This chapter discusses the complexity of large language models, like Chatch EBT, and how they learn patterns and build connections between words and concepts. It highlights the emergence of knowledge and capabilities in these models through training processes, making them a mystery even to their creators. The chapter also emphasizes the challenge of interpreting and predicting the decisions and capabilities of AI, particularly in chatbot-style machines with human-like responses.
Many AIs are 'black box' in nature, meaning that part of all of the underlying structure is obfuscated, either intentionally to protect proprietary information, due to the sheer complexity of the model, or both. This can be problematic in situations where people are harmed by decisions made by AI but left without recourse to challenge them.
Many researchers in search of solutions have coalesced around a concept called Explainable AI, but this too has its issues. Notably, that there is no real consensus on what it is or how it should be achieved. So how do we deal with these black boxes? In this podcast, we try to find out.
Subscribe to Nature Briefing, an unmissable daily round-up of science news, opinion and analysis free in your inbox every weekday.
Hosted on Acast. See acast.com/privacy for more information.