Thomas Kehler, a key member of the AIWS Natural AI Initiative:
Why we need to look to nature for a path forward
Earlier this week, the Boston Global Forum assembled a group of business, academic, and government leaders, calling for “A Natural AI Based on The Science of Computational Physics, Biology, and Neuroscience: Policy and Societal Significance.” I was a speaker at that event. The event resulted from discussions on finding a path forward given this year of AI success.
The accomplishments of AI in 2023 were so substantial that, for many, it is just a matter of pouring in money, data, and GPUs from this point forward. For many others, it is a destructive force to be regulated – even halted.
The power and productivity potential of generative AI is unparalleled in the history of AI. Pre-trained transformer models combined with the recognition that nearly all historical AI research areas can be mapped into a language encoding and decoding format fueled an explosion in results. This power is highly beneficial but also potentially destructive.
The success of today’s AI models is grounded in the mathematical apparatus of statistical physics. Specifically, deep learning neural nets derive from the physics of cooperative phenomena (magnetism and the flocking patterns of birds share the same math as deep learning). Principles from physics are foundational to today’s AI models. However, we are using the immense power of statistical thermodynamics without using physicists’ techniques to bind the generative power of solutions in real-world constraints. The call to ground AI research in development in a First Principles AI is a clear next step.
Pioneers in brain imaging and computational biology have made substantial leaps forward in linking mechanisms of perception and the physics of living systems to a fundamental principle of nature – homeostasis. The Free Energy Principle and the computational mechanism of Active Inference are based on the work of Karl Friston and other pioneers in brain science and computational biology. This work is tied to the profound insight that the principle of homeostasis applies to the physics of life and is an emerging general theory of the self-organizing principles of natural intelligence. Founders of the work in the Free Energy Principle and Active Inference are signatories to the attached letter.
Those who signed the attached letter believe the appropriate response to the tremendous advances in AI is to ground it in natural law. Generative AI must be grounded in cause and effect rather than filtered statistical possibilities.
The road ahead will benefit humanity and the preservation of biological lifeforms, provided we stay true to what brought us to this point of technological success. It rests upon our collective shoulders to bond to secure that road ahead.