A New Strategy for Artificial Intelligence: Training Foundation Models Directly on Human Brain Data

TL;DR

The paper identifies a fundamental limit of current foundation models: reliance on observable human actions $A(t)$, which may obscure deeper cognitive processes. It proposes training foundation models directly on brain data $B^*(t)$ as an approximation to $B(t)$, creating a brain-generated data stream to access a richer cognitive latent space. To operationalize this, it introduces two strategies—reinforcement learning from human brain (RLHB) and chain of thought from human brain (CoTHB)—and discusses extending these ideas toward brain-trained agents, AGI, and ASI, with attention to ethical and technical challenges. The proposed approach offers a realistic middle ground between scaling current architectures and pursuing neuroscience-inspired alternatives, while highlighting opportunities to improve alignment and trust through neurally grounded training data.

Abstract

While foundation models have achieved remarkable results across a diversity of domains, they still rely on human-generated data, such as text, as a fundamental source of knowledge. However, this data is ultimately the product of human brains, the filtered projection of a deeper neural complexity. In this paper, we explore a new strategy for artificial intelligence: moving beyond surface-level statistical regularities by training foundation models directly on human brain data. We hypothesize that neuroimaging data could open a window into elements of human cognition that are not accessible through observable actions, and argue that this additional knowledge could be used, alongside classical training data, to overcome some of the current limitations of foundation models. While previous research has demonstrated the possibility to train classical machine learning or deep learning models on neural patterns, this path remains largely unexplored for high-level cognitive functions. Here, we classify the current limitations of foundation models, as well as the promising brain regions and cognitive processes that could be leveraged to address them, along four levels: perception, valuation, execution, and integration. Then, we propose two methods that could be implemented to prioritize the use of limited neuroimaging data for strategically chosen, high-value steps in foundation model training: reinforcement learning from human brain (RLHB) and chain of thought from human brain (CoTHB). We also discuss the potential implications for agents, artificial general intelligence, and artificial superintelligence, as well as the ethical, social, and technical challenges and opportunities. We argue that brain-trained foundation models could represent a realistic and effective middle ground between continuing to scale current architectures and exploring alternative, neuroscience-inspired solutions.

Figures (5)

• Figure 1: Generative brain principle. Left: Human brain activity, B(t), can be approximated by neuroimaging data, B*(t), while generating observable actions, A(t). Right: The Venn diagram illustrates the relationship between the three sets, with the yellow area at the intersection of B*(t) and A(t) representing the observable actions that can be decoded from neuroimaging data. The teal area, included in B*(t) but not in A(t), represents the additional information that could be gained from incorporating neuroimaging data into foundation models. Future advances in neuroimaging technologies could gradually expand B*(t), resulting in a better approximation of B(t).
• Figure 2: Foundation model continuum. Foundation models are represented along two dimensions: the semantic gradient from predominantly device-generated data to predominantly human-generated data (x-axis), and the levels of organization (y-axis). Specialized foundation models are typically grounded in device-generated data, LMMs combine both data types, and LLMs rely primarily on human-generated data. Two empty regions highlight current constraints: the top-left corner, where it is currently difficult to imagine a foundation model of human cognition that does not significantly leverage human-generated data; and the bottom-right area, where it is equally difficult to imagine foundation models for natural sciences that do not primarily rely on device-generated data. Interestingly, a foundation model of human cognition grounded exclusively in neuroimaging data would naturally fall within the first empty region. However, we do not represent our hypothetical brain-trained foundation models there, since in the models we envision, neuroimaging data would serve as a complement, rather than a replacement, for more classical data types. The idea of fully brain-trained foundation models relying solely on neuroimaging data is intriguing, but beyond the scope of this paper.
• Figure 3: Cognitive levels.From left to right: Promising brain regions are represented for the perception (red), valuation (orange), execution (green), and integration (blue) levels, along with the corresponding limitation of foundation models that such neural signals could help overcome.
• Figure 4: RLHB and CoTHB.Left: In RLHB, a model output is presented to a subject, brain activity is recorded (here from valuation regions) as the subject evaluates the output, and neural signals are extracted to align the model. The grey arrow suggests that future outputs could be improved based on this feedback. Right: In CoTHB, a reasoning problem is presented to a subject, brain activity is recorded (here from execution regions) as the subject engages in reasoning, and neural signals are extracted to guide the model. The grey arrow suggests that the model could subsequently improve its ability to solve similar reasoning problems.
• Figure 5: Future of foundation models.Left: In the current strategy, classical foundation models are trained on classical training data, represented in yellow-green to emphasize the importance of observable human actions, A(t). Middle: In the new strategy, brain-trained foundation models would incorporate neuroimaging data, B*(t), alongside classical training data. The vertical grey arrow suggests that this strategy could also motivate further advances in neuroscience, unlocking a virtuous circle. Bottom-right: The promising brain regions and cognitive processes identified in this paper could help overcome the current limitations of classical foundation models. Top-right: The horizontal grey arrow suggests potential future advances toward brain-trained agents, AGI, and ASI.