OpenAI’s Noam Brown Advocates ‘System Two Thinking’ for Future of AI

Noam Brown, a prominent research scientist at OpenAI, took to the TED AI stage to discuss the transformative potential of OpenAI’s new o1 model and its impact on industries through deliberate, strategic reasoning. Known for pioneering AI advancements, such as Libratus, the poker-playing AI, and CICERO, which excels at Diplomacy, Brown envisions AI as more than a computational tool, but as an engine for breakthrough innovation and complex decision-making.

“The incredible progress in AI over the past five years can be summarized in one word: scale,” Brown remarked to a captivated audience of tech leaders, developers, and investors. He emphasized that while the transformer architecture has advanced AI, scaling data and compute alone is not enough. “The main difference is the scale of the data and the compute that goes into it,” he said.

Advocating for System Two Thinking

Brown emphasized the importance of shifting from simply scaling data toward what he calls "system two thinking"—a slower, deliberate reasoning process akin to human complex problem-solving. Reflecting on his PhD work with Libratus, he shared, “It turned out that having a bot think for just 20 seconds in a hand of poker got the same boosting performance as scaling up the model by 100,000x.” Initially, he thought this result was an error, but it proved to be a game-changer. Brown now sees system two thinking as the key to moving beyond data-centric models.

Popularized by psychologist Daniel Kahneman, system two thinking allows humans to tackle challenging problems with careful, step-by-step reasoning. Brown believes this approach could lead to major gains in AI performance without the demand for exponentially increased data and computing power.

The o1 Model’s Real-World Potential

Following the September 2024 release of OpenAI's o1 models, which integrate system two thinking, Brown shared real-world applications that showcased the model's capabilities in handling complex tasks. For instance, the o1 model achieved 83% accuracy in a qualifying exam for the International Mathematics Olympiad, a substantial leap from GPT-4o’s 13%. Brown highlighted that the model’s reasoning capabilities make it particularly beneficial for fields that rely heavily on data-driven decision-making, such as scientific research and finance.

In practical terms, he believes the o1 model will enhance decision-making in sectors such as healthcare, energy, and finance. Using cancer treatment as an example, he asked the audience if they would be willing to invest significantly in life-saving technology. Brown suggested that the o1 model could enable faster data processing and hypothesis generation, potentially transforming how breakthroughs are made in areas like renewable energy and medical research.

Balancing Speed with Deliberate Reasoning

Acknowledging that some may be skeptical about slower, more deliberate AI, Brown addressed potential concerns: “When I mention this to people, a frequent response that I get is that people might not be willing to wait around for a few minutes to get a response,” he noted. However, for crucial problems requiring high-accuracy answers, Brown argued that this slight delay and added cost are justifiable.

System Two Thinking: A Distinct Approach in the AI Race

OpenAI’s focus on system two thinking could reshape AI’s competitive landscape, particularly in the enterprise sector. While companies like Google and Meta have heavily invested in multimodal AI technologies, OpenAI’s emphasis on deep reasoning positions it uniquely. For example, Google’s Gemini AI is optimized for diverse tasks, but it remains to be seen if it can match the problem-solving accuracy OpenAI's o1 models promise.

Nevertheless, the high cost of implementing o1 could limit its immediate adoption. Running the o1-preview model currently costs $15 per million input tokens and $60 per million output tokens—substantially more than GPT-4o. Brown suggested that, despite the expense, businesses that prioritize accuracy may find the investment worthwhile.

What This Means

Brown’s talk underscored a critical turning point in AI development: “Now we have a new parameter, one where we can scale up system two thinking as well—and we are just at the very beginning of scaling up in this direction.” If successful, OpenAI’s o1 model could pioneer a new phase of AI innovation, enabling more thoughtful, precise AI applications. This shift toward strategic reasoning may offer industries, from finance to healthcare, powerful tools for informed decision-making that go beyond traditional, data-heavy methods.