There may be a case for optimism in AI’s transformation of scientific discovery but it’s too early to be sure