Co-Intelligence: The Definitive, Bestselling Guide to Living and Working with AI

by Ethan Mollick

AI is not a silver bullet, and there will be instances when it might not work as expected or may even produce undesirable outcomes. One potential concern is the privacy of your data, which goes beyond the usual questions of sharing data with large companies, and to the deeper concerns about training. When you pass information to an AI, most current LLMs do not learn directly from that data, because it is not part of the pretraining for that model, which is usually long since completed. However, it is possible that the data you upload will be used in future training runs or to fine-tune the model you are working with. Thus, while it is unlikely that the AI training on your data can reproduce exact details of what you shared, it isn’t impossible. Several of the large AI companies have addressed this concern by offering private modes that promise to protect your information, some of which meet the highest regulatory standards for things like health data. But you have to decide how much you trust these agreements. A second concern you might have is dependence—what if we become too used to relying on AI? Throughout history, the introduction of new technologies has often sparked fears that we will lose important abilities by outsourcing tasks to machines. When calculators emerged, many worried we would lose the ability to do math ourselves. Yet rather than making us weaker, technology has tended to make us stronger.