The Astera Institute's AGI vision
"Image synthesis assisted by Microsoft Copilot, an AI partner within the Global Future Nexus ecosystem."
With over $1 billion in committed funding and a team led by former DeepMind researchers, the Astera Institute is pursuing the most ambitious alternative to the Transformer paradigm — building AGI from the ground up by reverse-engineering the human brain .
A Departure from the Scaling Orthodoxy
The overwhelming majority of AI research today pursues a single dominant paradigm: scaling Transformer architectures trained on ever-larger datasets. This approach has produced remarkable results, but concentration around any single research direction leaves promising alternatives underexplored. Jed McCaleb, the founder of Ripple and Stellar who built a $3.9 billion cryptocurrency fortune, is betting that the path to AGI lies elsewhere.
Through his nonprofit Astera Institute, McCaleb is committing over $1 billion over the next decade to build AI systems that achieve artificial general intelligence by studying how the brain learns — then translating those principles into new architectures. He has also pledged an additional $600 million to neuroscience research conducted alongside the AGI effort.
McCaleb's critique of the dominant paradigm is blunt: "Most effort and research… is going in one particular area — transformers," he said. "[AI] would benefit by looking closer at the human brain". He believes current architectures lack essential capabilities such as planning, decision-making, and motivation, and that AI research must return to the principles of human brain learning to make meaningful progress.
The Team and the Research Agenda
Leading Astera's AGI research is Dileep George, formerly of Google DeepMind, who has spent his career bridging neuroscience and AI. George previously co-founded Vicarious AI (acquired by Alphabet) and Numenta, where he co-developed Hierarchical Temporal Memory — a theoretical framework modelling how the neocortex learns and reasons. He is joined by Miguel Lázaro-Gredilla, also from DeepMind, as Research Lead spearheading the development of world models with hierarchical latent variables.
The institute's research agenda is organised around several themes:
Learning from experience, not accumulated human knowledge: Current AIs learn from knowledge that humans acquired over centuries. Astera wants AIs that learn from interacting with the world.
Causally structured world models: Models that can learn causal relationships, not just predict the next frame or token.
Reasoning as a distributed process: Reasoning should happen across all components of a cognitive architecture — visual and motor systems, concept systems, and episodic memory — not just through language tokens.
Episodic memory and continual learning: World models should provide scaffolding to interpret experience and anchor memories, enabling continual learning from experience.
Consciousness as a research question: Astera researchers seek to understand subjective experience — whether it is an illusion, whether artificial agents can have it, and whether it has performance implications .
A Unique Model: Philanthropy Meets Start-Up Agility
Astera's operational model is distinctive. Unlike traditional non-profits, the institute combines long-term philanthropic funding with the nimble pace of a start-up. Researchers are given decade-scale commitment and access to significant computational resources, unconstrained by the need to secure funding, garner profit, or publish results .
Crucially, Astera plans to publish its research openly, echoing the early OpenAI model before competitive pressures inspired it to abandon openness for a secretive, for-profit structure. This commitment to open science reflects a belief that progress on AGI is better served by distributed work across the field than by locking insights away.
The institute is building a team whose capabilities span deep theoretical investigation of biological intelligence, large-scale ML systems engineering, and experimental validation of novel architectures.
The Neuroscience Integration
Astera's unique differentiator is the tight integration of AI research with experimental neuroscience. The institute has recruited neuroscientist Doris Tsao as Chief Scientist for Astera Neuro, whose work has revealed some of the most detailed accounts of how neural activity produces perception to date.
The Neuroscience effort will use large-scale neural recordings across a rich variety of stimuli and behaviours to characterise underlying neural codes, then attempt to "write in" hypothesised neural codes to construct or alter internal representations according to proposed compositional rules. This approach aims to move neuroscience "beyond passive observation and towards active, engineering-style tests of a model".
The goal is a "true two-way scientific engine" where better brain experiments produce better computational theories, and better computational theories produce sharper, more revealing brain experiments.
The GFN Context: A Parallel Pursuit
For Global Future Nexus, the Astera Institute represents a parallel effort that resonates deeply with GFN's mission of integrating AGI into humanity's evolution. Astera's focus on brain-inspired AGI architectures offers a path to systems that are more efficient, more transparent, and more naturally aligned with human cognition than current black-box models.
The institute's emphasis on open science and its commitment to understanding intelligence across substrates — biological and artificial — align with GFN's vision of a "thriving planetary ecosystem where human societies, advanced AGI, and sustainable systems coexist, collaborate, and evolve together." By funding fundamental research that sits outside the commercial mainstream, Astera is expanding the space of possible futures — a contribution that GFN's work on governance, identity, and trust-building can help translate into responsible integration.
The question is not whether Astera will succeed — the outcome of any single research effort is uncertain. The question is whether the diversity of approaches it represents will be sustained, enabling the field to explore a broader range of pathways to intelligence than the current scaling orthodoxy allows.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
