There is no rapid advancement

**The Limitations of AI: Understanding the Universe and Market Domination**

**AI’s Advancements in Gaming and Self-Play**

Back in 2017, Elon Musk expressed his concerns about AI, referring to it as “summoning the demon.” This fear stemmed from the rapid advancements of AI algorithms, such as Go agents, which were able to beat top human players through self-play. By the end of 2017, these algorithms also mastered Chess and Shogi. By 2020, they even became proficient at playing Atari games without the need for extensive simulation calls.

**The Complexity Gap: Go vs. the Universe**

Although these advancements in AI appeared threatening, it is important to consider the vast difference in complexity between the games AI conquered and the universe itself. Even without mathematical calculations, it becomes evident that the number of states in the universe far surpasses any Go game. One can imagine fitting the universe with numerous tiny Go boards, highlighting the significant difference in complexity between these two realms.

**The Role of Dynamics Models and Compression in AI**

Modern self-play systems, like MuZero, employ dynamics models, which predict the next state based on the current state and an action. These models essentially represent a world model that encapsulates the understanding of how the world functions. GPT-4, a dynamics model, operates under the same principles, but with the condition of the prior action space. This approach can also be simplified by understanding that the loss function for dynamics is compression.

According to the principle “compression is prediction is intelligence,” by creating a large compressive model from feeding the entire internet, one may expect to conquer the universe. However, the reality is much more complex.

**The Limitations of AI in Predicting the Stock Market**

While AI can provide valuable applications, such as predicting stock market trends, it does not guarantee dominance in the market. A hypothetical approach involving downloading historical stock market data and training a large AI model with numerous GPUs might seem promising. However, many hedge fund brokers have already adopted similar strategies, and most of them are not billionaires. The reason for this limitation lies in the fact that AI models would need to encompass all the computers actively participating in the market, as well as the presence of other market participants. To outperform the market, one would require superior computing power, which is unlikely.

**Contrasting AI in Go with AI in the Universe**

In the game of Go, AI models do not need to consider other computers as opponents since the game’s rules are fixed and independent of computing power. However, when it comes to modeling the universe, the inclusion of computers becomes crucial. Modeling computers within the universe is challenging, and even with extensive self-play, understanding and predicting their behavior would be difficult. Additionally, assuming equal distribution of compute power, no single system would possess a significant majority of compute power, preventing one system from dominating.

**The Importance of Compute Distribution and Regulation**

Even with the potential benefits of AI advancements, it is crucial to ensure that compute power is not concentrated in a single entity. Compute distribution typically follows a power law distribution, where the majority of compute power is spread among various systems. Instead of artificially restricting FLOPS (Floating Operations Per Second) in training runs, effective regulation should focus on preventing a 51% attack on compute power. By maintaining a balanced compute distribution, the risk of a single dominant system emerging can be mitigated.

**The Nature of Progress: Harder-to-Find Solutions**

In AI development, it is essential to understand that progress does not happen overnight. Similar to the industrial revolution, which gradually transformed energy sources, the information revolution will do the same for intelligence. Progress follows an exponential growth pattern, and time is required to unlock new solutions. While intelligence may experience a revolution, the complexity of the universe ensures that progress will be a gradual process.

**The Future of AI: Letting the Markets Evolve**

As long as the development of AI is not artificially restricted, there is no imminent risk of a “FOOM” (Fast Takeoff of Machine intelligence) scenario. Instead of rushing into regulating AI, it is beneficial to allow the markets to develop naturally. By avoiding unnecessary interference, the potential risks associated with AI dominance can be better mitigated.