Virginia Dignum: responsible AI past the compliance checkbox.

If Amy Edmondson is the QSP speaker who tells you how to run a team, Virginia Dignum is the one who tells you what you're accountable for when that team ships. She is among Europe's most influential voices on AI governance — and, less advertised on the international circuit, she is Portuguese: born and raised here, a University of Lisbon graduate whose first job, in 1986, was building an AI system to plan subsidised housing in Lisbon. A Portuguese researcher returning to a Portuguese stage to talk about the rules Europe is writing for AI is worth the room on its own.

Who she is

Dignum is Professor of Responsible Artificial Intelligence at Umeå University in Sweden, where she leads the AI Policy Lab. The titles around that are a map of where AI rules actually get made: she served on the EU High-Level Expert Group on AI that shaped the bloc's early ethics guidelines, sits on UNESCO's expert group on implementing its AI Ethics Recommendation, the OECD Expert Group on AI Futures, and the United Nations Advisory Body on AI. She is a Wallenberg Scholar, a member of the Royal Swedish Academy of Engineering Sciences, and holds the Medal of Honour of the City of Oeiras. Her book Responsible Artificial Intelligence is a standard reference; her recent The AI Paradox is the trade-facing argument.

The one idea to walk in with

Dignum's central claim is deceptively plain: AI is not something that happens to us. Every system encodes a choice, and responsibility belongs to the people who design, deploy, and govern it — not to the model, and not to "the technology" as an abstract force. The mystified narrative, where AI is either salvation or an uncontrollable threat, is in her telling a way of dodging that accountability. For a product manager this lands uncomfortably close to home: "the model did it" is not an available defence. You chose the model, the training data, the threshold, the fallback, and the cases you decided not to test.

A second point of hers cuts against an instinct most teams share: more data does not mean fairer AI. Data distributions encode existing inequalities, so flooding a system with more of the same entrenches them rather than washing them out. The fairness question is a design question, not a volume question — which means it lands on the product spec, not just the data pipeline.

Why this is operational, not philosophical

It would be easy to file Dignum under ethics-as-garnish. The EU AI Act makes that filing expensive. The Act is now in force, with obligations phasing in, and it regulates by risk tier: some uses banned outright, "high-risk" systems carrying documentation, transparency, human-oversight, and data-governance duties that attach to the people deploying them. A PM scoping an AI feature in Europe is making AI Act classification decisions whether or not anyone has labelled them as such. Dignum's recent work asks a sharp follow-up question — whether the EU's "Digital Omnibus" simplification drive strengthens the Act or quietly dilutes it — which is precisely the kind of moving regulatory ground a product roadmap has to price in.

Her governance-enables-innovation line is the part worth taking back to a sceptical engineering lead. Her argument is that every mature industry runs on rules, and that clear constraints reduce the uncertainty that actually stalls shipping. For European teams told daily that regulation is why they're behind, that reframe — rules as the precondition for trust, and trust as the precondition for adoption — is more useful than another round of catch-up anxiety.

What to listen for in Matosinhos

Three threads. First, how she draws the line between meaningful human oversight and oversight theatre — a human "in the loop" who rubber-stamps is a compliance artefact, not a safeguard, and she is blunt about the difference. Second, anything concrete on the Digital Omnibus: simplification or dilution is a live fight, and her read signals where European obligations are actually heading. Third — given the venue — whether she connects responsible-AI capacity to European, and specifically Portuguese, competitiveness: whether governance is a moat or a millstone for the companies in that room.

Perpenda's calibration tiers — settled, contested, unsettled — exist because honest AI products have to say what they don't know. Dignum makes the same argument one level up, about the institutions deploying those products: responsibility starts with naming the choice rather than hiding it inside the model. The session worth pairing with Edmondson's.