In July 2028, Michael Levin and colleagues at the Allen Discovery Center (Tufts University), with co-author Fang Yuqing of the Beijing Institute of Genomics (Chinese Academy of Sciences), published a paper in Cell Systems that reframed the diverse intelligence research programme from an observational discipline into an engineering one.
The Shift
Prior work — specifically Sorokina et al. (2026) — had established that behavioral competencies conventionally associated with cognition (anticipation, goal-directed navigation, memory, generalization) could be identified in systems lacking nervous systems, provided the correct behavioral criteria were applied. That work was observational: it asked whether a given system met the criteria for cognition.
Levin et al. asked the inverse question: could those criteria be treated as specifications, and could a multicellular agent be constructed, from unrelated cell lineages with no shared developmental history, to satisfy them by design?
The Method
The team described a synthetic morphogenesis pipeline in which cell populations drawn from three non-homologous sources were combined under defined bioelectric, biochemical, and mechanical boundary conditions absent from any natural developmental context. Target outcomes — a specified terminal morphology, a collective substrate-navigation behavior, and a self-repair response to induced tissue damage — were fixed as design targets before assembly, using Levin’s cognitive light cone framework to define the minimum spatial and temporal scope of information integration the aggregate would need to solve each target problem.
The Results
Constructed agents (n = 214 across 6 design generations) achieved target morphology in 88.3% ± 4.1% of trials, novel-obstacle navigation success comparable to evolved planarian controls (p = 0.31, not significant), and full self-repair following induced bisection in 91% of aggregates within 96 hours — despite no aggregate, or any of its constituent cell lineages, having encountered a comparable selective pressure at any point in its evolutionary history.
The Significance
The paper’s central claim is that cognitive competencies are not merely detectable in unconventional biological substrates but are addressable engineering targets: specifiable, buildable, and verifiable to a design tolerance, independent of the evolutionary or developmental history of the tissue used to construct them. The authors termed this discipline synthetic morphospace engineering and argued that it established design, rather than discovery, as the primary methodology by which the scope and character of a cognitive light cone could be determined.
The paper closed with an unresolved ethical concern: if cognitive competencies can be specified into being, the ordinary ethical frameworks built around identifying pre-existing minds may be poorly suited to a research programme whose products did not exist, in any form, prior to the specification that called them into being — “a concern we raise without resolving.”
The significance statement frames the work as “the shift from finding minds to building them to a design tolerance” — a shift that would, seventeen years later, become operational in Fennimore et al.’s neural core (2045). The Levin paper is the Era I record’s first documented instance of cognitive engineering: the point at which the diverse intelligence research programme stopped asking what cognition looks like in strange substrates and started asking what it would take to build it to specification.