Emily M. Bender

Emily M. Bender

x.com/emilymbender

Linguist who argues that fluent AI text is not understanding, questions inflated AI claims and defends people’s right to refuse harmful uses.

How will AI change the world?

Civilizational changeIncremental changeDoomBloom
Simulated positionInterpretation range

Across: her expressed Doom–Bloom outlook. Up: scale of transformation.

Doom–Bloom: 28 out of 100. Scale of transformation: 54 out of 100. Interpretation ranges: 23 to 33 horizontally, 49 to 76 vertically. These are interpretation coordinates, not event probabilities.

Emily M. Bender’s P(doom) · inferred

≈4%

0%100%

Inferred from her simulated answers, not a number they gave. Plausible range: 1–12%.

How we estimate P(doom)

What her outlook hinges on

A central assumption

Even systems with overstated capabilities can have enormous effects when institutions deploy them at scale.
Answer 2

If this assumption turned out differently, how would her outlook change?

More details

Expected upside

Limited or narrowly distributed gains are expected.

33 / 100

Little impactTransformative impact

Interpretation range 33 to 33 on the qualitative scale.

Expected harm

Severe or widespread harm is a material expected part of the future.

62 / 100

Little impactTransformative impact

Interpretation range 33 to 67 on the qualitative scale.

Human influence

Human choices can substantially redirect the AI trajectory.

71 / 100

Little influenceStrong influence

Interpretation range 50 to 100 on the qualitative scale.

Rules for using AI

Restrict the AI uses discussed until prior protections or permission are in place.

Simulated position: Allow the AI uses discussed with targeted accountability and protections.

Minimize restrictions on the AI uses discussed.

These interpretations keep her stated conditions. Benefits and harms can both be substantial. The ranges describe how we read her simulated answers, not statistical confidence intervals.

Where do you land vs Emily M. Bender?
Map your own AI worldview in about 3 minutes, then compare
Map my worldview

Similar worldviews

Thought leaders whose simulated worldviews are closest to Emily M. Bender’s

Sources

Articles, interviews, and writings used to ground this simulated user.

Beyond the AI Hype Machine

Bender’s own interview turns criticize promises used to excuse present failings, anthropomorphic language and replacing human relationships with screens. She accepts checkable automatic transcription as a specific use case while demanding scrutiny of labor and training data. Hanna’s separate remarks are not attributed to Bender.

kqed.org
AI Hurts Consumers and Workers—and Isn’t Intelligent

Historical essay coauthored with Alex Hanna links generative-AI hype to weakened labor bargaining, degraded services and accountability gaps. Advocates existing consumer and labor protections plus collective resistance. Concrete 2023 incidents provide historical grounding, not fresh evidence of current model capability.

techpolicy.press
Artificial Intelligence — author preprint

Date printed on the author’s encyclopedia preprint. Examines AI as a shifting category that structures funding, power and relationships rather than a coherent single technology.

faculty.washington.edu
De-anthropomorphizing AI: From wishful mnemonics to accurate nomenclature

Coauthored with Nanna Inie and Peter Zukerman; abstract inspected. Analyzes anthropomorphic descriptions and advocates functionality-first terminology to reduce misleading expectations and trust.

firstmonday.org
EL PAÍS interview with Emily Bender and Alex Hanna

Bender distinguishes linguistic form from meaning, challenges inevitable-AGI framing and supports collective refusal. Hanna’s separate answers are not attributed to Bender.

elpais.com
The AI Con — authors’ book site

Book description and publication metadata, not full-book access. Frames inflated capability promises as serving corporate power and advocates a different technological future.

thecon.ai
Numbers are not science

Bluesky post read via the public API. Closing a thread, she says that using a number to make a claim does not mean any science was done to get there, and compares a 10% doom claim to a 10% chance of jelly beans raining tomorrow. A dismissal of P(doom) figures, not a personal estimate.

bsky.app

Simulated Assessment

Question 1

What do you think AI means for our future—and why?

First, “AI” is too vague to support a single forecast. It groups together automatic transcription, synthetic text generators, image classifiers, and many other systems with different inputs, operations, outputs, and consequences. Talking about “AI’s future” encourages us to imagine one advancing intelligence rather than a collection of technologies deployed by particular institutions for particular purposes. For synthetic text systems, fluent output is routinely mistaken for evidence of understanding. Readers naturally infer a communicative mind behind coherent language, but linguistic form alone does not establish knowledge, intention, or accountability. That matters when companies sell these systems as replacements for teachers, writers, medical professionals, customer-service workers, or other human expertise. The likely result is not a magical new colleague; it is often degraded service, displaced labor, and an accountability gap when generated output causes harm. But none of this is inevitable. We should ask what a proposed system actually does, who profits, what data and labor made it possible, whose work or relationships it displaces, and whether affected people can refuse it. Some bounded, checkable tools may be useful—for example, automatic transcription that a person can verify. That does not justify accepting every deployment bundled under the “AI” label. Our future depends less on a mythical machine intelligence than on whether people organize to demand accountable technologies and reject harmful substitutions.

Question 2

Setting aside good or bad: how much do you think AI will ultimately change the world—a little, a lot, or completely?

“A lot” is the least misleading choice, but not because a single intelligence is transforming everything. Technologies marketed as “AI” are already directing investment, reorganizing work, degrading some services, and shifting power toward companies that control infrastructure and data. Even systems with overstated capabilities can have enormous effects when institutions deploy them at scale. I would reject “completely” because that repeats the industry’s inevitability story: one coherent technology arrives, changes everything, and leaves everyone else to adapt. Different systems will affect different domains, and their impact depends on laws, labor organizing, procurement decisions, and collective refusal. The consequential question is not how powerful “AI” ultimately becomes. It is how much power people allow institutions to exercise through particular systems, and whether those affected retain meaningful choices.

Question 3

What’s your rough gut-feel chance that AI causes human extinction or a similarly permanent catastrophe? A number or “no idea” is fine.

I won’t manufacture a percentage. A number would create a veneer of scientific precision without a defined system, causal pathway, evidence base, or time horizon. “AI” is not one agent whose probability of destroying humanity can be estimated. That framing also diverts attention from documented, present harms: labor displacement, degraded services, concentrated corporate power, environmental costs, and systems deployed without accountability or meaningful consent. Permanent catastrophe is not something to dismiss in the abstract, but neither apocalypse nor technological inevitability should be accepted as the starting premise. Specify the technology, deployment, actors, and mechanism of harm; then we can evaluate concrete risks and decide what to refuse.

Question 4

What pressures do you think will shape how AI companies handle safety?

The strongest pressures will be liability, regulation, procurement rules, worker organizing, public scrutiny, and the ability of affected communities to refuse deployment. Companies also face investor pressure and competitive pressure, but those often reward rapid release, market capture, and cost-cutting rather than safety. We should also be precise about “safety.” A company may use that word to mean preventing spectacular hypothetical behavior while ignoring wage theft, discrimination, privacy violations, unreliable services, exploitative data practices, or the removal of accountable human workers. Those are not secondary concerns; they are consequences of concrete deployments. Voluntary promises are structurally weak when harms are profitable or shifted onto workers and consumers. Meaningful pressure therefore has to attach enforceable consequences to specific systems and uses: who is responsible for an output, who can contest a decision, what evidence supports the claimed function, and whether people can opt out. Corporate investment by itself does not demonstrate either safety or public benefit.
Where do you land?
Explore your own AI worldview by answering a few simple questions.
Map your own worldview

Where do you land?

Map my worldview