AI doomsday hype obscures urgent ethical failures, expert warns

Alarmist predictions about artificial intelligence wiping out humanity are diverting attention from tangible harms already unfolding across workplaces, courts and online platforms, according to a senior academic who says the fixation on speculative futures is allowing powerful technology firms to sidestep accountability for present-day abuses.

Speaking amid intensifying debates over AI governance, Professor Tobias Osborne argues that the dominance of “doomsday” narratives has distorted public understanding of risk. He contends that attention is being drawn away from labour exploitation in data supply chains, large-scale copyright appropriation and algorithmic bias that affects hiring, credit decisions and access to public services.

The warning comes as governments in Europe, North America and parts of Asia push ahead with regulatory frameworks aimed at managing AI’s rapid commercial deployment. While existential risks are not dismissed, Osborne maintains that focusing policy debate on far-off scenarios weakens scrutiny of harms that can be measured and mitigated now.

At the centre of his critique is the way dramatic rhetoric shapes priorities. Claims that advanced AI could soon surpass human intelligence and threaten civilisation dominate conferences, investment pitches and media coverage. According to Osborne, this framing benefits technology companies by positioning them as guardians against catastrophe, while obscuring their role in practices that raise immediate ethical and legal questions.

One such practice is the treatment of workers who underpin AI systems. Large language models and image generators depend on vast volumes of labelled data, much of it produced by outsourced contractors paid a fraction of wages in developed economies. These workers often review violent or explicit material to train content filters, a process that researchers have linked to psychological harm. Osborne argues that debates about hypothetical super-intelligence rarely acknowledge these human costs.

Copyright disputes form another fault line. Publishers, artists and software developers have accused AI firms of training models on protected material without permission or compensation. Courts in several jurisdictions are considering whether existing intellectual property law applies to machine learning at scale. Osborne says that by steering public discussion towards abstract future threats, companies blunt pressure for clearer rules and fairer licensing arrangements today.

Bias embedded in algorithms presents a further concern. Studies across sectors have shown that automated systems can replicate and amplify discrimination present in historical data. Hiring tools have been criticised for disadvantaging women, while facial recognition systems have shown higher error rates for people with darker skin tones. Osborne notes that these outcomes have real consequences for livelihoods and civil liberties, yet struggle to compete for attention with apocalyptic forecasts.

The academic also questions the assumption that existential risk narratives are neutral or purely precautionary. He suggests they can act as a strategic shield, enabling firms to call for voluntary self-regulation while arguing that only they possess the expertise to manage future dangers. This, he says, weakens democratic oversight and delays binding standards.

Regulators are beginning to grapple with these tensions. The European Union’s AI Act categorises systems by risk level and imposes stricter obligations on applications that affect fundamental rights. Elsewhere, lawmakers are examining transparency requirements, audit obligations and liability rules. Osborne welcomes these moves but warns that enforcement will falter if political energy is consumed by speculative fears rather than documented harm.

Industry responses have been mixed. Some companies emphasise safety research and long-term risk mitigation, pledging investment in alignment and control mechanisms. Others have begun negotiating licensing deals with content owners or publishing summaries of training data sources. Critics, including Osborne, say such steps remain uneven and insufficient without clear legal mandates.

Civil society groups echo the call for a recalibration of debate. Labour advocates want minimum standards for data workers, including mental health protections and fair pay. Creators seek enforceable consent and remuneration frameworks. Rights organisations argue for mandatory bias testing and avenues for redress when automated decisions cause harm.