Alan Turing’s Biggest AI Assumption May Have Been Wrong

Alan Turing’s pioneering work in artificial intelligence laid the foundation for modern AI research. However, his ideas about machine intelligence may have been built upon a fundamental flaw: the assumption that machines can replicate human-level intelligence by simply processing more data and increasing computational power.

This notion has driven decades of AI research, with researchers focusing on building larger and more complex models to mimic human behavior. But what if this approach is inherently flawed? Peter J. Denning’s new book, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, challenges the status quo by arguing that our pursuit of artificial general intelligence (AGI) may be doomed to fail.

Denning identifies tacit knowledge – those unwritten rules and intuitions that govern human behavior – as a major obstacle in replicating human intelligence. Humans possess a vast reservoir of implicit understanding that cannot be easily articulated or encoded into computers. This is not just a matter of data storage; it’s the very essence of what makes us intelligent.

One of the most compelling examples of tacit knowledge in action is our ability to understand and navigate social contexts. We can recognize sarcasm, humor, and emotion without needing definitions. However, when we try to teach machines to do the same, we’re faced with a seemingly insurmountable task: how do you encode the nuances of human interaction into a computer program? The answer is that you can’t.

This realization has significant implications for our understanding of intelligence and what it means to be intelligent in the first place. Are machines truly intelligent if they can only mimic human behavior, lacking the underlying context and culture that gives it meaning?

The consequences of this flawed assumption are already manifesting themselves in AI systems today. Large Language Models (LLMs) like ChatGPT and Claude may generate human-like text, but they lack the very thing that makes language meaningful: context. They can’t grasp the subtleties of human conversation or truly understand cultural nuances.

Denning’s book offers a compelling case for re-examining our assumptions about AGI and its potential uses. Rather than building machines that mimic human behavior without truly understanding it, perhaps we should take a step back and ask some harder questions.

The representation problem lies at the heart of Denning’s argument: computers are fundamentally unable to capture the nuances of human understanding. Words are mere symbols for meaning, not the meanings themselves. Yet, we continue building AI systems that rely on manipulating words without grasping their underlying context.

This creates a false promise of what machines can do. We’ve been sold on the idea that bigger language models will eventually unlock human-level intelligence, but Denning’s book suggests this is little more than a myth. Until we address the fundamental limitations of machine representation, we’ll continue building AI systems that are fundamentally flawed.

The cultural context is another crucial oversight in our understanding of human intelligence – its dependence on cultural norms, values, and history. We’ve long recognized culture’s importance in shaping human behavior but have failed to incorporate it into our approach to building machines that can understand and replicate this behavior.

This challenge facing AI researchers today is how to build a machine that understands the complex web of cultural norms, values, and history underlying human conversation. Denning’s answer is clear: you can’t. Yet, we continue pushing forward with LLMs unable to grasp this context.

As we move forward in AI research, it’s time to confront the fundamental limitations of machine intelligence head-on. Denning’s book offers a compelling case for re-examining our assumptions about AGI and its potential uses. Rather than building machines that mimic human behavior without truly understanding it, perhaps we should shift our approach.

We need to acknowledge the limitations of machine representation and culture, focusing instead on building systems that understand and respect these fundamental aspects of human behavior. The future of AI depends on this shift in perspective, as we can no longer afford to make the same mistakes over and over again.