Near-term harms first.

Argues AI's effects on knowledge work depend on whether organizations adopt it as a substitute for human judgment or as a scaffolding for it; treats this as a managerial choice, not a technological inevitability.

Argues 'fairness' framings underweight the concept of violence done through information systems; calls for moral language adequate to the structural harms produced by data-driven systems.

Argues AI's impact on the labor share of income is the central macroeconomic concern, and that policy needs to start preparing for transformative scenarios where labor becomes a smaller fraction of value.

Argues AI's labour effects are following a historical pattern that the original Luddites resisted intelligently; calls for organized labor responses rather than passive acceptance.

Argues universities have no plan for AI's effect on student writing and assessment; current institutional inertia means policy will be set by individual instructors making contradictory choices.

Argues AI literacy must reach K–12 to give the next generation agency over AI rather than leaving design choices solely to labs and platforms.

Argues we are heading toward 'technological unemployment', not because AI will eliminate all jobs, but because it will reduce demand for human labor across enough domains that distribution becomes the central political question.

Argues educational and journalistic systems urgently need AI-text detection to maintain trust; founded GPTZero specifically as a response to ChatGPT's release.

Argues AI's most damaging effects are mediated through advertising and engagement business models; safety must address those incentive structures, not just model capabilities.

Argues independent third-party audits, not lab self-attestation, are how AI systems should be regulated; pioneered audit methodologies for government and platform AI.

Argues AI deployment is fundamentally a cultural and institutional question that current technical framings miss; calls for cyber-physical-social systems thinking as an alternative.

Argues algorithmic decision-making in finance, criminal justice, and healthcare is the high-stakes ground for AI ethics; pushes back on framings that treat the question as primarily about future AGI.

Argues artists should hold property and consent rights over AI training; co-built Spawning's 'Have I Been Trained?' tool to make this possible at scale.

Argues workplace algorithmic management is a primary surface for AI harm and that labor law and civil-rights law should govern it; AGI debates distract from where harm is happening.

Argues system-level audits, rather than dataset audits or model cards alone, are required to capture real-world harms; supports legal frameworks that mandate access to deployed models.

Argues AI hype outpaces capability and that practical reporting on what models actually do, and fail to do, is more useful than amplifying lab claims.

Argues both x-risk and tech-criticism camps under-weight the social and political effects of AI on knowledge work and the texture of public conversation.

Argued in 1995 that automation would dramatically reduce demand for human labor; updates this thesis to argue AI is the long-delayed continuation of that trajectory.

Argues recommender-system harms are a primary AI safety question; deployed engagement-maximizing recommenders may already be the highest-impact AI in the world.

Argues algorithmic systems already cause large-scale harm in criminal justice, hiring, and credit; calls for adversarial journalism and audit infrastructure rather than corporate self-policing.

Argues the human cost of AI's data and moderation pipelines is invisible by design and that surfacing it is essential to honest debate about AI governance.

Argues the AI policy conversation should center documented sociotechnical harms, labor, copyright, surveillance, hallucination harms, rather than speculative existential scenarios.

Argues AI's labor and bias harms are real and disproportionately affect Global Majority populations; pushes back on narratives that center only existential risk.

Argues AI failures are structural rather than glitches; calls for regulatory enforcement of audit requirements over voluntary lab self-governance.

Argues the most consequential AI risks may be social and epistemological rather than existential; warns against AI-mediated 'editing of reality' that destabilizes shared truth.

Argues LLMs systematically violate users' contextual privacy norms in ways that current alignment work does not address; privacy is a first-order alignment problem.

Argues cognitive liberty, the right to mental privacy and self-determination over one's own thoughts, must be enshrined as a foundational right in the AI/neurotech era.

Argues mainstream coverage of AI accepts industry framing too readily; pushes back on both x-risk and corporate techno-optimism in favour of focus on labour, environment, and material costs.

Argues 'AI safety' framings collapse when separated from the data extraction and labour systems that produce models; advocates regulating those substrates rather than the models alone.

Argues every technology embeds choices about whose interests it serves; AI systems are not neutral and must be evaluated against the political economy of their deployment.

Brings a historian's perspective: argues each new wave of computing has been heralded as transformative, and that AI is similar, real and overstated at the same time.

Argues ML safety must integrate ecological, economic, and labor harms, not only model misalignment, into a single research agenda.

Argues the field over-weights existential risk relative to documented harms from deployed systems and that solid engineering for robustness, validation, and oversight does most of the safety work.

Argues computer vision systems encode racialized and political assumptions through their training data, and that revealing these assumptions through art is part of the political work.

Argues AI hype obscures concentrated power and labor disruption; calls for global coordination and public-interest research investment as the alternative to lab-led regulation.

Editorial work consistently treats AI hype claims with skepticism while taking the underlying technical advances seriously; frames safety reporting around concrete failure modes rather than meta-debates.