AGI-26 keynote
Hananel Hazan
Hananel Hazan, Ph.D., is an interdisciplinary computer scientist at the Allen Discovery Center at Tufts University, where he has worked since 2019 in Michael Levin’s lab.
“My primary research interest lies in understanding how biological cells interact and cooperate to achieve shared goals through modeling.”
— Hananel Hazan, Allen Discovery Center profile
One question, five substrates
Read the archive in order and the same question keeps getting asked of larger and stranger things: how much competence is already latent in a system’s structure, before you train it?
-
Reservoirs
Haifa, from 2010Reservoir computing and liquid state machines — architectures in which a large, fixed, randomly connected pool of units does the computing through its own dynamics, and only a small readout is trained. Hazan’s early work asks what makes such a reservoir hold up: how topology and stability bear on whether it survives damage to its own components.
Topological constraints and robustness in liquid state machines (2011) ↗ -
Spiking networks
UMass AmherstThe same commitment, engineered: spiking neural networks trained by unsupervised, reinforcement, and evolutionary learning, and the open-source toolchain to simulate them. Hazan led BindsNET, described on his Tufts page as “a state-of-the-art framework designed for rapid constructions of rich simulations of spiking networks.”
Locally connected spiking neural networks for unsupervised feature learning (2019) ↗ -
Living tissue
TechnionInstrumentation for talking to real neurons in real time. At the Technion he built a closed-loop experimental platform for studying cortical neuronal network interactions; the resulting Closed Loop Experiment Manager is published as an open and inexpensive solution for multichannel electrophysiological recording.
Closed Loop Experiment Manager (CLEM) (2017) ↗ -
Cells
Allen Discovery Center, TuftsIn Michael Levin’s lab the question turns biological. One paper opens by naming the premise it means to drop — “A common view in the neuroscience community is that memory is encoded in the connection strength between neurons” — and stores memory in spike timing instead. Another searches the parameter space of bioelectric circuits for patterning behavior.
Memory via Temporal Delays in weightless Spiking Neural Network (2022) ↗ -
Ecosystems
2026Habituation and sensitization located in a predator–prey model. The paper places itself, in its own words, “at the intersection of ecology, basal cognition, and mathematics.” It describes learning-like dynamics in a simulation — and claims nothing stronger than that.
Training Ecosystems (2026) ↗ -
Transformers
2026And the 2010 idea comes back. His newest first-author paper freezes a random backbone and trains only low-rank adapters, stating in its own abstract that the construction “is formally analogous to Reservoir Computing unfolded along the depth axis of a feedforward network.”
A Little Rank Goes a Long Way (2026) ↗
At a conference called AGI-26
Hazan is one of sixteen authors of a 2025 position paper titled “Stop treating `AGI’ as the north-star goal of AI research.” The paper argues that “focusing on the highly contested topic of `artificial general intelligence’ (`AGI’) undermines our ability to choose effective goals.”
It names, in its own words, “six key traps — obstacles to productive goal setting — that are aggravated by AGI discourse: Illusion of Consensus, Supercharging Bad Science, Presuming Value-Neutrality, Goal Lottery, Generality Debt, and Normalized Exclusion,” and calls on the community to “(1) prioritize specificity in engineering and societal goals, (2) center pluralism about multiple worthwhile approaches to multiple valuable goals, and (3) foster innovation through greater inclusion of disciplines and communities.”
He co-signed it; he did not lead it. We surface it here because he appears on the roster of a conference named for the thing that paper asks the field to stop treating as its north star, and because that is worth a real conversation rather than an omission.
Read the paper ↗The Levin seam
Michael Levin is the most frequent co-author in this archive. The shared work runs from bioelectric circuits explored by evolutionary heuristic search, through control flow in active-inference systems, to diffusion models read as evolutionary algorithms — the point where Hazan’s question about latent structure meets Levin’s about collective and basal intelligence.
Selected work
16 entries drawn from the archive, titled exactly as the records read, with his position in the author list. Everything else — including earlier Haifa-era work applying machine learning to human neuroimaging and speech data — is in the corpus.
- 2026Training Ecosystems: A Computational Approach to Uncovering Learning Behavior in Unconventional ContextsarXiv (Cornell University) second author, with Samanta and Levin
- 2026A Little Rank Goes a Long Way: Random Scaffolds with LoRA Adapters Are All You NeedarXiv (Cornell University) first author, with Zhang, Hartl and Levin
- 2026Heuristically Adaptive Diffusion‐Model Evolutionary StrategyAdvanced Science co-author, with Hartl, Zhang and Levin
- 2025Reservoir Computation with Networks of Differentiating Neuron Ring OscillatorsAnalytics last author
- 2025Quantifying Misalignment Between Agents: Towards a Sociotechnical Understanding of AlignmentProceedings of the AAAI Conference on Artificial Intelligence third author
- 2025Stop treating `AGI' as the north-star goal of AI researcharXiv (Cornell University) fourth of sixteen authors
- 2024Diffusion Models are Evolutionary AlgorithmsarXiv (Cornell University) third author
- 2023Control Flow in Active Inference Systems—Part I: Classical and Quantum Formulations of Active InferenceIEEE Transactions on Molecular Biological and Multi-Scale Communications fifth of seven authors
- 2022Memory via Temporal Delays in weightless Spiking Neural NetworkarXiv (Cornell University) first author, with Caby, Earl, Siegelmann and Levin
- 2022Exploring the Behavior of Bioelectric Circuits Using Evolution Heuristic SearchBioelectricity first author, two-author paper with Michael Levin
- 2019Locally connected spiking neural networks for unsupervised feature learningNeural Networks third author
- 2019
- 2017Closed Loop Experiment Manager (CLEM)—An Open and Inexpensive Solution for Multichannel Electrophysiological Recordings and Closed Loop ExperimentsFrontiers in Neuroscience first author, with Noam Ziv
- 2015Non-parametric temporal modeling of the hemodynamic response function via a liquid state machineNeural Networks second author
- 2011Topological constraints and robustness in liquid state machinesExpert Systems with Applications first author, with Larry M. Manevitz
- 2010Stability and Topology in Reservoir ComputingLecture notes in computer science co-author
Background
- Now
- Interdisciplinary computer scientist, Allen Discovery Center at Tufts University (since 2019), in Michael Levin’s lab.
- Previously
- Postdoctoral Research Associate, College of Information and Computer Sciences, UMass Amherst. Postdoctoral Researcher, Network Biology Research Laboratories, Technion — where he built a closed-loop experimental platform for studying cortical neuronal network interactions.
- Education
- Ph.D. Computer Science (2013), University of Haifa · M.Sc Computer Science (2007), University of Haifa · B.A. Exact Sciences (2002), Ashkelon College, Israel.
- Software
- Led BindsNET, “a state-of-the-art framework designed for rapid constructions of rich simulations of spiking networks.” He is listed first among the contributors on the 0.3.4 release record.
About this archive
This is an independent archive of Hananel Hazan’s publication record, built for Society of Minds Aligned and AGI-26. Records span 2007–2026 and are drawn from OpenAlex; preprints and versions of record are both kept, so the archive is a reading index, not a bibliometric count.
Biography, degrees, prior posts, and both direct quotations come from a single source: his Allen Discovery Center profile. Nothing here is asserted that neither source supports.
Not written by, reviewed by, or endorsed by Hananel Hazan. If you are Hananel and something here is wrong, the feedback control at the bottom of the page reaches us directly — and the claim bar will hand you the site.