A synthesis of insights from the panel “Putting Responsible AI into Practice: Roadmaps and Governance” and related flash talks
Beyond Ethical Checklists: The Business Case for Responsible AI
“Don’t think of AI responsibility as well-understood,” cautioned one panelist during the discussion moderated by Ricardo Baeza-Yates. This sentiment captured the evolving nature of AI governance—shifting from abstract principles to actionable frameworks that acknowledge context and position in the AI value chain.
Kerry McLean from Intuit articulated this evolution clearly: “The folks who create the foundational models have a different set of responsibilities than the folks who use the foundational models to create applications or services.” This distinction is crucial as organizations operationalize responsible AI practices.
At Intuit, responsible AI isn’t merely a compliance exercise but aligns with their mission of “power and prosperity for our customers.” McLean emphasized that AI could potentially “level the playing field” for small businesses—which “fail at a rate of 50 percent in their first five years”—by giving them “services that only very large companies or very wealthy individuals have access to.”
The Challenge of Regulatory Navigation
The global patchwork of AI regulation presents significant challenges. While over 100 AI-related bills were proposed in the U.S. last year, none passed at the federal level. Meanwhile, approximately 800 bills were introduced across 45 states, creating a fragmented landscape.
McLean highlighted this complexity: “If you’re a company trying to operate even in just the United States or globally, your privacy governance world… is incredibly complex.” Adding AI regulations to this mix creates further uncertainty that “is not good for [companies] who are trying to figure out how they comply with those regulations, it’s not [predictable], it’s not good for universities, it’s not good for investment.”
Sudha Jamthe illustrated this divided approach with a concrete example: “Whenever you go to any website or any mobile app or anything… we have to opt out. And the minute you step into… Europe… I see that every standard app that I’m using pops up, because then I have to opt in.” The result? “When we’re building products in a global company today, there’s two rounds of products” with different interfaces depending on jurisdiction.
Trust Infrastructure and Digital Identity
Rod Boothby’s flash talk on digital identity illuminated a central challenge for responsible AI: establishing trust in digital interactions. “Most random humans you bump into. Could you give them your wallet for 10 minutes? The keys to your house for 10 minutes?” he asked, highlighting how trust varies dramatically across societies.
In countries like Norway and Sweden, bank-based digital identity systems have reduced online fraud by “99.4%” according to Boothby. He contended that such trust infrastructure is essential in an era where AI makes it increasingly difficult to verify authenticity.
“Trust is the most important resource that we have. More valuable than gold. It’s more valuable than oil and data,” Boothby emphasized. “AI means we can create data, add information, verified data, data you can trust—that’s valuable.”
The Personal Nature of AI Interactions
The panel highlighted how AI experiences vary significantly between users. As Sudha Jamthe noted: “When you interact with AI and I interact with the same AI, it’s not the same. Like, it’s going back to its own memory, trying to bring all its biases and learning, and it’s going to give [answers that are] stochastic.”
This personalization creates challenges for testing and governance. “When you test it as an engineer,” Jamthe explained, systems may “behave differently based on who they are.” She illustrated this with a classroom example where a computer vision model trained on dogs and mops struggled to categorize human faces: “And given my hair, I think it usually (ridiculously) thinks I may be a dog.”
From Theory to Practice: Creating Governance Structures
Wael Mahmoud from Airbnb outlined practical steps for integrating responsible AI throughout the product lifecycle. He emphasized that responsible AI must be “integrated early” rather than added as an afterthought, with “centralized team oversight” providing consistency across products.
This requires embedding ethics into education and practice. As Jamthe noted, “I teach courses on AI and in all my courses I have like ethics… In recent times, I started incorporating ethics when I [teach] AI.” Diverse teams are central to this approach: “You bring people of multiple perspective, multiple [lived] truths, their ethics.”
Jorge Sanz of IBM Research emphasized that responsible AI varies depending on an organization’s role in the AI value chain. “It’s not the same to be responsible for creating models as it is to deploy them at scale,” he explained, noting that accountability differs for those designing foundational AI versus those integrating it into applications. He highlighted how interpretations of responsibility shift across jurisdictions, influenced by legal frameworks and cultural norms.
“Even within the European Union, where laws are harmonized, interpretations can still differ,” he said, underscoring the complexity of global complianceSanz also addressed the challenge of balancing AI performance with interpretability. “You can always use a simpler model, but it might be less accurate,” he noted, pointing to the trade-off businesses face between transparency and predictive power. He stressed that governance at scale requires conscious investment, particularly as companies grapple with the question: “Do we build a single global monitoring system, or do we tailor compliance regionally?” The decision, he suggested, has profound ethical and operational implications for organizations navigating an increasingly regulated AI landscape.
Andreen Soley challenged the notion that technology should be shaped solely by engineers and policymakers, advocating instead for a truly multidisciplinary approach. “Public interest must be a priority in technology development, not an afterthought,” Soley asserted, emphasizing the necessity of bringing social scientists, ethicists, and community voices into AI and digital governance. Her work at PIT-UN reflects this vision, fostering a network where technologists are trained not just to build, but to build with intention—prioritizing equity, transparency, and societal impact.
Her reflections on the barriers to interdisciplinary collaboration underscored a broader issue within tech culture. “We speak different languages, have different incentives, and evaluate success in fundamentally different ways,” she noted, pointing out that true innovation requires bridging these divides. Through initiatives like the AI for Impact Co-Op Program, PIT-UN is actively equipping students and professionals to navigate this complexity, ensuring that the next generation of technologists isn’t just proficient in code, but in crafting technology that serves the public good.
The First Ethical Test
Perhaps the most fundamental test of responsible AI is what Ricardo Baeza-Yates posed as a simple question: “Would I use my product or not?” This personal accountability represents a starting point for responsible development that goes beyond compliance checklists.
As organizations navigate the complex landscape of AI governance, they face the challenge of balancing innovation with responsibility, speed with ethics, and global standards with local contexts. The unfinished revolution of responsible AI requires intentional stewardship—not just to prevent harm, but to realize AI’s potential as a force for inclusive prosperity and human flourishing.
As the summit drew to a close, one theme resonated throughout: Responsible AI is not a static concept, but an evolving challenge requiring thoughtful engagement across disciplines, industries, and global contexts. Each speaker—whether focusing on governance, digital identity, fairness in AI systems, or public interest technology—emphasized that the true measure of Responsible AI lies in its ability to balance innovation with accountability.
Ricardo Baeza-Yates’ guiding question, “Would I use my own product?”, underscored the ethical responsibility at the heart of AI development, while Kerry McLean, Sudha Jamthe, and Jorge Sanz illustrated the need for adaptable governance structures that navigate legal complexity without stifling progress.
The conference, hosted by the Institute for Experiential AI, reaffirmed that AI is most powerful when it extends human intelligence rather than replacing it. Andreen Soley’s vision of a multidisciplinary approach, Rod Boothby’s call for trust infrastructure, and Wael Mahmoud’s insights on integrating responsibility into fast-moving teams all pointed to a shared goal: AI that serves society equitably, ethically, and effectively. The future of Responsible AI will be shaped not by any single framework, but by a collective commitment to continuous learning, adaptive regulation, and the intentional design of technology that fosters trust, prosperity, and inclusion.
