Even though advancements like AlphaGo and LLMs have made significant strides in AI, we’re still far from creating AI that can navigate the complexities of the real world with its endless possibilities, incomplete information, and diverse goals.
While AlphaGo mastering a game like Go is impressive, it’s much simpler than predicting real-world events. In Go, the number of possible moves is limited, but in the real world, the options are nearly infinite. Similarly, engaging in a conversation with ChatGPT is simpler than real-world navigation, as conversations have limited context and responses, whereas real-life scenarios are vast and unpredictable. Additionally, in Go and chat interactions, the necessary information is available or inferable, but in real life, information can be unclear or missing. Lastly, the goals in Go and chat interactions are relatively straightforward—winning a game or providing an answer—but real-life goals and needs are often fluid and can change over time.