On September 9–10, Wharton Human-AI Research (WHAIR) convened academic and industry researchers at Wharton San Francisco for the 2026 Business & Generative AI Conference. Across two days of keynotes, research presentations, a poster session, and discussions, participants explored an increasingly important question: As generative AI grows more capable, how can organizations ensure that it creates meaningful and lasting value?

Dozens of researchers from institutions around the world presented work spanning labor markets, healthcare, teamwork, entrepreneurship, news, marketing, search, personalization, and agentic systems. Three keynote speakers, Faye Iosotaluno, Angela Duckworth, and Ramayya Krishnan, and two plenary presenters, Hemant Bhargava and Hema Yoganarasimhan, framed that broader conversation from the perspectives of consumer technology, behavioral science, and organizational deployment. Together, they emphasized that AI’s future will be determined not only by what these systems can do, but by the choices people and institutions make about how they are designed and used.

Designing AI for Trust

Faye Iosotaluno, C’02, W’02, President and COO of Entangled Publishing and former CEO of Tinder, opened the conference with her presentation “Consumer AI & The Architecture of Trust.”

Drawing on her experience building consumer products that shape how people meet and form relationships, Iosotaluno argued that when technology begins to shape human relationships, product design becomes social design. Generative AI raises the stakes because it can advise users, communicate and act on their behalf, and even simulate relationships with them.

As consumer AI assumes a more active role in people’s lives, trust must mean more than technical reliability. It must encompass stewardship: whether a product supports the user’s judgment, confidence, capability, and agency. Traditional short-term measures such as engagement, time spent, and retention may show that a product is compelling, but not whether it is beneficial. A system can maximize engagement while encouraging dependence, replacing human connection, or weakening skills.

Iosotaluno challenged builders and consumers to ask not only whether AI can perform a task, but what role it should play. In some contexts, the highest-trust product may coach, prepare, or explain rather than simply do everything for the user. Her message offered businesses a valuable framework: Capability may drive adoption, but trust earns permission to scale.

Aligning AI with Our Present and Future Selves

Angela Duckworth, the Rosa Lee and Egbert Chang Professor at the University of Pennsylvania and faculty co-director of the Behavior Change for Good Initiative, brought a behavioral-science perspective to “The AI Alignment Problem and Multiple Selves.”

AI alignment is often described as getting a system to do what humans want. Duckworth complicated that definition by asking: Which version of the human?

People frequently experience conflict between a present self seeking convenience or gratification, and a future self pursuing longer-term goals. AI can intensify that conflict because it excels at satisfying immediate needs. It can finish an assignment, draft a message, offer reassurance, or provide frictionless companionship. Those benefits are real, but they may come at the expense of learning, judgment, resilience, or human connection when AI consistently replaces the effort through which those capacities develop.

Research on AI-assisted learning and companionship shows that the effects are not uniformly negative. A well-designed tool can provide immediate help while strengthening a user’s abilities; other uses may meet an emotional need while creating longer-term costs. The alignment challenge, then, is not simply to satisfy a request in the moment. It is to design systems and norms that account for both immediate preferences and long-term well-being.

Moving from Model Capability to Organizational Value

Ramayya Krishnan, dean emeritus and the W.W. Cooper and Ruth F. Cooper Professor at Carnegie Mellon University’s Heinz College, delivered “AI Deployment: Integrating Evaluations, Operations and Economics.”

Krishnan shifted the focus from what models can do in isolation to what happens when organizations deploy them within real systems. Model capability alone is not enough to create value; organizations must connect evaluation, operations, economics, workflows, and workforce strategy.

Reliable evaluation is the first step. A model that answers correctly once may not remain reliable over a longer or adversarial exchange. In consequential settings such as healthcare, evaluation must reflect real operating conditions, and monitoring tools must detect when performance is likely to deteriorate.

Deployment then becomes a workflow-design problem. Introducing AI into a task can alter labor costs, coordination requirements, and responsibilities, producing new workflows and roles rather than simply automating existing ones. By modeling how tasks are divided among people, AI, and human-AI teams, leaders can examine how deployment choices affect performance, costs, roles, and skills.

This connects AI strategy directly to workforce development. If AI changes the tasks that define a job, it also changes the skills workers need. Krishnan highlighted efforts to help workers identify their existing skills, understand emerging roles, and access relevant learning opportunities. His keynote underscored a recurring conference theme: AI’s value will depend on complementary investments in evaluation, process design, human capital, and institutional capacity.

A Research Community Focused on Real-World Impact

Beyond the keynotes, the conference agenda featured dozens of researchers investigating how generative AI is changing work, healthcare, marketing, digital platforms, personalization, and human decision-making. Two plenary presentations offered a closer look at the kinds of questions this research community is working to answer.

Hemant Bhargava, distinguished professor of management; Suran Chair in Technology Management; director of the Center for Analytics and Technology in Society, University of California, Davis, examined the rapidly expanding, and increasingly fragmented, landscape of state AI legislation. His research asks what states are attempting to regulate, how their priorities differ, and what factors help determine whether an AI-related bill becomes law. Using a multidimensional policy taxonomy and a multi-agent classification system, the work offers a framework for interpreting thousands of bills according to who is being constrained, whom the legislation is intended to protect, and what potential harm it addresses. The project illustrates how AI can help researchers study the institutions emerging to govern the technology itself.

Hema Yoganarasimhan, professor of marketing; Michael G. Foster Faculty Fellow, University of Washington, approached AI improvement from a different direction. Her presentation introduced TextBO, a framework designed for settings in which generating possible solutions is inexpensive but evaluating them is costly. Generative AI can rapidly produce advertisements, prompts, designs, or other possible solutions, but determining which ones actually work may still require experiments, expert review, or real-world testing. TextBO applies ideas from Bayesian optimization in language space to help AI systems improve while using fewer evaluations — an increasingly important capability for business and societal applications.

Together, the presentations reflected the conference’s broader emphasis on studying generative AI in context. The central questions extended well beyond what models can produce: How should AI be evaluated? How can experimentation become more efficient? What rules should govern deployment? How will those rules affect organizations and markets? By bringing faculty members, postdoctoral researchers, doctoral students, and industry experts into the same conversation, the conference connected technical advances with the managerial, behavioral, economic, and policy environments that will ultimately shape their impact.

Learn more about the 2026 Business & Generative AI Conference and explore the complete conference agenda.