerful interests try to poison the inputs and manipulate the outputs of large language models, can distort the knowledge base. The increasing use of synthetic data, created by AI, can also corrupt such models. In the worst-case scenario, this might even lead to model autophagy disorder (Mad), in which LLMs ingest their own output and eventually break down."}],[{"start":225.85,"text":"Carissa Véliz, an Oxford university philosophy professor and author of Prophecy, argues that data should always be regarded as a human artefact and that predictions tend to be downstream from power. “We create a mirror and we position it to reflect one thing or another, but it doesn’t reflect the whole world,” she says. “We forget all the time that the map is not the territory.”"}],[{"start":248.95,"text":"Prediction markets, in particular, can be exploited by insiders and can even face attempted after-the-fact manipulation. In March, the Israeli journalist Emanuel Fabian came under extreme pressure — including death threats — to change his report that missiles had struck a town near Jerusalem, invalidating some Polymarket bets."}],[{"start":268.59999999999997,"text":"AI forecasting models cannot be gamed in this way. But their long-term resilience has yet to be tested. Vinci says they will constantly have to adapt to a fast-changing world, alluding to the Red Queen’s warning to Alice in Lewis Carroll’s Through the Looking-Glass. “It takes all the running you can do, to keep in the same place.”"}],[{"start":294.15,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1781839282_8848.mp3"}