How Can We Build Inclusive AI Science Infrastructure? | AI for Science Forum

TL;DR

Inclusive AI science infrastructure requires shared education, research opportunities, accessible data, efficient computing, collaboration, appropriate regulation, and sustainable system design. People worldwide must be able to create AI tools, not merely use technologies developed elsewhere, while institutions prevent rising energy and resource demands from deepening inequality. Read on to explore the High Performance Data Facility, practical data-access requirements, and the case for transformative sustainability goals.

Transcript

[APPLAUSE] PAUL HOFHEINZ: Thank you to everyone who's come here today to listen to this presentation. Look, I have two rules of thumb, which are pretty simple, and I'm breaking both of them today. One of them is never speak to a group that knows a lot more about a topic than you do, and there wasn't a lot I could do about that except not come. I ju... Read More

Key Insights

• AI and scientific progress are at an inflection point, creating an urgent opportunity to avoid repeating past mistakes in science and technology while preventing new innovations from widening the digital divide or leaving parts of society behind.
• Inclusive technology requires participation in creation, not merely access to finished products. Educational opportunities, research resources, and avenues for discovery must enable more people and regions to develop AI tools rather than remain dependent on technologies created elsewhere.
• AI infrastructure has substantial energy and resource demands, so sustainability cannot remain the responsibility of environmental specialists or policymakers alone. Researchers, institutions, and technology developers must consider efficiency when designing, expanding, and using large computing systems.
• Renewable energy investment alone is not sufficient because broader access will increase the number of people using AI systems. Sustainability therefore requires transformative changes in system design and use, rather than small increases in renewable supply or incremental efficiency improvements.
• Ambitious computing sustainability targets can produce results even when they initially appear unrealistic. The Exascale Computing Project included an explicit sustainability goal, and the achievement of that goal was presented as evidence that similarly demanding objectives should guide current investments.
• The High Performance Data Facility is designed to make vast scientific datasets easier to move, access, and use. It connects data from national scientific user facilities with high-performance computing resources for real-time AI and other computational applications.
• Scientific data infrastructure must preserve integrity, privacy, usability, and accessibility at the same time. Open access alone does not guarantee practical use, because people also need understandable formats, efficient movement, suitable computing connections, and protections for sensitive information.
• Scientific capacity is closely connected with economic development, yet many Global South countries have not strongly applied science to their challenges. Solutions imported from elsewhere may fail when population growth, governance constraints, and changing conditions create fundamentally different problems.

Questions & Answers

Q: How can AI science infrastructure support inclusive progress?

It can combine education, research opportunities, accessible data, efficient computing, collaborative communities, public engagement, and appropriate regulation. Inclusion means enabling different societies and regions to create AI tools and pursue discoveries, not merely giving them access to technologies produced elsewhere.

Q: Why must people participate in creating AI tools?

Access to finished technologies alone can leave some regions dependent on tools created elsewhere. Shared education, resources, and avenues for discovery allow more people to shape research priorities, build relevant tools, and participate in scientific progress.

Q: Why is sustainability central to AI infrastructure?

AI and large computing systems have substantial energy and resource demands that can worsen existing inequalities. Researchers, institutions, technology developers, and policymakers therefore need to consider efficiency and sustainability when designing, expanding, and using these systems.

Q: Why are renewable energy and incremental efficiency gains insufficient?

Broader access to AI means more people will use these systems, increasing overall demand. The panel argues that small additions of renewable energy or modest sustainability improvements will not be enough; transformative changes are needed in how computing systems are designed and used.

Q: What does the Exascale Computing Project demonstrate about sustainability goals?

The Exascale Computing Project included an explicit sustainability goal that was initially viewed as unrealistic. That goal was achieved, supporting the argument that similarly ambitious requirements should guide new computing investments.

Q: What is the High Performance Data Facility?

The High Performance Data Facility is an infrastructure initiative for managing enormous volumes of data from national scientific user facilities. It is designed to move data efficiently, connect it with high-performance computing, support real-time AI and other computational applications, and broaden access to federally funded research.

Q: What scientific data is the High Performance Data Facility intended to support?

It is intended to support data from national scientific user facilities, including particle accelerators, light sources, genomic facilities, chemistry facilities, and materials research facilities. The infrastructure aims to make those large datasets easier to transport, access, and use with AI and other computational systems.

Q: Why is open access alone insufficient for scientific data?

Data can be formally open yet remain difficult to understand, move, process, or connect with suitable computing resources. Useful infrastructure must also support understandable formats, efficient transfer, real-time applications, data integrity, usability, and appropriate privacy protections.

Summary & Key Takeaways

• Scientific success depends on infrastructure beyond laboratories, including shared data systems, educational opportunities, public engagement, regulations, collaborative communities, and sustainable computing. Future discoveries will be shaped not only by technical achievement, but also by how effectively institutions work together to realize, distribute, explain, and preserve the benefits of research.

• Inclusive AI development requires more than expanding access to finished technologies. People across different societies and regions need education, resources, research opportunities, and the capacity to create tools themselves. Without deliberate action, rapid innovation could widen the digital divide and leave disadvantaged communities bearing a greater share of environmental and resource costs.

• The High Performance Data Facility was presented as a practical infrastructure initiative for managing enormous volumes of data from national scientific user facilities. Its purpose is to move data efficiently, support real-time AI and computational applications, broaden access to federally funded research, connect data with high-performance computing, and protect integrity and privacy.