New Hampshire has two openings to improve its workforce this fall. Governor Kelly Ayotte proposed a Community College Promise Program in October that would cover tuition for qualifying students from households earning up to $100,000, subject to future legislative approval. The New Hampshire Tech Alliance also published an October 2026 report on the state of AI. These are separate efforts, but policymakers should connect their ambitions around one practical question: Will graduates be better at making sound decisions when AI supplies a convincing mistake?
Free or more affordable education can help people enter skilled occupations. AI training may help them advance. Neither promise tells an employer whether an apprentice can distinguish a plausible answer from a correct one. That distinction belongs at the center of any publicly supported AI course or workforce partnership.
Imagine a small Granite State manufacturer hiring a new production planner. An AI assistant creates a purchasing recommendation and confidently suggests a substitute component. The substitution looks sensible until someone checks a certification requirement that the software omitted. The competent employee examines the source document, rejects the recommendation, and explains the change to a supervisor. An entry-level worker needs structured opportunities to learn that judgment.
Make those opportunities explicit. For each AI course funded by a workforce grant, ask participating employers to supply three realistic examples of where an incorrect answer could cause rework, delay, or harm. Instructors can anonymize those examples, deliberately insert errors into simulated outputs and ask students to identify what went wrong. The exam should reward evidence, correction and escalation rather than fast prompting.
This also means preserving an apprenticeship path. If software prepares every routine draft, the novice loses the small assignments that once taught how a process works. Training designers should restore that learning through supervised exercises, compare an AI-generated recommendation with the real source, and then let trainees handle progressively harder work. The employer should name a human mentor who can explain why an answer fails.
New Hampshire already has educational approaches worth building on. An MIT pilot with Lakes Region Community College introduced high school students to the mathematics behind AI and practical problems. A Keene precision-optics workforce partnership emphasizes specialized technical education tied to real manufacturers. These programs serve different groups, and neither should be reduced to a generic chatbot class. Their shared lesson is that practical competence needs more than exposure to technology.
Policymakers can make an AI training grant accountable without adding much bureaucracy. Before the course, record how often trainees catch a realistic error. After supervised practice, run a different test and measure improvement. Ask employers whether graduates can work independently sooner, whether avoidable rework falls, and how much mentoring still matters. Publish those outcomes in aggregate so applicants and taxpayers can judge the results.
The same standard should apply to organizations buying AI education with private money. Count successful corrections, useful decisions, and responsible handoffs. Do not call a course effective merely because hundreds finished it or because an application reports heavy use. Usage can rise while mistakes and cleanup costs rise with it.
New Hampshire’s affordability and technology plans could open doors for many residents. The opportunity becomes more valuable when those residents leave training able to challenge the tools they are expected to use. A workforce prepared to say, ‘Show me the source,’ will be better equipped for an AI-assisted economy than one trained simply to accept the answer.
========================================
Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026) https://disasteravoidanceexperts.com/aibook
Authors’ and Speakers’ opinions are their own and may not represent those of Grok Media, LLC, GraniteGrok.com, its sponsors, readers, authors, or advertisers.
Disagree, agree, Got Something to say? We Want to Hear It. Comment or submit Op-Eds to steve@granitegrok.com
