Thalamic AI builds AI models with perceptual coherence so that, in moments of crisis and conflict, they may help us find the shortest path to peace and prosperity.

Is whether we should build artificial general intelligence (AGI) even the right question? We spent years studying neuroscience and AI and came to believe that the most essential aspect of human insight—from the arts to the sciences—is not intelligence, as it is widely understood. Our intelligence solves problems; but it is our clarity and depth that tell us what problems to solve and why. We believe that’s something different entirely.

In rapidly evolving, complex situations, clarity is what is hardest to maintain and depth is what is most needed. History’s darkest moments have come when our clarity collapses and our depth gives way to fear. On the battlefield, for example, we call this the fog of war. As language-based intelligence scales and AI systems are deployed into crisis situations, we risk an analogous fog of words. That risk reveals a deeper truth: intelligence has never guaranteed clarity or depth in man, nor should we expect it to in machine.

Transformers—the architectures behind current-generation AI models—are excellent tools for language-based reasoning. But a picture is worth a thousand words. While transformers can be applied to images and video, they work by breaking the world into correlated parts and fail systematically on tasks that require perceptual coherence: the brain’s ability to see the world’s parts and their causal relations as a unified whole. In the most complex situations—war, financial crises, natural disasters, epidemics—we need AI systems with this coherence property. A coherent AI does not answer your questions in a chat; it senses the whole situation, identifies what matters, and asks you the question.