Here, I provide the first historical, theoretical, and empirical analysis of prophetic examples. I collect and analyze a novel dataset of over 2 million U.S. patents and applications from the biology and chemistry industries. I find that at least 17% of experiments in this population are fictional. Through both empirical and theoretical analyses, I weigh the potential costs and benefits of prophetic examples and find that the costs prevail. Prophetic examples could be beneficial if they help patentees; but I find little evidence that they do so, even in the specific situations in which they should be the most useful. Instead, prophetic examples likely hinder innovation because they prevent others from conducting their own experiments – even after the patent has expired and even if the prophetic example is incorrect. Prophetic examples also hopelessly confuse scientists – a shocking 99% of scientific articles incorrectly cite prophetic examples as if they contained factual information – which means that made-up results from patents contaminate the scientific literature.
Given these harms, I argue for a shift from prophesies to more clearly delimited hypotheses – roadmaps for future research, but nothing more – preserving what value there is in speculation while mitigating the clear harms of the practice. Beyond these concrete policy recommendations, my findings also have rich implications for theoretical debates about the physicality of invention, when and to whom patents should be granted, how patents transmit information, and, ultimately, how best to incentivize innovation.
June 25, 2018