Terence Tao on AI: The Rise of "Artificial General Cleverness"

Terence Tao, a renowned mathematician and Fields Medalist, has recently shared his thoughts on the capabilities of current artificial intelligence (AI) systems. In a post on the Mastodon server mathstodon.xyz, Tao discusses the concept of "artificial general cleverness," which he believes is becoming increasingly prevalent in AI applications.

Why It Matters

Tao's perspective provides valuable insight into the limitations and potential of modern AI. His distinction between genuine artificial general intelligence (AGI) and what he terms "artificial general cleverness" highlights a nuanced understanding of the current state of AI technology. This distinction is crucial for policymakers, technologists, and businesses as they navigate the rapidly evolving landscape of AI.

The Current State of AI

According to Tao, true AGI-defined as an AI system capable of performing any intellectual task that a human can-is not yet within reach. However, he argues that a weaker form of intelligence, which he calls "artificial general cleverness," is becoming more common. This type of cleverness involves the ability to solve complex problems through ad hoc means, often using stochastic processes or brute force computation.

Tao emphasizes that these solutions may be ungrounded, fallible, and uninterpretable. They might also be derived from similar tricks found in an AI's training data, making them less a product of true intelligence and more the result of sophisticated pattern recognition and computational power.

The Opportunity

Despite its limitations, artificial general cleverness can still achieve impressive results, particularly when combined with stringent verification procedures to filter out incorrect or unpromising approaches. Tao suggests that this technology can be very useful and impressive in solving a wide spectrum of tasks at scales beyond what individual humans could manage.

For example, AI systems can now generate creative content, optimize complex logistics, and even assist in scientific research. These applications are not only practical but also have the potential to drive significant advancements across various industries.

Key Risks

While artificial general cleverness offers substantial benefits, it is not without risks. One of the primary concerns is the lack of interpretability. When AI solutions are derived from uninterpretable processes, it can be challenging to understand how decisions are made, which is particularly problematic in fields like healthcare and finance where transparency and accountability are crucial.

Additionally, the fallibility of these systems means that they may produce incorrect or suboptimal results. This risk is exacerbated by the scale at which AI operates, as even a small percentage of errors can have significant consequences when applied to large datasets or critical systems.

A New Perspective

Tao proposes a shift in perspective for understanding and utilizing current AI tools. Rather than viewing them as approximations of human intelligence, he suggests seeing them as stochastic generators of sometimes clever-and often useful-thoughts and outputs. This perspective can help in managing expectations and leveraging AI's strengths more effectively.

By recognizing that cleverness and intelligence are decoupled traits for AI, policymakers and businesses can better align their strategies with the capabilities of current technology. For instance, focusing on developing robust verification mechanisms to ensure the reliability of AI-generated solutions can mitigate some of the risks associated with artificial general cleverness.

Conclusion

Terence Tao's insights into the nature of modern AI provide a balanced view of its potential and limitations. While genuine AGI remains elusive, artificial general cleverness is already making significant contributions across various domains. By adopting a more nuanced understanding of AI, stakeholders can better harness its capabilities while mitigating associated risks.