¹⁸In fact, calculations of the amount of data that humans observe in real life show that the number of data points needed in AI is often much larger than what humans need. For example, children seem to learn to speak from a relatively small number of “input” words (Cristia et al., 2019); current AI systems need orders of magnitude more. Perhaps even more strikingly, humans can learn from a single example they see or hear, as I have pointed out earlier, see e.g. Lake et al. (2015). This shows that there is a lot of room for improvement in AI compared to the brain in terms of efficiently using all the data.