STRATEGYOngoing practice91% confidence

AGI as Shapeshifting Narrative

AGI has no fixed definition — ambiguity is the feature, not the bug

Problem it solves

Explains how AI companies simultaneously invoke existential doom and consumer abundance without cognitive dissonance

Best for

Evaluating the sincerity of AI safety rhetoric, identifying when capability and risk narratives are instrumental rather than principled

Not ideal for

Timing market moves around specific capability announcements — this is a structural lens, not a catalyst detector

Overview

Why this framework exists

Hao documents that 'artificial general intelligence' has no stable scientific definition — the field was named in 1956 specifically by choosing a term whose endpoint (recreating human intelligence) has no consensus scientific meaning. This definitional ambiguity is not a bug awaiting resolution; it is a structural feature that allows AI companies to simultaneously invoke catastrophic risk (justifying capital accumulation, locking out competitors, and demanding regulatory deference) and imminent consumer benefit (driving adoption). The same company can say AGI represents existential danger to humanity and also that their AGI assistant will cure cancer and solve climate change — without any logical inconsistency, because the term means something different in each sentence.

Sam Altman's deployment of four mutually inconsistent AGI definitions for four different audiences is the canonical example: existential risk for Congress (justifying no regulation that would create barriers to US competitiveness), consumer utopia for users (driving adoption), revenue machine for Microsoft (unlocking investment), and autonomous agent for OpenAI website (establishing technical legitimacy). Each definition mobilizes the target audience toward a desired behavior.

The deeper mechanism, which Hao illustrates with the Dune analogy, is that constant embodiment of the myth eventually dissolves the distinction between sincere belief and instrumental performance. Executives who begin by strategically invoking existential risk lose the ability to distinguish their genuine beliefs from their public positioning — making the mythology self-reinforcing.

Core principles

5 total
  1. AGI has no scientific consensus definition — any company invoking it is filling a blank canvas with whatever their target audience needs to hear.
  2. Existential risk rhetoric and consumer benefit rhetoric serve identical strategic functions when deployed by the same company to different audiences.
  3. The definitional moving goalpost protects against accountability — you can never fail to achieve AGI if you can redefine it after each milestone.
  4. Constant myth embodiment dissolves the line between strategic performance and genuine belief — the Dune effect.
  5. Matching the language of your target audience's existing concerns is more effective than making original claims — Altman mirrored Musk's existential risk framing precisely.

Steps

5 steps
  1. Collect all definitions from the same source
    Gather every public statement from the executive or company about AGI, ASI, or general intelligence across different venues — congressional testimony, investor calls, consumer marketing, research publications, and internal documents if available. Do not average them or assume the most recent supersedes earlier ones.
    Pro tipThe audience is the key variable — sort statements by who heard them, not when they were made.
  2. Map each definition to its audience's required behavior
    For each statement, identify: who was the audience, what action did the speaker need from them (provide capital, withhold regulation, adopt product, provide data, join company), and how does the definition of AGI in that statement make that action more likely? This reveals whether definitions are sincere technical claims or mobilization instruments.
    WarningStatements that appear contradictory when compared directly often become coherent when analyzed by audience requirement — this is a sign of sophisticated narrative management, not confusion.
  3. Identify the null accountability structure
    Ask: what would it look like for this company to fail to achieve AGI in a way that anyone could verify? If no such failure mode exists (because AGI can always be redefined), the term is functioning as an unfalsifiable narrative device rather than a technical milestone.
    Pro tipThe goalpost shift is the tell. When a company achieves what they previously called AGI and then moves the definition forward, the narrative function of the term is confirmed.
  4. Apply the Dune test for sincere belief
    Assess whether executives can distinguish their genuine technical beliefs from their public narrative by looking for inconsistencies in private versus public communications. When public and private language align perfectly with external audience needs rather than internal evidence, the myth has consumed the operator. Dario Amodei's 10-25% catastrophic risk estimate is offered in the same breath as aggressive capability development — test whether these can coexist as sincere beliefs.
    Pro tipHao's framing: 'They are purposely trying to create this feeling within the public that they are summoning the demon because it is a crucial part of their power.'
    WarningDo not assume cynicism — the Dune mechanism is that genuine belief and strategic narrative can become indistinguishable even to the speaker.
  5. Track the evil empire rotation
    Identify the current 'evil empire' competitor being invoked to justify the company's own accumulation. Note when the previous evil empire became cooperative or irrelevant and a new one was named. The rotation confirms the narrative function — if the evil empire were a sincere factual claim, it would not need replacement.
    Pro tipThe sequence for OpenAI: first Google (early period), then China (from approximately 2019 onward). Each rotation coincided with a new capital raise or regulatory threat.

Checklist

Saved in your browser

Examples

2 cases
Altman's four AGI definitions

Sam Altman deployed four distinct and mutually inconsistent definitions of AGI across four audiences: to Congress, AGI is the system that will cure cancer, solve climate change, and end poverty (mobilizing against restrictive regulation); to consumers, it is the most amazing digital assistant you will ever have (driving adoption); to Microsoft, it is a system that will generate hundreds of billions in revenue (unlocking investment); and on the OpenAI website, it is defined as highly autonomous systems that outperform humans in most economically valuable work (establishing technical credibility with the research community).

OutcomeEach audience received the definition that maximized their cooperation. The inconsistency was never challenged in any single forum because no single audience heard all four definitions simultaneously.
The 2015 existential risk letter

In 2015, Sam Altman privately wrote that 'development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity.' Hao notes that the language precisely mirrors what Elon Musk was publicly saying at the time — Musk was focused on existential AI risk, and Altman needed to recruit him as a co-founder. The letter was not a general-audience document but a targeted mobilization instrument written in the target's existing language.

OutcomeMusk joined as a co-founder and brought significant capital and credibility. When Musk later sued over the nonprofit-to-for-profit transition, the same letter became evidence of Altman's instrumental use of existential risk framing.

Common mistakes

3 traps
Evaluating capability claims against a single definition
When you assess 'has OpenAI achieved AGI?' you are implicitly adopting one of their four audience-specific definitions. Any answer you reach is relative to whichever definition you chose. The correct analysis is: which definition is operationally active for the decision you are making (investment, regulation, adoption) and what evidence applies to that specific definition?
Treating private communications as more sincere than public ones
The Dune mechanism means private communications may be as mythologized as public ones. The 2015 Altman letter was private but written specifically to mirror Musk's language — private audience, instrumental construction. Private sincerity cannot be assumed.
Assuming the definitional ambiguity will be resolved
Companies benefit from the ambiguity directly — any resolution would eliminate the narrative flexibility that allows simultaneous doom and abundance claims. There is no internal incentive to resolve the definition, and external pressure (regulatory definitions) is actively lobbied against.

Origin story

How this framework came to be

The definitional instability dates to 1956 when John McCarthy coined 'artificial intelligence' over colleagues' objections that it pegged the field to recreating human intelligence — an undefined target. Hao traces Altman's specific manipulation of this ambiguity from 2015 through the November 2023 firing, drawing on internal emails disclosed in the Musk/Altman lawsuit. The private 2015 Altman letter warning that superhuman machine intelligence is 'probably the greatest threat to the continued existence of humanity' was written specifically to recruit Musk, mirroring Musk's own language — suggesting instrumental construction rather than sincere conviction from the outset.

Source

Traced to primary
Source · PODCAST
AI Whistleblower: We Are Being Gaslit By AI Companies, They're Hiding The Truth!
Karen Hao · 2025
Open source →

Related frameworks

Browse all Strategy →