AI forecasting models edge out human superforecasters as startups chase the market
Several startups are commercializing AI forecasting models that aim to predict stock moves, geopolitics and even pop-chart performance, according to an AFP report. FutureSearch co-founder Dan Schwarz said his models are already competitive with top "superforecasters" and predicted that by 2030 people will view past decision-making as "Stone Age thinking"; two AI models took the top two spots in a competition hosted by forecasting platform Metaculus last month, where humans and non-humans both competed. Preseen co-founder Veniamin Veselovsky said his system turned a $35 stake into $1.9 million in seven months, while colleague Theo Summer stressed that AI forecasting assigns probabilities rather than guaranteeing outcomes, cross-checking linked assumptions such as oil prices and Middle East conflict. Mantic co-founder Toby Shevlane said some clients already use its tools for trading decisions and that leading models now generate their own questions about the future and allocate probabilities across outcomes.
If forecasting models keep matching or beating human experts, decision-making in finance, government and insurance—and retail traders' odds in prediction markets—could shift substantially.
看好人工智慧（AI）能做出比人類專家更準確的預測，數家新創公司正將預測模型轉化為龐大商機，應用範圍像是股市走勢、地緣政治，甚至流行歌手的排行表現。
法新社報導，預測公司FutureSearch共同創辦人施瓦茲（Dan Schwarz）表示，到了2030年，「我們會覺得過去用來做決策的方式，完全是石器時代思維」。
AI預測模型憑著自公開與私人來源取得的大量資料，已與最優秀的「超級預測者」（superforecasters）不相上下。超級預測者是指對事件的估算結果，明顯優於一般人的天賦異稟者。