Artificial Minds and Learning by Thinking:

Some of the most groundbreaking discoveries don’t come from observation alone but from thinking. Einstein developed his theories of relativity through thought experiments, while Galileo used mental simulations to derive insights about gravity. A new review published in Trends in Cognitive Sciences on September 18 reveals that this process isn’t exclusive to humans—artificial intelligence (AI) is also capable of learning by thinking.

What is Learning by Thinking?

Learning by thinking refers to a process where both humans and AI can self-correct and generate new conclusions without external input. Tania Lombrozo, a professor of psychology and co-director of the Natural and Artificial Minds initiative at Princeton University, emphasizes that AI systems, particularly large language models, demonstrate this capability. For instance, ChatGPT can often correct itself without explicit instruction, which mirrors human learning.

Examples of Learning by Thinking in Humans and AI

Lombrozo identifies four ways both humans and AI can learn by thinking:

Explanation: Humans often gain new insights by explaining concepts to others, revealing gaps in their understanding. AI, similarly, can refine its answers when asked to explain a complex topic.

Simulation: Before physically rearranging furniture, we mentally simulate different layouts. AI models can do the same through simulation engines, refining their outputs with each run.

Analogy: Comparing downloading pirated software to stealing physical goods helps us understand the moral implications. AI uses analogy to enhance its ability to respond accurately to questions.

Reasoning: If you know that tomorrow is a leap day, and a friend's birthday falls on leap days, you can reason that tomorrow is their birthday. AI, when prompted to use step-by-step reasoning, can arrive at more accurate conclusions.

Why is Learning by Thinking Valuable?

According to Lombrozo, learning by thinking functions as an "on-demand learning" tool. When humans or AI encounter new information, they don’t always know when it will become useful. This stored knowledge becomes relevant when the right context arises, leading to more efficient learning.Debating AI's Cognitive Abilities

The review raises important questions about the boundaries between AI thinking and simple data processing. While AI systems have become more sophisticated, Lombrozo acknowledges that there is ongoing debate about whether these systems are truly thinking or merely mimicking human cognitive processes.

Implications for Future Research

AI’s ability to learn through thinking opens new avenues for studying both human cognition and advancing AI technology. By comparing the two, we can gain deeper insights into how we learn and enhance AI systems by making them more aligned with natural cognitive processes.

In the end, the ability to think, reason, and learn without external input is not just a human trait—AI has proven capable of similar feats. As AI technology evolves, the distinctions between natural and artificial minds will continue to blur, providing fertile ground for future research.

Read Original article here