The Language of Deception

Chapter 6The Imitation Game

While modern LLMs are a fantastical spectacle of lights and mirrors, they are ultimately only that. They are a complex and magnificent illusion. They are capable of presenting themselves as complex entities with awareness, social intelligence, and even emotion. But all these factors are learned from training samples consisting of human expressions of the same, and are established as patterns and correlations of the language(s) that we use so frequently to describe the human experience. Murray Shanahan of the Imperial College of London wrote:

Turning an LLM into a question-answering system by a) embedding it in a larger system, and b) using prompt engineering to elicit the required behaviour exemplifies a pattern found in much contemporary work. In a similar fashion, LLMs can be used not only for question-answering, but also to summarise news articles, to generate screenplays, to solve logic puzzles, and to translate between languages, among other things. There are two important takeaways here. First, the basic function of a large language model, namely to generate statistically likely continuations of word sequences, is extraordinarily versatile. Second, notwithstanding this versatility, at the heart of every such application is a model doing just that one thing: generating statistically likely continuations of word sequences (Shanahan, 2023).

Of Shanahan's two key takeaways here, the second—namely, that these systems are non-sentient statistical ...