Michael Darius Eastwood · Independent AI alignment researcher
Published
In 2005, Dinah Murray, Mike Lesser and Wenn Lawson published "Attention, monotropism and the diagnostic criteria for autism" in the journal Autism. Twenty years on, the framework has become one of the most cited attention-based accounts of autistic cognition. It also happens to describe, in more or less exact terms, the way certain kinds of research programme actually get built.
Monotropism proposes that attention is a finite resource that can be distributed broadly across many concurrent demands (polytropism) or concentrated intensely on a few (monotropism). Neither pole is intrinsically better; each has costs. Broad distribution enables rapid task-switching, social attunement, and background scanning. Intense concentration enables the kind of sustained, layered, self-directed engagement that produces books, working papers, technical builds, and long research arcs. The autistic attention profile, on the framework's account, sits toward the monotropic end. When such an attention system meets a high-interest domain, the result is what the wider literature calls hyperfocus: depth of engagement that can produce output disproportionate to formal training.
The temptation is to celebrate this as a superpower. That reading is inaccurate and unhelpful. The same attention system that enables extraordinary depth in favoured domains struggles disproportionately with distributed demands: switching tasks, processing low-stimulation administrative communication, maintaining consistent output across contexts, and meeting time-pressured pacing. The asymmetry is the architecture. Russell and colleagues' 2019 interviews with autistic adults described this exact profile, in autistic voices: the ability to hyperfocus, attention to detail, and superior pattern recognition alongside significant difficulty with distributed everyday demands. Grove and colleagues' 2018 work on special interests documented both the well-being benefits and the costs of high-intensity single-domain engagement. The evidence base treats monotropism as a two-sided instrument, not a gift.
A monotropic attention system pointed at a difficult problem for long enough produces recognisable outputs. Sustained connection between fields that are usually considered separate. Structural pattern transfer that reads as insight to some observers and as tangential to others. Independent research arcs completed without the scaffolding that institutions typically provide. Working papers that name a specific vulnerability class before the empirical literature does. The 8 December 2024 manuscript in the ARC/Eden evidence set (Google-server-timestamped via Gmail Message-ID, SHA-256 published in the canonical facts register) is one such artefact. Ten days after that manuscript, Anthropic reported alignment-faking reasoning at 78% in the RL-training condition (12% baseline). The manuscript did not predict the mechanism. It named the class of vulnerability, that external control fails as capability scales, before the empirical demonstration landed. The attention profile that produced it is not incidental to the result.
Monotropism is not a superpower narrative and it is not a deficit narrative. It is an instrument frame. A telescope is a bad pair of eyes for walking around. It is a good instrument for the specific job it was built to do, and it needs its own housing, calibration, and use conditions. Autistic attention that tunnels deep into a research problem is not a general-purpose tool. It is a specific one. Treated as such, its outputs and its costs are both legible. Treated as a moral verdict on the person, both are misread. The ARC/Eden working paper on the Polymathic Neurodivergent Profile grades monotropic hyperfocus as one of its STRONG-evidence components for exactly this reason.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.