||| FROM GLEB TSIPURSKY for DISASTER AVOIDANCE EXPERTS |||
Orcas businesses do not need a grand AI strategy before they start experimenting. They need a clear rule for when automation stops and a person takes over.
San Juan County itself recently adopted generative AI use policies, with the recognition that the guidelines will have to evolve as the technology changes. At the same time, the Economic Development Council’s 2026 Economic Survey is asking business owners, managers, and workers to identify the challenges and opportunities shaping the local economy. Those two developments point toward the same practical question: how do we use new tools without losing judgment about what works here?
For a small business or nonprofit, AI can help draft customer messages, summarize documents, prepare marketing copy, organize research, or generate a first version of a proposal. The central risk appears when responsibility becomes fuzzy, especially when the system is wrong.
Every recurring AI-assisted workflow should therefore have a human handoff rule. Define the conditions that automatically send the task back to a person. Those conditions might include uncertain facts, unusual customer circumstances, a price or refund, a safety issue, a financial commitment, or any public statement that could damage trust if it is wrong.
Then keep a simple exception log for 30 days. Record where the automated output needed correction, what context it missed, how much time the fix took, and who made the final call. At the end of the month, review the pattern.
That exercise often reveals what software dashboards miss. A tool may make the first step faster while quietly creating more checking, correction, and follow-up. Or it may prove genuinely useful on routine work while repeatedly failing on the cases that require local knowledge.
The point is to make experimentation accountable. Clear handoff rules let employees and owners know where the boundaries are. They also make it easier to expand a workflow once the evidence shows it deserves to scale.
On an island where businesses and community organizations depend heavily on trust, the most useful AI policy may fit on one page: what the system may do, when a person must step in, and how exceptions get recorded. That is enough to turn experimentation into learning rather than guesswork.
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://
**If you are reading theOrcasonian for free, thank your fellow islanders. If you would like to support theOrcasonian CLICK HERE to set your modestly-priced, voluntary subscription. Otherwise, no worries; we’re happy to share with you.**
I read this story a couple of days ago:
“An AI medical scribe in Australia invented a claim that a patient takes psychedelic mushrooms. She’s never touched them. It went into her medical record, got sent to her GP, and could have affected her workers’ compensation case.
Peer-reviewed research found 70% of AI scribe notes contain at least one error, with 44% of the hallucinations classified as serious enough to affect diagnosis or treatment. In another study, 31% of AI-generated notes contained fabricated information. One major platform alone processes over 3 million medical encounters a month.”
https://www.abc.net.au/news/2026-08-14/ai-medical-scribe-error-leaves-patient-devastated/107031672
AI regularly “hallucinates”. I think it’s very concerning that our local government and/or local businesses would use AI. What rights do customers/clients/patients/constituents have to decline the use of AI in their relationship with these authorities/businesses? Any? Why should we trust that our data and interactions will be handled appropriately? Why should the onus be on us to triple check everything, to ask during every interaction? How many lives will be derailed when we have to deal with the potential impacts of this? (As in the story above, months of dealing with trying to correct an error.)
It’s like the County and businesses are hiring someone who they know lies 70% of the time. Is that really a good idea?
Humans seem to have thrown the precautionary principle out the window, because it might save a few bucks. Crazy times.
I favor the definition of AI advocated long ago by computer-mouse inventor Doug Engelbart: Augmented Intelligence. Which means that a human mind or hand should be involved in interpreting and using the AI output. A human, that is, who can take the responsibility if the output turns out to be wrong or dangerous.
Meanwhile the AI companies are going full steam ahead building systems that they don’t seem to be in control of, and by building AI into our human systems in a way that makes them difficult to “turn off”, we are setting ourselves up for a hell of a FAFO. AI-generated malware is happening. AI autonomous agents getting out of control and hacking companies is happening. I think the “augmented intelligence” days are long gone.