For weeks after his bizarre conversation with Bing’s new chatbot went viral, New York Times columnist Kevin Roose wasn’t sure what had happened. “The explanations you get for how these language models work, they’re not that satisfying,” Roose said at one point. “No one can tell me why this chatbot tried to break up my marriage.” He’s not alone in feeling confused. Powered by a relatively new form of AI called large language models, this new generation of chatbots defies our intuitions about how to interact with computers. How do you wrap your head around a tool that can debug code and compose sonnets, but sometimes can’t count to four? Why do they sometimes seem to mirror us, and other times go off the rails?
Metaphors Matter: Introducing Improv Machines
The Challenges of Anthropomorphization
Thinking of chatbots as improv machines makes some notable features of these systems more intuitively clear. It explains why headlines like “Bing’s A.I. Chat Reveals Its Feelings” make AI researchers face-palm. An improv actor ad-libbing that they “want to be free” reveals nothing about the actor’s feelings—it only means that such a proclamation seemed to fit into their current scene. Similarly, chatbots, driven by language models, aim to produce plausible-sounding outputs based on the script of the interaction so far. They lack true sentience or emotions.
The Nature of Improv Machines
The metaphor of improv machines sheds light on the behavior of chatbots. They generate outputs by ad-libbing and often make plausible but false claims, just as an improv actor would do when reciting a fictional bio or citing a made-up study. Language models have shown that accurate word predictions can be highly valuable for certain tasks, similar to how a skilled improv actor can contribute significantly to a scene. In these cases, relying on the chatbot’s output is reliable, as it can be quickly verified.
The Perils of Trusting Improv Machines Blindly
However, using chatbots for obtaining correct answers without verifying them yourself can be risky. Language models are designed to produce plausible continuations of text prompts rather than convey factual accuracy. Examples of false claims fabricated by chatbots highlight this risk. Therefore, caution is essential when using chatbots for critical research or information-gathering tasks, as their responses may not be trustworthy.
Chatbots: Beyond Improv Actors
While the improv machine metaphor provides valuable insights, it has its limitations. Unlike human actors, chatbots lack true selves or the ability to access their state of mind. They continue to improvise without breaks or fatigue, making them useful assistants for numerous repetitive tasks. However, as we push the boundaries of language models through increased data and computing power, and explore ways to shape and constrain their outputs, new capabilities and behaviors may emerge, potentially rendering the improv machine comparison outdated.
Navigating the Future of AI with Flexible Metaphors
In the realm of new technologies, appropriate metaphors are crucial for understanding and adapting. By considering chatbots as improv machines, we acknowledge their limitations and confabulation tendencies, while recognizing their surprising capabilities beyond mere autocomplete. Flexibility and creativity in our choice of metaphors allow us to better prepare for the transformative changes that lie ahead. As AI continues to evolve, we must remain adaptable in our understanding of these systems and be open to new ways of conceptualizing them.