No direct regret survey exists for proactive AI retraining decisions — the rates below are forward-looking concern and aspiration proxies, not retrospective measurements. Among US adults, 56% are extremely or very concerned about AI-driven job loss and 64% believe AI will lead to fewer jobs over the next 20 years, according to a Pew Research Center survey of 5,410 US adults fielded in August 2024. Only 23% of the public believes AI will have a positive impact on how people do their jobs. Workers in the most AI-exposed roles — approximately 19% of the US workforce by Pew’s 2023 occupational analysis — face the highest structural risk: these tend to be higher-wage roles ($33/hour on average versus $20/hour in least-exposed jobs), meaning the displacement stakes for waiting are significant.
The decision to wait is not simply a bet on AI disruption failing to materialise — it is also a bet on timing. The specific AI tools and skills that are most valuable today (prompt engineering, workflow automation with a particular platform, specific coding assistants) evolve faster than most training curricula. A worker who retrains today for a particular AI application risks finding portions of that skill set superseded within a few product cycles — a plausible obsolescence risk given the pace of AI capability change, though no study has yet tracked how often this actually happens to early retrainers. This obsolescence risk generates the primary mechanism for action-side regret: early retrainers may find that one-time credential acquisition is insufficient and that the decision is better framed as beginning continuous learning earlier versus later. Only 36% of organisations operate as career development champions with structured learning programs (LinkedIn 2025 Workplace Learning Report), which means the majority of workers who do retrain are doing so without employer support — the condition most associated with higher credential mismatch rates.
Gilovich and Medvec’s temporal asymmetry research predicts that inaction regrets tend to grow over time while action regrets fade, a pattern especially likely in AI disruption because consequences accumulate slowly. Workers in AI-exposed roles may experience years of gradual wage stagnation and narrowing opportunity before attributing the outcome to delayed adaptation. By the time the inaction regret crystallises, the gap to close is substantially larger than it would have been with earlier action. The 56-to-15 inaction-to-action concern ratio should be read as directional given the prospective nature of the evidence — but the pattern is consistent with the broader Gilovich/Medvec literature on career deferral decisions, and with the structural employment data showing that AI-exposed workers are already experiencing measurable wage and opportunity divergence from less-exposed peers.