Can a Virtual Fish Outsmart Your Cat? The Surprising Link Between Zebrafish and the Future of AI
January 12, 2026
By: Marylee Williams
Email: [Author's Email]
Media Inquiries: Aaron Aupperlee, School of Computer Science, [Email]
Aran Nayebi, an assistant professor at Carnegie Mellon University’s School of Computer Science, jokingly remarks that his robot vacuum has a larger computational capacity than his two cats, Zoe and Shira. Yet, while the vacuum mindlessly follows pre-programmed paths, his cats exhibit a remarkable ability to explore, play, and adapt with genuine autonomy. This contrast sparked a groundbreaking question: Can we create AI that mimics this natural curiosity and self-directed behavior?
And this is the part most people miss: It’s not just about making AI smarter—it’s about making it curious. Nayebi and his team at CMU have taken a giant leap in this direction by developing a virtual zebrafish that behaves like its real-life counterpart without any prior training. This isn’t just a simulation; it’s a glimpse into a future where AI agents could explore complex datasets autonomously, uncovering insights humans might overlook due to bias or limited perspective.
But here’s where it gets controversial: What if AI scientists, free from human biases, could revolutionize scientific discovery? Nayebi believes such AI agents could replicate moments of serendipity, like the discovery of penicillin, by processing vast amounts of interconnected data more efficiently than humans. For instance, in biology, where relationships between data points are intricate, AI agents could navigate these complexities with unparalleled precision. However, this raises a provocative question: Are we ready to trust AI with the reins of scientific exploration?
The team chose zebrafish for their research due to prior studies on glial cells in the fish’s brain. Initially overlooked, these cells were found to play a crucial role in the larval zebrafish’s ability to swim and explore. When biologists severed the zebrafish’s ability to use its tail, it entered a state of futility-induced passivity—a behavior where the fish, after repeated failed attempts to swim, temporarily stopped trying. Remarkably, interactions with glial cells eventually prompted the fish to try again. This resilience inspired Nayebi’s team to develop a computational model called 3M-Progress (Model-Memory-Mismatch Progress), which enables AI agents to explore and adapt without external rewards or labeled data.
Here’s the kicker: The model incorporates memory primitives—fixed assumptions about the world that the AI can reference. When new sensory experiences contradict these priors, the AI updates its understanding, much like a zebrafish learning to navigate its environment. Reece Keller, a Ph.D. student on the team, emphasizes that this approach captures not only zebrafish exploration behavior but also predicts whole-brain activity at single-cell resolution. This suggests that animal intelligence is deeply rooted in biological priors—a bold claim that challenges traditional AI design.
Unlike reward-based AI, like robot vacuums, the virtual zebrafish is driven by intrinsic motivation. It doesn’t seek rewards; it seeks understanding. Nayebi explains, “We didn’t show this virtual zebrafish how real zebrafish move. We created a simulated environment, let it explore, and evaluated its behavior afterward.” When researchers recreated the futility-induced passivity scenario, the virtual zebrafish exhibited similar behavior—a significant milestone, as it demonstrated this complex response without prior knowledge.
But here’s the real question: If AI can replicate such nuanced behaviors, are we closer to creating machines that truly think like animals? Nayebi believes this is just the beginning. As researchers tackle more complex problems, the solutions increasingly mirror biological processes, not because they’re imitating them, but because they’re converging on the same efficient principles.
The CMU team, including Alyn Kirsch, Felix Pei, Xaq Pitkow, and Leo Kozachkov from Brown University, is now exploring how this autonomy can be applied beyond zebrafish. Their next steps? Expanding this framework to different embodiments, potentially revolutionizing fields from robotics to neuroscience.
What do you think? Is AI ready to take the lead in scientific discovery? Or are we treading into territory better left unexplored? Share your thoughts in the comments—this conversation is just getting started.
