Responsible AI Part 1

I saw the poster in a hotel elevator in Wilmington. It was an ad for the hotel’s restaurant, and it read:

“Stop by our bar to enjoy our thoughtfully sourced menu, featuring classic fare with a twist and premium drinks.”

I read the phrase “thoughtfully sourced” and stopped. What does that even mean? It isn’t a certification like organic, all-natural, or vegan, where a standard exists and someone checks the claim against it. “Thoughtfully sourced” could literally mean the restaurant thought about it and decided to serve the cheapest food possible, charging as much as it could get away with. The phrase sounds good. It commits the restaurant to nothing.

That poster is what got me thinking about “Responsible AI.” The term is vague in exactly the same way, and Katie Robbert, my Trust Insights co-founder, and I have been saying so since we started this series in 2024. On the very first episode, I put it this way:

“Responsible AI seems like one of those blanket terms that says nothing, in the same way, like, responsibly sourced food. What does that mean?”

Katie’s answer, on the spot, still holds up:

“If someone says, ‘Our company practices responsible AI,’ I’m definitely going to raise an eyebrow at them, because that doesn’t mean anything.”

Two years later, a hotel elevator poster proved her right about an entirely different industry. The term still sounds like it should make you feel good. On its own, it still doesn’t actually mean anything.

This article opens a four-part series on our RAFT framework, Trust Insights’ answer to that problem: Respect, Accountability, Fairness, and Transparency. Katie and I built RAFT together to make “responsible AI” mean something specific enough for a company to act on, rather than a phrase that only feels good to say. We start with Respect, because a definition of responsible AI that skips over human values isn’t a definition. It’s a slogan.

Why This Moment Demands a Firmer Definition

Katie and I first sketched out RAFT together back in 2024. This series isn’t a rerun of that year’s thinking — the world the framework has to operate in has changed underneath it, and two shifts explain why Respect needs sharper teeth now than it did then.

The first shift is raw capability. Two years ago, AI models were, frankly, dumb as a bag of hammers. Today, they’re better than PhDs in almost everything. That doesn’t mean today’s models are right, moral, or ethical — capability and character are different things entirely. It means they’re more capable, and more capability amplifies the harm that misuse can cause. A cordless drill, in the wrong hands, can only do so much damage. A Milwaukee Hole Hawg — an industrial drill strong enough to bore a hole through a car — is a different order of danger altogether. Today’s AI tools sit much closer to the Hole Hawg end of that range than they did two years ago, which means our definition of responsible AI has to be that much more firm, clear, and defined to keep pace.

The second shift is in how companies actually deploy that capability, and it compounds the first. Rubber-stamping of AI is becoming standard practice: companies are cramming it into every product, whether or not it belongs there. Agentic AI and agentic workflows mean less human review at the exact point where a decision happens, more automation, more wholesale handoff of tasks that used to require a person to sign off. When the system pauses to ask for clarification, people mindlessly click through: “okay, okay, okay, okay.” That’s the opposite of responsible. It’s the opposite of thoughtful. Saying what you do and doing what you say requires knowing what you’re actually agreeing to, and a blind click tells you nothing about what you approved.

Responsible AI sits above legal compliance, not inside it. A system can clear every regulatory hurdle a company faces and still fail the more basic test of whether it respects the people it touches. That test is what Respect is for.

Respect, Defined: A Test You Can Actually Apply

A value that can’t be tested isn’t a value. It’s a decoration. Respect, as we define it in RAFT, comes with three concrete tests that hold up under pressure, including in the hard cases where intentions alone won’t settle the question.

The Golden Rule Tiebreaker

The first test is the simplest, and it’s the one we reach for when every other analysis stalls: would you want this done to you? Every major tradition has some version of it — do unto others as you would have them do unto you; that which is hateful to you, do not do to your neighbor. The test is pass-fail. Either you’d want it done to you, or you wouldn’t, and there’s no partial credit and no room to argue your way to a comfortable answer.

Why “The Good of the Many” Isn’t Good Enough

The Golden Rule matters because the more common ethical shortcut, utilitarian reasoning, has a real flaw hiding inside it. Most ethical frameworks lean on some version of “the good of the many outweighs the good of the few.” The problem: if “the many” already holds privilege, that framework can justify harming the few for the benefit of the privileged. The better pivot takes quantity out of it entirely. Ask instead whether a decision benefits those with privilege at the expense of those without. It isn’t a question of who benefits more. It’s a question of who’s being harmed.

Saying and Doing Aren’t Enough, Either

Even a company with sincere intentions can lean on a definition of “alignment” that’s too thin to catch real harm. Utilitarian ethics basically says that if you do what you say and say what you do, you’re ethically aligned. The uncomfortable implication is that a genuinely harmful company that discloses its harmful intent and then acts on it is, by that definition, “aligned.” The answer isn’t to abandon the say-do test. It’s to pair it with the Golden Rule, and with a harm-distribution principle: strive to avoid harm where you can, and where harm is unavoidable, put it on the people who can bear it, not the people who can’t. A $10,000 tax bill is backbreaking for someone making $20,000 a year. For someone making $20 billion a year, it’s coins in the couch cushions.

Say-do consistency alone can’t be the finish line. A company that says exactly what it intends to do, and then does exactly that, can still cause serious harm if it never asks who absorbs the cost.

Turning Principle Into a Checklist You Can Run

Respect can’t live only as language in a values statement. It has to show up in the systems a company actually builds and operates, which means turning the abstraction into an artifact someone can point to and run. Here’s the exercise: take the RAFT framework, your company’s existing values, your employee handbook, and your terms of service, and feed them into your AI tool of choice. Have it build a YAML checklist testing your policies against Respect, Accountability, Fairness, and Transparency. That checklist becomes a knowledge block: background context fed into every workflow and automation where harm is possible, and especially where harm is probable — loan approvals, healthcare claim decisions, anything with a real decisioning stake. The machine becomes a backstop, not an amplifier, against the biases that creep in naturally when humans are the only check.

That single checklist does more than one job. In this installment, it operationalizes Respect. In Part 4 of this series, on Transparency, the same YAML artifact returns as the self-audit instrument a company uses to check its own claims against its own behavior. Build it once, and it keeps working across the rest of RAFT.

Sidebar: RAFT vs. The 5P Framework By Trust Insights™

The 5P Framework By Trust Insights™ — Purpose, People, Process, Platform, Performance — structures how Trust Insights approaches any project. RAFT is the ethical lens applied within each P. Purpose: is ethical behavior actually part of your stated purpose? People: who you hire. Process: how you do business. Platform: who you choose to do business with, and who you don’t. Performance: the hardest question of all — if doing the right thing is less profitable than doing the wrong thing, which one wins at your company? That’s not a question we can answer for you. But it’s one of the real underpinnings of every conversation about responsible AI: what matters more, people or profit?

A poster in a hotel elevator can promise anything it wants. “Thoughtfully sourced” only means something once someone can check the claim against a real standard. RAFT is Trust Insights’ attempt to build that standard for AI, starting with whether a system respects the humans it touches. Part 2 takes on Accountability: who answers for the system when it gets something wrong.

Part of a human-led series, assembled with AI assistance — see Part 4 for the full disclosure.

Need help with your marketing AI and analytics?