NSF AI Institute for Artificial and Natural Intelligence (ARNI)

When it comes to AI, “we are still in the early stages of understanding intelligence itself,” says Richard Zemel, the Trianthe Dakolias Professor of Engineering and Applied Science at Columbia Engineering. Zemel spoke on March 27 at Columbia’s Morningside campus as part of the Lecture Series in AI. Zemel leads a team of researchers at NSF AI Institute for Artificial and Natural Intelligence (ARNI), a National Science Foundation-funded initiative that brings together experts from AI, neuroscience, and cognitive science. Read more about Zemel’s lecture: https://bit.ly/4uMc75u

ARNI was excited to offer two engaging talks this spring exploring some of the most fundamental questions in AI and NeuroAI research. As part of the Columbia Engineering Lecture in AI Series, ARNI Director Richard Zemel presented “Integrating Past and Present in Continual Learning,” highlighting recent advances in how AI systems can continuously learn and adapt without forgetting previously acquired knowledge. The talk showcased ARNI’s ongoing cutting-edge research on continual learning, memory, and generalization inspired by human intelligence. ARNI also welcomed Ellie Pavlick, Assistant Professor at Brown University and Director of the NSF ARIA Institute, for our Distinguished Seminar Series. Her talk, “(How) Does AI Think?” explored recent findings in AI interpretability, examining how AI systems develop complex internal representations while remaining highly context-sensitive, and reflecting on the broader implications of comparing AI and human cognition. Both talks highlighted interdisciplinary topics and attracted students and faculty from across Columbia University, reflecting ARNI’s mission to advance dialogue at the intersection of AI, neuroscience, and cognitive science. #ContinualLearning #AI #HumanCognition #Neuroscience #CognitiveScience #Research #NeuroAI

Excited to support Andreas Tolias' impressive work on OmniMouse, which will be presented at ICLR2026, as a truly important step forward for neurofoundation models. The model introduces a unified multi-modal, multi-task architecture that integrates neural activity, visual stimuli, and behavior, enabling flexible tasks such as neural prediction, behavioral decoding, and neural forecasting within a single framework. Trained on 150B+ neural tokens from 3M+ neurons across 73 animals, OmniMouse achieves state-of-the-art performance and reveals a key insight: while increasing model size yields diminishing returns beyond moderate scales, expanding dataset size consistently improves performance across tasks. This suggests that, unlike language and vision models where parameter scaling drives breakthroughs, brain modeling is fundamentally data-limited—requiring richer, more diverse, and better-integrated neural datasets to unlock the next phase of progress. You can access this work at: https://lnkd.in/eTsie3aK As part of our neurofoundation models flagship efforts, this research helps position ARNI at the frontier of a rapidly emerging field, where AI and neuroscience come together to build unified models that can predict brain activity, decode behavior, and generalize across tasks. A powerful example of how NeuroAI is not just applying AI to the brain, but reshaping how we think about intelligence itself. #NeuroAI #FoundationModels #AIforScience #ICLR2026 #ARNI #AIResearch #Neuroscience

Excited to share this new paper, co-authored with an incredible group of Managing Directors across the 29 NSF National AI Institutes NSF AI Institutes Virtual Organization (AIVO) I truly enjoyed working with this team, bringing together diverse perspectives, challenges, and experiences into a shared vision. One of the key takeaways: project management in large, interdisciplinary AI initiatives isn’t just about keeping things running. It’s a strategic force that enables real collaboration, integration, and ultimately, better science. Proud to be part of this effort to rethink how we support and lead complex AI research. Grateful to everyone who contributed! #AIResearch #ResearchManagement #AIInfrastructure #ResearchLeadership #TeamScience #NeuroAI

NSF AI Institute for Artificial and Natural Intelligence (ARNI) participated in the Foundation Models for the Brain Symposium (NeuroNYC + e184), a timely convening on both the promise and the real-world constraints of this emerging space, from neural data and sensing to scalable modeling and deployment. Strong ARNI presence: Liam Paninski and Andreas Tolias on key bottlenecks, with Tolias delivering a keynote on advancing foundation models for the brain. Sean Escola, MD, PhD (Protocol Labs, ARNI collaborator) moderated the second Panel “from research to real-world systems”. Spyridon Mylonas, PhD, in his capacity as ARNI's managing Director, also participated in a breakout on NeuroAI ecosystem coordination, spanning technical, institutional, and governance dimensions. These discussions are increasingly central to aligning data infrastructure, partnerships, and shared frameworks needed to move the field forward at scale. At ARNI, neurofoundation models are a core priority. Work led by Liam Paninski, Andreas Tolias, Blake Richards, and Eva Dyer focuses on integrating neural data across tasks, modalities, and species, advancing both AI and neuroscience. Building on momentum from NeurIPS 2025 workshop, the direction is clear: progress in this space will require coordinated effort across disciplines, institutions, and sectors. #NeuroAI #FoundationModels #Neuroscience #ArtificialIntelligence #AIResearch #Neurotechnology #AIforScience

Strengthening U.S.–Canada collaboration in next-generation AI We were thrilled to welcome Luc Vinet (CEO of IVADO) and a multidisciplinary IVADO delegation of 30 outstanding researchers spanning AI, neuroscience, engineering, and AI policy, to Columbia for a dynamic exchange across AI, neuroscience, and data science. IVADO is a leading Canadian AI consortium that brings together academia, industry, and government to advance robust, responsible, and impactful AI, while accelerating talent development and real-world adoption. This engagement reflects a growing strategic connection between Columbia Engineering and the Canadian AI ecosystem. NSF AI Institute for Artificial and Natural Intelligence (ARNI) plays a central bridging role, particularly through our close ties with Mila - Quebec Artificial Intelligence Institute, where ARNI faculty member Blake Richards co-leads our work on neurofoundation models. These connections are further strengthened through our Director, Richard Zemel, and ARNI’s early development, which benefited from the support of Yoshua Bengio, highlighting ARNI’s role as a catalyst for cross-border collaboration.. Our discussions spanned large language models, multi-agent systems, robotics, neuroscience, and AI governance and safety, and crystallized shared priorities in continual learning and neurofoundation models. We are now actively exploring joint workshops, coordinated research efforts, and opportunities for trainee exchange as next steps toward deeper collaboration. We were particularly excited to connect with colleagues from Polytechnique Montréal, Université de Montréal (University of Montreal)éal, McGill University, and UNIQUE (Unifying Neuroscience and Artificial Intelligence in Quebec), further expanding this growing network. Looking forward to building on this momentum and advancing joint efforts across research, training, and innovation. #NeuroAI #AIResearch #GlobalAI

Happy Brain Awareness Week! We’re excited to share a new four-video series introducing ARNI’s work and the emerging field of NeuroAI, created in collaboration with Columbia Engineering Communications. The four short videos introduce the mission of the NSF AI Institute for Artificial and Natural Intelligence (ARNI) and explore a core question driving our research: What are the principles of intelligence shared by brains and machines? The videos explore how connecting advances in neuroscience and cognitive science with AI can help uncover the principles of intelligence, across both biological and engineered systems. Watch the ARNI video series: https://lnkd.in/e_ehzC-V Learn more about the project: https://lnkd.in/eTKN5Awj hashtag #NeuroAI hashtag #AIResearch hashtag #MachineLearning hashtag #Neuroscience hashtag #FutureOfAI

The 2026 NEUROAI Course is now accepting applications! Applications are open for the 2026 NeuroAI course, co-designed by Neuromatch Academy and ARNI. The NeuroAI Course provides participants with core concepts and practical skills in the rapidly evolving interdisciplinary field of NeuroAI. It emphasizes shared principles across neuroscience, cognitive science, and machine learning, with a focus on generalization, and offers hands-on experience implementing NeuroAI models. Neuromatch offers a suite of intensive, live, online programs organized around small collaborative groups (“pods”). Participants work closely with peers and a dedicated Teaching Assistant, culminating in a mentored group research project. Pods are matched by time zone, research interests, and when possible, language preference. 2026 Neuromatch Courses 6–24 July: Computational Neuroscience; Deep Learning 13–24 July: NeuroAI; Computational Tools for Climate Science These courses are ideal for advanced undergraduates, MSc and PhD students, post-baccalaureates, research staff, and early-career researchers. They are designed for anyone preparing for research at the intersection of neuroscience, machine learning, data science, and computational modeling. Participants will benefit from a structured, collaborative learning environment that culminates in a hands-on research project within an international cohort. There is no cost to apply. Tuition is adjusted based on local cost of living, and waivers are available for those who need them. Course details, dates, and FAQs: https://lnkd.in/gaEhbVHz Application portal: https://lnkd.in/eGKjFtup #Neuromatch #NeuromatchAcademy #AI #Neuroscience #NeuroAI #Courses #Research #ResearchTraining #Online #ApplyNow

Proud to see our former ARNI postdoc, Mehdi Azabou, become a startup founder—congratulations to Mehdi and the Constellation team on their $10M raise! Mehdi was a driving force behind several institute-wide collaborations at ARNI, and it’s exciting to see his vision translate into real-world impact. Constellation is hiring software engineers, AI researchers, and data scientists. #AI #AIresearch #Constellation #neuroscience #neuroAI #softwareengineers #datascientists