It’s simple to state the primary limitation of predictive AI: it’s hard to predict the future. But with generative AI, to which we turn next, things are more complicated. The technology is re markably capable, yet it struggles with many things a toddler can do. It is also improving quickly. So, to understand what the limitations are and have some sense of where things might be going, it’s important to understand the technology. In chapter 4, we hope to demystify how generative AI works.
We’ll also discuss the many harms that arise from generative AI. In some cases, harms arise because the product is flawed. For instance, software that claims to detect AI-generated essays doesn’t work, which can lead to false accusations of AI-based cheating. In other cases, harms arise because the product works well. Image generators are putting stock photographers out of jobs even as AI companies use their work without compensation to build the technology. Of course, there are many applications of generative AI that both work well and are broadly beneficial, such as automating some parts of computer programming (although, even here, there are minor risks that programmers should watch out for, such as the possibility of bugs in AI-generated code that might give hackers an advantage). Given the focus of the book, we won’t spend much time on these beneficial applications. But we should emphasize that we are excited about them and about the potential of generative AI in general.