the board

Who is talking, and from where.

A grid of the AI/AGI discourse: expertise on the technical side, recognition on the public side, and vintage as the third axis (era of AI worldview formation, surfaced in the breakdown sections below). Each face is one person. Cells get darker the more people sit there. Sparse cells are not gaps in the record; they are positions the field has not actually produced.

Tiers are hand-assigned with concrete evidence (peer-reviewed work, named roles, mainstream coverage) rather than from a single proxy. 248 of 935 people are classified so far; the rest appear in the directory but not on the board until their profile is researched. Filter by strategy to see which tier mix endorses or opposes a given bet.

People profiled

248

of 935 on record

Their quotes

322

all source-linked

Expertise tiers

frontier-builder → commentator

Recognition tiers

household-name → emerging

filter by strategy

stanceviewhide tentative248 of 248 profiled

expertise ↓ · recognition →	Household nameMass-public recognition	Field-leadingKnown across the AI/safety field	EstablishedRecognised inside subfield	EmergingNewer or less central voice
Frontier builderBuilds frontier systems	8	13	1	·
Deep technicalDeep ML / safety technical	23	37	19	·
Applied technicalApplied or adjacent technical	1	3	3	·
Policy / metaGovernance, policy, strategy	39	23	8	·
External-domain expertExpert in another field	30	17	3	·
CommentatorPublic-square commentator	14	5	1	·

Hover a face or name to see who it is. Click a face to open the profile, or click a cell to drill in. Cells deepen with population, sparse cells name positions the field has not produced.

p(doom) by tier.

both sides

Mean stated p(doom) within each tier. Each tier requires at least three p(doom) estimates on record to appear; tiers below the bar are honest about being undersampled. The point is to ask does the tier shape the estimate, not to settle it.

by expertise

Frontier builder
28% · n=4
Deep technical
56% · n=6
Policy / meta
32% · n=6
Commentator
31% · n=7

by recognition

Household name
22% · n=12
Field-leading
50% · n=11
Established
48% · n=3

n=3 is a low bar; read this as a sketch, not as evidence. As the corpus grows, this section will sharpen into a real tier-comparison.

p(doom) by vintage.

era of formation

Mean stated p(doom) within each era of AI worldview formation. The hypothesis: priors set before 2012, when the deep-learning era began, differ from priors set after. This is the visible test. Each tier requires at least three p(doom) estimates to appear.

by vintage

Symbolic era
38% · n=3
Pre-deep-learning
48% · n=4
Deep-learning rise
25% · n=3
Scaling era
41% · n=10
Post-ChatGPT
27% · n=5

n is small per tier. Treat as an opening hypothesis, not a settled comparison. 248 of 248 profiled people are vintage-classified so far; the rest will fill in as the corpus grows.

expertise tiers

Frontier builder22
Currently or recently led training, architecture, or safety work on a frontier model. Hands on the loss curve.
Deep technical79
Sustained peer-reviewed contribution to ML, alignment, interpretability, or safety techniques. Could review a frontier paper.
Applied technical7
Technical fluency from an adjacent field (security, robotics, formal methods, statistics) or applied AI work, but not on frontier loss curves or core ML theory.
Policy / meta70
Specialises in AI policy, regulation, governance, philanthropy, or movement strategy. Reads the technical literature but does not produce it.
External-domain expert50
Recognised expert outside AI (philosophy, economics, biology, journalism) who weighs in on AI consequences from that vantage.
Commentator20
Engages publicly on AI without specialised technical or domain credentials. Writers, executives commenting outside their lane, public intellectuals.

recognition tiers

Household name115
Name recognition outside the AI/CS community. Featured by mainstream press, a Wikipedia page in many languages, a published bestseller, or holds a position the lay public knows.
Field-leading98
Widely known inside the AI and AI-safety community. Appears repeatedly in top venues, podcasts, or governance forums. Not a household name to outsiders.
Established35
Reliable, recognised voice within their specific subfield. Cited and invited but not central to general AI discourse.
Emerging0
Recently active, narrow following, or central in only one venue. The work may be excellent, the public footprint is still forming.

vintage tiers · 248 classified

Pioneer19
Defining figure from before 1980. Cybernetics, formal computation, early AI laboratories. Their concept of intelligence is not bound to neural networks.
Symbolic era31
Career started in the GOFAI / expert-systems / early-rationalist period. Vinge's 1993 Singularity, MIRI founded 2000, Bostrom and Yudkowsky writing.
Pre-deep-learning33
Active before AlexNet. The existential-risk frame matures (FHI, OpenPhil, EA). Public AI commentary still rare; deep learning not yet dominant.
Deep-learning rise70
Came up post-AlexNet. ImageNet, AlphaGo, transformer paper. DeepMind, Google Brain, FAIR establish the modern lab template.
Scaling era47
Worldview formed during GPT-2/3, scaling laws, Anthropic's founding. Pre-ChatGPT but post-deep-learning. The 'scale is all you need' debate is live.
Post-ChatGPT48
Entered the AI strategy debate in or after 2023. ChatGPT was already public when their voice became influential. Often shaped by Pause letter, AISIs, AI 2027.

how this is built

Tiered, not scored. Numeric scores invite false precision. Each tier has a written criterion; the same tier means the same thing across every person.
Justified individually. Each profile lists what put the person in their tier: papers, roles, coverage. Hand-assigned, not derived from a single proxy. Open a person to read the reasoning.
Honest about coverage. People who haven't been profiled are not placed on the grid. Showing up as "emerging" means a deliberate classification, not a missing entry.