THE AGE OF AI // future of AI

AI is advancing at a remarkable pace, what’s coming next 5–10 years? // global scale, integration

The event focused on several key topics related to AI:

1. Transformative Power of AI. Discussed how AI is revolutionizing academia, industry, and society, with a focus on accelerating scientific discovery and solving complex problems like climate change and healthcare.

2. AI in Scientific Research. Explored how AI tools, such as OpenAI’s Deep Research, can streamline research processes, enabling scientists to focus on higher-level tasks and potentially leading to major discoveries.

3. AI in Business and Education. Highlighted the role of AI in transforming business processes, improving efficiency, and personalizing education through AI-driven tutoring systems.

4. Challenges and Risks. Addressed concerns about AI’s ethical implications, including explainability, data quality, and the potential risks of over-regulation, particularly in Europe.

5. Open Source and Collaboration. Emphasized the importance of open-source AI models and partnerships between academia and industry to drive innovation and ensure transparency.

6. Energy Consumption and Efficiency. Discussed the energy demands of AI systems and the need for efficient data processing and sustainable practices, while balancing the benefits AI brings to solving global challenges.

7. Future of AI and AGI. Speculated on the development of Artificial General Intelligence (AGI), its potential to self-improve, and the responsibilities of organizations in managing its societal impact.

8. Preparing for an AI-Driven Future. Provided advice for students and professionals on adapting to rapid technological changes, emphasizing critical thinking, adaptability, and the importance of understanding AI tools.

9. Regulation and Innovation. Debated the balance between regulation and fostering innovation, particularly in Europe, to ensure access to cutting-edge AI technologies while addressing societal concerns.

10. Exciting Developments in AI. Panelists shared their enthusiasm for AI’s potential in fields like healthcare, logistics, and scientific research, predicting significant advancements by 2025.

1. AI Progress and Scale

- The primary driver of AI advancement over the past five years has been “scale,” referring to increased data and compute resources, rather than significant algorithmic changes.

- Frontier models today are based on the same Transformer architecture introduced in 2017 but are much larger and more computationally expensive, with some costing hundreds of millions of dollars to train.

2. Concerns About Plateauing

- There are concerns that AI progress may plateau due to the high costs of scaling, questioning whether models costing billions or trillions of dollars will be feasible.

3. System One vs. System Two Thinking

- Inspired by Daniel Kahneman’s “Thinking Fast and Slow,” system one thinking is fast and intuitive, while system two thinking is slower and more deliberate.

- In poker AI research, incorporating system two thinking (allowing the AI to “think” for 20 seconds per hand) improved performance as much as scaling the model by 100,000 times.

4. Poker AI Success Story

- A poker-playing AI initially lost to top human players in 2015 despite being trained on trillions of hands.

- By redesigning the AI to prioritize system two thinking, the same team defeated top poker professionals in 2017 by a wide margin, surprising both the poker and AI communities.

5. Examples Beyond Poker

- Similar patterns of improvement from system two thinking have been observed in chess (IBM’s Deep Blue) and Go (DeepMind’s AlphaGo).

- Scaling thinking time by 10x in these games was equivalent to scaling model size and training by 10x.

6. OpenAI’s O1 Language Models

- OpenAI recently introduced the O1 series of language models, which incorporate system two thinking by taking more time to respond to complex queries.

- This approach opens new possibilities for solving important problems like developing cancer treatments or advancing scientific theories.

7. Cost vs. Value

- While system two thinking increases query time and cost, it can be justified for critical applications where accuracy and depth are prioritized over speed and cost.

8. Future of AI Scaling

- Combining system one (intuitive) and system two (deliberative) scaling paradigms could accelerate AI progress and unlock new capabilities.

- The speaker believes this marks the beginning of a new paradigm in AI development, emphasizing that the revolution is happening now.