The science underpinning this study is unfortunately incredibly limited. For instance, there isn’t even basic significance testing provided. Furthermore, the use of historic events to check forecasting accuracy, and the limited reporting of proper measures to prevent the model utilising knowledge not available to the human forecasters (with only a brief mention of the researchers providing the search link with pre-approved) is also very poor.
I’m all for AI tools improving decision making and have undertaken several side-projects myself on this. But studies like this should be highlighted for their lack of scientific standards and thus, we should be sceptical of how much we use them to update our judgments of how good AI should be at forecasting currently (which to me is quite low given they struggle to causally reason)
+1. While I applaud the authors for doing this work at all, and share their hopes regarding automated forecasting, by my lights the opening paragraphs massively overstate their bot’s ability.
The science underpinning this study is unfortunately incredibly limited. For instance, there isn’t even basic significance testing provided. Furthermore, the use of historic events to check forecasting accuracy, and the limited reporting of proper measures to prevent the model utilising knowledge not available to the human forecasters (with only a brief mention of the researchers providing the search link with pre-approved) is also very poor.
I’m all for AI tools improving decision making and have undertaken several side-projects myself on this. But studies like this should be highlighted for their lack of scientific standards and thus, we should be sceptical of how much we use them to update our judgments of how good AI should be at forecasting currently (which to me is quite low given they struggle to causally reason)
+1. While I applaud the authors for doing this work at all, and share their hopes regarding automated forecasting, by my lights the opening paragraphs massively overstate their bot’s ability.