Experts Say Divergent Definitions Stall Global AI Governance

Newser reports an opinion piece by Sarosh Nagar and David Eaves, affiliated with University College London, arguing that divergent definitions of AI and differing timelines for its impact are a primary reason international talks on AI governance keep stalling. The authors write that some actors frame AI as `ChatGPT`-style systems, others as potential superintelligence, and others as routine algorithms, and they quote the view that "governance is impossible" until those definitions converge. Newser also reports the op-ed says governments predicting rapid, total overhauls are likely to align with leaders, like the United States and China, to secure access. Governments confident in slower impacts in certain sectors, meanwhile, may sit back and build their own domestic systems, and the power players dominating computing power and top models have little incentive to hand authority to global bodies.

Fragmented definitions create scope problems for regulation, because a rule that fits one interpretation of "AI" may be irrelevant under another. Industry-pattern observations show that mismatched risk time horizons, from immediate safety and robustness to long-term existential risk, produce divergent prioritization among regulators and stakeholders.

For practitioners, this means international regulatory uncertainty will likely persist until standard-setting bodies, technical communities, or coalitions of states converge on shared terminology and risk taxonomies. Concentration of compute and model development also shapes bargaining leverage, an industry-wide structural factor separate from any single actor's intent.

Indicators include efforts to publish shared taxonomies or definitions by technical bodies, bilateral agreements among major computing powers, and dominant model providers participating in multistakeholder standard-setting. Observers should note whether these signals reduce definitional variance across jurisdictions.