THE HUMAN IDEAS ARCHIVE
The AI Ideology Atlas
A field guide to what people believe about AI, why they believe it, and where the argument gets complicated. No single axis. No personality scores.
Progress & pace
Defensive accelerationism
d/acc is a self-described approach to accelerating technologies that strengthen defense and distribute power. Vitalik Buterin’s formulation links defensive, democratic and decentralized development. It is broader than AI and does not mean either unconditional acceleration or a blanket refusal to build.
MovementEffective accelerationism
Usually written e/acc. A self-described movement favoring rapid technological development, competitive experimentation and the expansion of intelligence. Its early manifesto draws on thermodynamics and markets; those arguments are advocacy, not a scientific proof of the optimal pace of AI deployment. Participants do not share one detailed policy platform.
Policy positionPausing frontier AI
A family of proposals to temporarily restrict training or deployment of especially capable general-purpose systems while safeguards and governance develop. PauseAI is an advocacy organization with its own proposal; it is not a synonym for every person worried about AI, nor a call to ban every present-day AI application.
ConceptTechnological optimism
The broad expectation that technological progress can improve human life. It encompasses conditional, institution-conscious optimism as well as stronger faith in markets or invention. It is not identical to e/acc, and it does not require endorsing every product labeled AI.
Capabilities & futures
AGI timelines
Forecasts about when systems might reach a specified level of broad cognitive capability. The date is meaningful only alongside a definition, evaluation method and uncertainty range. A forecast of benchmark performance is not automatically a forecast of cheap, reliable replacement of an occupation.
ConceptThe AI bubble thesis
The argument that some AI-related valuations or investment expectations exceed plausible future returns. This is an economic claim, not proof that the technology is useless. Useful infrastructure and unsuccessful investments can coexist; an overvalued firm can sell a genuinely helpful product.
Risk & safety
AI alignment
Research and engineering concerned with making system behavior match intended goals and constraints. The term covers different technical problems and different ideas about whose intentions matter. A model following a user’s request is not automatically behaving in a socially acceptable way.
Research fieldExistential risk from AI
The study and debate of scenarios in which AI contributes to human extinction or a permanent, severe loss of humanity’s future potential. It is not a claim that today’s chatbot is secretly conscious. Proposed mechanisms, probabilities and appropriate responses are contested.
Research fieldPresent-day AI harms
An umbrella for observable harms associated with current systems: discriminatory decisions, privacy failures, fraud, unsafe outputs, exploitation and unequal burdens. These are different problems, not one mechanism. Attribution requires examining the specific system and the institutions deploying it.
Power & governance
AI sovereignty
The aim of retaining meaningful control over AI infrastructure, data, capabilities and critical dependencies. Governments, regions and institutions use the term differently. It can mean resilient procurement and local capacity, or a more expansive program of national technological autonomy.
Policy positionDemocratic AI governance
Approaches giving affected people a meaningful role in decisions about AI through accountable institutions, participation and contestable rules. This can involve legislatures, public consultation, worker representation or other mechanisms. It is not equivalent to taking an online vote on every technical change.
Policy positionPublic compute
Publicly supported access to computational infrastructure for research and other public-interest uses. It can include shared facilities, subsidized access and associated datasets or training. A national research resource is one institutional example, not the only possible model.
Openness & ownership
Closed models
Systems whose weights or other important development materials are not publicly released. Access is often mediated by a hosted product or API. Closed development is a distribution and governance choice, not by itself proof of superior safety, inferior quality or fraudulent intent.
ConceptOpen-source AI
An approach to AI development emphasizing freedoms to use, study, modify and share a system. The Open Source Initiative’s AI definition specifies what must be available to exercise those freedoms. An accessible download alone does not establish that a release qualifies.
ConceptOpen-weight models
Models whose learned parameters are made available for others to obtain, subject to the release’s license. This can allow local inference and modification. It does not necessarily include training data, training code or an unrestricted license, so open weights and open source should not be used interchangeably.
Work & distribution
Human augmentation
Using AI to extend what people can do while retaining meaningful human participation. Examples include drafting assistance, translation, analysis and accessibility tools. Augmentation describes a relationship within a workflow; it is not a guarantee that employment, autonomy or bargaining power will improve.
ConceptAutomation and displacement
The substitution of machines for tasks previously performed by people, and the resulting changes in work. Task exposure, technical automability and actual job loss are different measures. Deployment decisions depend on cost, reliability, organization and demand, not capability alone.
Art & culture
AI and art
A broad category of practices using AI in making, selecting, modifying or presenting artistic work. It includes different levels of human control and different relationships to existing work. “AI art” names a production feature, not a single style, quality level or ethical verdict.
Policy positionCreator consent
The position that creators should have meaningful control over specified uses of their work in AI development or distribution. Proposals differ: opt-in licensing, opt-outs, collective bargaining, attribution or payment. Ethical consent and legal permission are related but distinct questions.
Human & machine
AI moral status
The question of whether an artificial system could have interests or experiences that deserve moral consideration. It is distinct from intelligence, legal personhood and whether users form attachments. Fluent claims about feelings do not independently establish that feelings are present.
ConceptCognitive dependence
Reliance on a system for remembering, evaluating or deciding, especially when independent checking and skill maintenance weaken. Not all cognitive offloading is harmful: writing and calculators are also external aids. The concern is whether reliance preserves or erodes the ability to judge the output.
MovementTranshumanism
A family of ideas favoring the use of technology to expand human capabilities and overcome some biological limitations. It extends beyond AI to health, longevity and enhancement. The Humanity+ declaration also recognizes risks and individual choice; transhumanism is not simply “replace humans with machines.”
Planet & resources
Language & tribes
“Doomer” and “decel”
Loose internet labels often applied to people seen as pessimistic about AI or as wanting to slow development. They are frequently hostile descriptions rather than the person’s own identity. Their use in a debate reveals the speaker’s framing, not a reliable summary of the target’s beliefs.
Our editorial stanceEvidence-first centrism
This publication’s editorial stance: apply the same evidential standards to optimistic and pessimistic claims, including our own. “Center” describes independence from the two rhetorical camps, not a rule that the true answer must lie halfway between them. It is not a recognized scientific school.
No doctrine found. A rare opportunity to think for yourself.