Updates
Publications
NeuroAI for AI Safety ↗
Nov 2024, Preprint
As AI systems become increasingly powerful, the need for safe AI has become more pressing. Humans are an attractive model for AI safety: as the only known agents capable of general intelligence, they perform robustly even under conditions that deviate significantly from prior experiences, explore the world safely, understand pragmatics, and can cooperate to meet their intrinsic goals. Intelligence, when coupled with cooperation and safety mechanisms, can drive sustained progress and well-being. These properties are a function of the architecture of the brain and the learning algorithms it implements. Neuroscience may thus hold important keys to technical AI safety that are currently underexplored and underutilized. In this roadmap, we highlight and critically evaluate several paths toward AI safety inspired by neuroscience: emulating the brain's representations, information processing, and architecture; building robust sensory and motor systems from imitating brain data and bodies; fine-tuning AI systems on brain data; advancing interpretability using neuroscience methods; and scaling up cognitively-inspired architectures. We make several concrete recommendations for how neuroscience can positively impact AI safety.
Can a MISL Fly? Analysis and Ingredients for Mutual Information Skill Learning ↗
Oct 2024, Preprint, NeurIPS IMOL Poster & ICLR Oral
Self-supervised learning has the potential of lifting several of the key challenges in reinforcement learning today, such as exploration, representation learning, and reward design. Recent work (METRA) has effectively argued that moving away from mutual information and instead optimizing a certain Wasserstein distance is important for good performance. In this paper, we argue that the benefits seen in that paper can largely be explained within the existing framework of mutual information skill learning (MISL). Our analysis suggests a new MISL method (contrastive successor features) that retains the excellent performance of METRA with fewer moving parts, and highlights connections between skill learning, contrastive representation learning, and successor features. Finally, through careful ablation studies, we provide further insight into some of the key ingredients for both our method and METRA.
Research Roadmap: Replacing Brain Cells to Combat Brain Aging ↗
Nov 2023, Whitepaper
Creating new brain cells to replace aged, dysfunctional ones is one “moonshot” strategy to combat brain aging. One hurdle is that the brain is a complex multicellular community comprising different cell-types (e.g., neurons, astrocytes, oligodendrocytes, microglia, vascular cells, mesenchyme, and others) enmeshed in networks of extracellular matrix and other components. Creating new brain cells to replace aged, dysfunctional ones may be a viable strategy to combat brain aging. Either new external cells can be transplanted, or alternatively, new cells can be created endogenously. Safely replacing aging brain cells with brand new ones could be transformative and will require intense focus and funding to realize.
Bottlenecks of Aging ↗
Oct 2023, Whitepaper
By 2029, the United States will spend $3 trillion dollars every year — half its federal budget — on adults aged 65 and older. By the same year, nearly 20 million Americans will die from age-related illnesses. Yet research on the biology of aging remains overlooked. Despite a 70-fold increase in funding for aging research since the last decade, the incentives of governments and for-profit investment do not always lend themselves to early bets on ambitious science or field-building. We outline initiatives in this whitepaper, which, if executed, could meaningfully accelerate the advancement of aging science and other life-extending technologies.
Curing Alzheimer’s by Investing in Aging Research ↗
Jan 2022, Policy Memo
Congress allocates billions of dollars annually to Alzheimer’s research in hopes of finding an effective prophylactic, treatment, or cure. But these massive investments have little likelihood of paying off absent a game-changing improvement in our present knowledge of biology. Funds currently earmarked for Alzheimer’s research would be more productive if they were instead invested into deepening understanding of aging biology at the cell, tissue, and organ levels. Fundamental research advances in aging biology would directly support better outcomes for patients with Alzheimer’s as well as a plethora of other chronic diseases associated with aging — diseases that are the leading cause of mortality and disability, responsible for 71% of annual deaths worldwide and 79% of years lived with disability. Congress should allow the National Institute on Aging to spend funds currently restricted for research into Alzheimer’s specifically on research into aging biology more broadly. The result would be a society better prepared for the imminent health challenges of an aging population.