Fireside Chat, Interview
How AI is Powering Payments, with Greg Ulrich of Mastercard
- The future of AI remains undefined, with expectations that reasoning models will increasingly admit uncertainty to reduce confident inaccuracies.
- Major advancements are anticipated in multi-modal capabilities integrating text, images, voice, and video to establish a "single source of truth" for tasks like invoice reconciliation.
- Trust, responsibility, and transparency are identified as critical operational pillars for the foreseeable next few years.
- As scaling laws advance, data and inference quality will serve as primary differentiators for AI solutions rather than just model size.
- Key performance indicators and specific targets will be established for every new initiative to measure value for customers, employees, and the ecosystem over time.
- A "human in the loop" approach will be maintained for customer-facing solutions while awaiting further improvements in model accuracy and latency.
- Generative AI initiatives will prioritize use cases involving manual tasks or unstructured data where the technology is deemed the appropriate tool.
- The organization plans to leverage the StartPath program to enable early-stage companies to access the MasterCARD network and receive support.
- A "hub and spoke" governance model will facilitate two-way collaboration, allowing enterprise-wide teams to provide learnings to business units while the central team receives new product ideas.
- Data safeguarding is positioned as the highest priority to maintain the organization's status as a trusted brand.
- Generative AI will be integrated into existing fraud solutions to improve scoring accuracy rather than replacing current systems entirely.
- Digital assistants using RAG and fine-tuned models are expected to reduce onboarding times for MasterCard products by automating manual tasks.
- Partnerships will be pursued only with entities that share commitments to governance, security, and shared values.
- Leadership intends to stay informed on rapid technological progress and organizational structures through podcasts, external news, and networking.
- The organization will present AI updates to its board of directors utilizing external speakers from investor, LLM, and academic sectors to provide varied perspectives.
- While enthusiasm exists, significant concern regarding accuracy and hallucinations may cause other organizations to delay direct consumer applications until proven.
- AI activities will not be fully centralized given the organization's scale and the ubiquitous nature of AI, described as comparable to electricity.