The Montréal Declaration, Translated for a Store Owner
What does the Montréal Declaration for Responsible AI actually ask of a business owner? It asks you to run AI that protects people's well-being, respects their autonomy, keeps their private data private, and stays honest about what it is and isn't doing. Université de Montréal built these ten principles through public consultation in 2017–2018 (Montréal Declaration, 2018). None of it is abstract if you run a call- and appointment-driven business. It is a strict checklist for how your AI must treat callers.
Where did the Montréal Declaration come from and why should an operator care?
Université de Montréal launched the Declaration in November 2017 after a public co-construction process. Instead of a corporate policy team, researchers, ethicists, and citizens debated what AI owes the people it touches (Université de Montréal, 2017). They produced ten principles: well-being, autonomy, privacy and intimacy, solidarity, democratic participation, equity, diversity and inclusion, caution, responsibility, and sustainable development (Montréal Declaration, 2018). This matters to you. You will not face an audit against it, but it remains the clearest plain-language articulation of what "responsible AI" means before regulators defined it.
What does the "well-being" and "autonomy" principle mean for a phone line or booking desk?
The well-being principle holds that AI development should ultimately serve the psychological and physical well-being of the people it interacts with, not just efficiency metrics (Montréal Declaration, 2018). In plain terms: if your AI system is measured only on calls handled per hour, you have already violated its spirit. The autonomy principle requires that AI systems be designed to increase people's control over their own lives, not diminish it (Montréal Declaration, 2018). A caller must always be able to reach a human, cancel a booking, or say no to a suggestion without friction designed to trap them. If your booking AI makes cancellation harder than booking, the principle fails in practice.
What does the privacy principle require when your AI is handling phone numbers, appointment history, and payment intent?
The protection of privacy and intimacy principle states that AI must guarantee people's privacy and control over their personal data, limiting collection to the declared purpose (Montréal Declaration, 2018). For call- and appointment-driven businesses, the mandate is clear. Do not let your AI vendor default to storing every call transcript indefinitely "in case it's useful," and never let scheduling data quietly become marketing data without telling the customer. This is the same discipline covered from the regulatory side in customer lists are not training data. The Declaration and GDPR-style purpose limitation reach the same conclusion from different directions.
Which principles govern honesty and caution — and what do they forbid?
The caution principle requires prudence and holding off deployment of AI whose risks aren't well understood; the responsibility principle requires that responsibility for AI decisions never be diffused so completely that no one is accountable for the outcome (Montréal Declaration, 2018). Translated to a front desk: if your AI cannot reliably tell a customer whether their appointment is confirmed, do not let it say "confirmed." And when something goes wrong — a double-booking, a missed callback — there needs to be a specific person accountable for fixing it, not a shrug toward "the system."
A plain-language operating table
| Declaration principle | What it asks of AI development | What it means for your phone/booking line |
|---|---|---|
| Well-being | Serve people's welfare, not just efficiency | Don't optimize solely for call-handling speed |
| Autonomy | Increase people's control | Easy human handoff and easy cancellation |
| Privacy & intimacy | Limit data to declared purpose | No silent data reuse; scoped retention |
| Caution | Don't deploy poorly understood risk | Don't let AI assert what it can't verify |
| Responsibility | Keep accountability traceable | A named owner fixes AI-caused errors |
What should an owner actually do with this?
Use it as a due-diligence lens on any AI vendor, including us. Ask what data the system retains and why. Ask whether a customer can always reach a person. Ask who owns the failure when the AI gets something wrong. Those three questions map directly onto the Declaration's privacy, autonomy, and responsibility principles. Any vendor should answer them without hedging. Our approach to data boundaries lives on /security, and the escalation duty this implies in practice is laid out in Human-in-the-Loop Is a Duty.
The Declaration was never written for a vendor to comply with in a filing. Researchers wrote it for the person on the other end of the interaction: your customer. Read it through their eyes, then demand your vendor meets the standard.
Keep reading: why we build this way.