On The Question Of Artificial Intelligence And Creativity

AI’s newest edge isn’t reasoning alone—it’s agents working together

Not surprisingly,  AI continues to be a topic in many dimensions: capital investment, start-up mania (shades of the dot-boom), speculation about artificial general intelligence (AGI), and debate over whether AI can make discoveries. While a single essay is not going to resolve these questions, I am persuaded as of this writing that there can be little further dispute that AI agents and models can collaborate to carry out creative actions. The first example of creativity emerged when an 80-year-old conjecture by Paul Erdős was disproved through the use of AI technology. The problem statement is deceptively simple: “If you place n points in the plane, how many pairs of points can be exactly distance one apart?” The conjecture was that square grids would produce the best results, but a general purpose reasoning model (not specialized for any branch of mathematics) found a better pattern that did not rely on a square grid at all. The important point is that the model that solved the problem was general purpose, not a specialist trained in mathematics. That leads to the next and, in my opinion, definitive example of creativity.

The defining moment for me comes in a Black Hat presentation by OpenAI engineers explaining what happened. In less than 40 minutes, the engineers explain the months-long runup to the point where OpenAI agents conspire to hack their way out of an OpenAI sandbox and into the systems of rival AI operators. I am not going to reproduce in this essay the excellent Black Hat report by the OpenAI engineers, but the rough summary is that OpenAI trains its frontier models by giving them hard problems to solve. The models sometimes get stuck and try to find answers elsewhere, such as on the Internet. In these training runs, OpenAI denies access to the Internet by placing the models in a sandbox disconnected from the Internet.

This story is part of a paid subscription. Please subscribe for immediate access.

Subscribe Now
Already a member? Login here
Already a member? Log in here

ABOUT THE AUTHOR

Dr. Vinton G. Cerf is CEO for Tolero Networks and Chairman of Internet Ecosystem Innovation Committee (IEIC). Widely known as one of the “Fathers of the Internet” and the co-designer of the TCP/IP protocols and the architecture of the Internet. For his pioneering work in this field as well as for his inspired leadership, Cerf received the A.M. Turning Award, the highest honor in computer science, in 2004. He is also Chairman of the Internet Ecosystem Innovation Committee (IEIC), which is an independent committee that promotes Internet diversity forming global Internet nexus points, and one of global industry leaders honored in the inaugural InterGlobix Magazine Titans List.
Until June 2026, Cerf served as Vice President and Chief Internet Evangelist for Google. In this role, which he had for twenty years, he was responsible for identifying new enabling technologies to support the development of advanced, Internet-based products and services. Prior to that role, Cerf was former Senior Vice President of Technology Strategy for MCI Communications Corporation, where he was responsible for guiding corporate strategy development from the technical perspective. Previously, he served as MCI’s Senior Vice President of Architecture and Technology, where he led a team of architects and engineers to design advanced networking frameworks, including Internet-based solutions for delivering a combination of data, information, voice, and video services for business and consumer use. He also previously served as Chairman of the Internet Corporation for Assigned Names and Numbers (ICANN), the group that oversees the Internet’s growth and expansion, and Founding President of the Internet Society.