sids blog

We discover new knowledge by thinking from first principles—building knowledge from the ground up. Current AI models have access to past data, but they also need real-time access to the world, to see and perceive it the way we do. They must make falsifiable assumptions about the world and then test those assumptions against reality in different situations.

So, if I have to create an AI scientist, it means I need to find a way for the AI to access real-time data from the world by observing it through sensors. Then, I need that AI to think about the data from first principles and not rely solely on its training data. It must make assumptions about the world and then go out into the real world to test those assumptions against reality.

If I am going to develop a scientific AI, what area should it focus on? Personally, I find physics fascinating, and fundamentally, everything boils down to physics. So, we can create an AI that tries to figure out fundamental truths about the universe—astrophysics, quantum physics. Finding a theory of everything would be a great achievement. One intriguing idea is an AI capable of building simulations to test the laws of physics or any hypothesis. An AI that can perfectly mimic reality in simulations could perform all possible scenarios and understand their outcomes. This would allow us to deploy solutions for real-world problems without causing harm to civilization, the environment, or any species.

I am very optimistic about what we can build with AI and the things we can achieve with it. Even in today’s world, most inventions happen without any significant contribution from any single individual. Even those creating breakthroughs contribute only a small fraction to the total number of discoveries occurring. Most people do not invent anything; they simply consume most things that others have invented or discovered. If an AI can discover fundamental truths about the universe faster than us, then I believe we should definitely develop such an AI. We can later try to understand those discoveries for the sheer fun of it, just as we read books to grasp concepts that others have explained well. Our goal would simply be to understand and satisfy our genuine curiosity.

I believe such a society would flourish, and we would outlast any other species that has ever existed on this planet. We would become truly multi-planetary and explore the entire universe. If we create a superintelligence so capable, we can certainly develop one that explains the mysteries of the universe in a way humans can understand. Perhaps the meaning of life will be lifelong learning, where every human on the planet is driven by genuine curiosity. This would lead to a blissful existence.