NeonThe research behind Neon
Neon draws on research into autobiographical memory, reflection, metacognition and long-term AI recall. The useful question is how each finding changes what we build and what we test.
These studies did not test Neon. The design choices below are our inferences from the research. They do not establish health benefits, reliable consciousness measurement or product efficacy. The researchers did not participate in or endorse the app.
McLean and Pasupathi studied how personal meanings developed through autobiographical conversations. This informs Neon's separation of original words and later understanding: a new account can sit beside the first.
The study concerned human conversation. It does not establish that AI reflection has the same effect.
McLean & Pasupathi · Journal of Personality, 2011Finley and colleagues studied memory after reviewing images captured with SenseCam. This points us toward testing recovery and review, rather than treating the volume recorded as the result.
A small image-review study cannot prove that wrist audio or Neon improves memory.
Finley, Brewer & Benjamin · Memory, 2011 · PDFGrossmann and colleagues examined a third-person diary method in two longitudinal experiments. It informs voluntary prompts that invite another view while preserving the original account.
This does not qualify an AI therapist, a universal reflection technique or a benefit from Neon.
Grossmann et al. · Psychological Science, 2021Gilbert's experiments examined external reminders and intention offloading. This informs our distinction between accepting a reminder, delivering it and fulfilling the intention.
A lab task does not establish that reminders improve someone's everyday life.
Gilbert · Quarterly Journal of Experimental Psychology, 2015 · PDFFleming and Lau distinguish different measures of metacognition. This informs inspectable sources and room for uncertainty: the confidence of a statement is not evidence that it is correct.
Free speech does not yield a validated metacognition or consciousness score.
Fleming & Lau · Frontiers in Human Neuroscience, 2014Chan and colleagues tested deliberately misleading conversational interviews in a simulated witness setting. Pataranutaporn and colleagues also studied misleading conversational information. These findings motivate neutral prompts, direct access to originals and explicit labels for AI interpretation.
These tasks are not estimates of risk in ordinary journaling. Adding citations alone does not establish protection.
Chan et al. · Conversational AI and false memories · 2024 preprintLoCoMo and LongMemEval evaluate long conversational histories. They inform tests for changes over time, relevant source recovery, updated facts and questions that cannot be answered from the record.
These benchmarks differ from real Watch transcripts. We do not claim a Neon benchmark score or that a large context window makes autobiographical recall reliable.
Maharana et al. · LoCoMo · ACL 2024Roediger and Karpicke studied retrieval practice and delayed retention. This informs an optional flow in which you try to recall a selected factual note before revealing its original words.
Learning prose is different from interpreting an experience. Neon does not turn personal accounts into scored lessons.
Roediger & Karpicke · Psychological Science, 2006 · PDFThe Cogitate Consortium's adversarial study tested predictions from two consciousness theories using neural and visual experiments. It reinforces the need to keep subjective reports, theoretical interpretations and empirical measurements distinct.
Neon preserves chosen accounts of experience. A transcript cannot establish the preservation or transfer of consciousness, reveal unspoken thoughts or constitute a complete identity.
Cogitate Consortium et al. · Nature, 2025Keep the original available. Let a person correct it. Distinguish a later interpretation. Ask whether a source still exists and is still permitted. Leave a proposal unaccepted until the person chooses it.
Our validation work separates source and policy checks, simulator views, native builds, physical-device operation and live-service trials. The development preview establishes none of the missing device or service results by itself.
Explore Inside NeonSelected primary sources checked on 3 October 2026. Access varied across full manuscripts, introductions and publisher abstracts. This is a bounded design bibliography, rather than a systematic or clinical review.