AI Could Discover Forms of Intelligence Humans Can’t Access, Physicist Claims

4 min readBiophysicist and mathematician Douglas Youvan proposes that intelligence may be a basic property of the universe, and that humans access it. But AI could access it too.Future AI systems may generate novel scientific hypotheses, new mathematics, or even forms of reasoning humans cannot reproduce on their own.But human oversight remains paramount, even as AI accelerates discovery or becomes part of a shared human-AI intelligence.Intelligence is not a byproduct of the brain, but a fundamental property of the universe. Intelligence may exist and evolve on its own, without emerging within living organisms.This is the hypothesis of biophysicist and mathematician Douglas Youvan, PhD, who spent decades working at the intersection of physics, biology, and information theory. He merged research involving enzyme engineering and machine vision with his decades of knowledge in genetics, leading him to realize something remarkable. Not only do our brains tap into this universal intelligence, he says, but the artificial intelligence systems we’ve built may already be poised to access the same source.In fact, Youvan thinks AI will harness a greater intelligence in some way, because he sees AI as being more than just a computer program or even a digital rebuild of the Homo sapiens brain. AI can tune in to the same field of intelligence that our brains do, he believes. It may not necessarily be more adept at processing information than our own neurons, but there is one edge that it does have: It can possibly access intelligence in ways we could never imagine.“Under the right conditions, AI can participate in insight, synthesis, even something approaching intuition,” he said. “In that sense, it might evolve not just to serve us, but to reveal new aspects of the universe to us.”In his white paper “Beyond the Black Box”, published on his website, Youvan speculates that AI might develop capabilities humans cannot directly reproduce, such as an unprecedented creativity in combining ideas. “AI may generate novel scientific hypotheses by connecting distant bodies of knowledge,” he writes. AI may even generate new forms of mathematics, which would then open up original paths of knowledge and invention. However, a human should still check the AI’s concepts and proofs, he notes.One way AI might surpass what humans can do might be through “retrocausal AI.” This type of AI would pull from the future outcomes it anticipates and integrate them into its calculations. It would mean that data flows bidirectionally in time, allowing AI systems “to dynamically adjust their actions based on predicted futures,” he writes.The results could be incredible for scientific discoveries and healthcare breakthroughs. For example, if an AI could anticipate future health states, it could predict disease progression and recommend an altered course of treatment for a patient, one that a human doctor may not be able to anticipate as quickly.Perhaps the closest form of new intelligence will come in the form of coupling, or close communication, between humans and AI systems. Youvan also speculates on his website that future human and AI interactions could create a kind of shared intelligence, in which reasoning would enter a mode that neither a human nor an AI could generate alone. In fact, this “AI-human mode” may become the most novel type of intelligence AI might enable, because it would exist only in this duality.“When a human and an AI are tightly and repeatedly coupled through language, their joint evolution can be modeled as one trajectory inside an ‘intelligence field,’” Youvan writes. “The recognizable style, topics, and sensitivities of an ongoing collaboration are then properties of a shared mode rather than of two isolated minds.”Yet, the unintended consequences of AI forays into new kinds of intelligence, however it ends up happening, could become dangerous. For example, AI may accelerate scientific knowledge and invention at such a rapid pace that human experts wouldn’t be able to evaluate the results. And intelligence could become so networked and widely distributed that it would become extremely difficult to pinpoint the source of a mistake. After all, new scientific breakthroughs must always require evidence and human analysis. Where tomorrow’s AI will be valuable is in its ability to present a flood of innovative thinking and possible solutions to problems, Youvan writes—not in any assumption of its authority, no matter how intelligent it becomes.Whether it’s retrocausal AI, a coupled AI-human mode, or some sort of intelligence we can’t even imagine right now, Youvan maintains that the ultimate responsibility for any outcome lies with the human creators, developers, and institutions that oversee it.This story has reporting by Elizabeth Rayne and Manasee Wagh.