Article Operations Organizational Strategy 4 min read
Why Associations Should Apply Their Own Review Standard to AI Use
Associations built the machinery to certify, accredit, and police standards for their fields. Their own AI use deserves the same treatment.
When the Royal College of Physicians received its charter in 1518 (opens in a new tab), it established the role of censors to “enquire about all practitioners of medicine … to examine, correct and govern them, if necessary to prosecute them.” Scientific academies followed suit a century later. Their foundation was the same: standards for quality, for consistent practice, for ethical conduct. These remain the very premise upon which certification, accreditation, and every ethics process depend. Someone will check. Standards that no one enforces are merely suggestions.
Now read your association’s AI policy against that last sentence. Is it a standard or a suggestion?
A Policy Versus a GateÂ
A policy is a statement of intent. What we expect. A gate is a control that’s positioned so something stops when a condition isn’t met. The real test isn’t what the document says. It’s what happens when someone doesn’t follow it. If the honest answer is “nothing … until we notice,” you have a policy masquerading as a gate.
That distinction mattered less when tools behaved predictably. Daniel Wallance, a senior expert at McKinsey & Company, put it simply in a recent Blackbaud webinar: (opens in a new tab) Today’s tools “are producing outcomes that are nondeterministic.” A deterministic tool can be governed by a rule you remember, because you test the rule once and it holds. A nondeterministic one can’t. The same prompt from the same person might be just fine on a Tuesday and wrong by Friday with no changes at all.
The Harm Isn’t the ErrorÂ
A fabricated citation can be corrected. A clumsy summary can be rewritten. Permanent damage won’t be inflicted that way. It’ll be cumulative and invisible—made of unchecked output such as paraphrased field guidance, a misplaced item that slipped into a certification bank, or member communications drafted by a tool and only scanned by a person. Then a member, an auditor, or a reporter pieces a few of them together. A single error won’t cause irreparable harm. The discovery that no one was checking will.
For a company, an ungoverned AI error might be considered an operating problem. Fix it, apologize, move on. I’d argue an association is fundamentally different. Its authority rests on policing standards on other people’s behalf, on assuring members, employers, and the public that a credential means that someone looked. For an association, “We had a policy, and no one followed it” isn’t just a process failure. It’s a statement about whether the organization is what it actually claims to be.
Where the Gate Should GoÂ
I’ve been working with a nonprofit that supports refugees. The organization does a lot with very little, so securing grants is a matter of survival. Interns had been using AI to complete sections piecemeal. Examples were duplicated, figures were inconsistent, and errors were untraceable. Now we’re building an agent that assembles a draft from vetted material, identifies gaps, and compounds learning over time. Nothing will be submitted to a funder until a program lead signs off on each section.
The operative question isn’t what to ask about a new AI tool. It’s where the machinery needs to sit so the work can’t move forward until someone says so.
- Position it at moments of consequence. Those moments happen before member-facing guidance goes out; before an item enters a certification bank; inside credentialing, editorial, or psychometric review; before a grant application is submitted.Â
- Stop the next step from proceeding. In a small association, the gate may only be one person. If that one person can skip it, then it isn’t a gate. What makes it real requires repeatable, unskippable structure: an approval field, a second name on the record, a queue that can’t advance without explicit sign-off.Â
- Gate the use, not the organization. An internal meeting summary and accreditation feedback don’t present the same risk, and they shouldn’t have the same bar to clear. Calibrating the gate to the stakes is the answer to “we’re not big enough.”Â
- Route it through what you already run. Most associations have a gated review process in place, whether for member-facing work or something else. The guidance isn’t “build governance.” It’s “send this through machinery you already have and trust.”Â
Professional associations have spent five centuries building the machinery that makes a standard mean something: censors, committees, and accreditations that an organization must earn and can fail. You already own the answer. Your reputation is built on trust, and that trust is built on always checking. Now the question is whether you’ll point that machinery at your own use of AI or wait to learn—perhaps much later and in aggregate—that no one was checking.
Comments
Comments are for ASAE members. Learn more about comments on this site.
Log in to your member account or join ASAE!