Skip to main content

Birds of a Feather at PEARC23: Exploring the Impact of AI Computing Paradigms on Science Gateways and National Compute Resources

By: Claire Stirm, Jeanette Sperhac, Michael Zentner, Sandra Gesing, Joe Stubbs, and Rob Quick

At PEARC23 in Portland, OR, the SGX3 team led a Birds of a Feather (BOF) session called "The Impact of AI Computing Paradigms on Science Gateways and National Compute Resources." With forty expert and enthusiastic community members in attendance representing diverse scientific backgrounds, we delved into the intersections of artificial intelligence (AI) and computational science.

Embracing AI in the Realm of Science Gateways

Science Gateways play a pivotal role in democratizing access to scientific software, data collections, and high-performance computing resources. They enable researchers across various domains to harness the power of complex computations through intuitive web interfaces. As AI-based approaches gain prominence in problem-solving across disciplines, it has become imperative that Science Gateways should adapt and evolve to accommodate these novel computational techniques.

The PEARC BoF session illuminated the increasing adoption of AI within scientific workflows, highlighting the profound impact it has on research methodologies and outcomes. Discussions were marked by the acknowledgement that AI is no longer confined to specialized domains; it is permeating fields from genomics and material science to climate modeling and beyond. This broadening scope underscores the urgency for Science Gateways to evolve and cater to the burgeoning demand for AI-driven computations.

Throughout the BoF session, participants engaged in rapid ideation sessions called Brain Trusts. The room was split into four separate Brain Trust areas, each with its own focus question. The Brain Trust sessions ran for 15 minutes, and were repeated three times to allow attendees to participate in three of the four total focus questions, which were:

  • What are novel modes of computing implied by AI? Which will have the greatest impact on CI for gateways and national compute resources?

  • What are the interesting aspects related to data sharing? How do the various compute loads and modes impact how data are shared, moved, and accessed

  • What are or should be the goals of broadening access to compute resources for AI purposes? Who can be brought into the community that previously was not?

  • How can AI be used in a science gateway to make it more effective, efficient, or otherwise better?

Each 15 minute Brain Trust was structured as follows: 

  • 2 minutes: The facilitator posed the focus question and encouraged thinking out of the box.

  • 5 minutes: Attendees engaged in rapid-fire writing of ideas on sticky notes and were encouraged not to think to eliminate their ideas out of a feeling they might not be right or good enough.

  • 5 minutes: Attendees placed their sticky notes on a grid in any of the following categories and engaged in discussion about their ideas after placement:

    • Easy, just do it

    • Very high impact

    • Wild and crazy

    • Important but requires programmatic funding

  • 3 minutes: Attendees chose another Brain Trust area in which to participate next

The discussions resonated with the collective vision of paving the way for the next 5-10 years of computational advancement. 

Key Takeaways and Insights

Several key takeaways emerged from the PEARC BoF session’s discussions. Participants acknowledged that AI workloads are complex and that creating and running them can be daunting. They recognized that helping users manage this complexity provides gateways with the opportunity to shine.

Judged on the ideas they voiced during the discussions, participants were especially keen to see gateways employ AI to automate and assist with the tasks involved in preparing and running AI workloads. Gateway users, not being computing specialists, will benefit from this assistance and guidance. Participants told us gateways should use AI itself for streamlining how users build workloads, from guiding and automating data preprocessing, to assisting with resource and parameter selection, to helping with training and job submission. 

They voiced numerous concerns about the challenges presented by data management in AI workloads, including data cleaning, ethics, provenance, and access. By comparison, participants had far fewer concerns about the availability of suitable computational resources for AI workloads than they did about issues surrounding data.

Participants also acknowledged that intuitive gateway user interfaces and effective user assistance are key for user adoption and retention. Participants want to ensure that gateway users are well supported by quality documentation and help features such as chatbots and interactive assistants. They asked for gateway usability to be enhanced through judicious use of AI features such as natural language to API translation, as well as assistance with workloads. 

Users mentioned these themes frequently:

  • Task automation
  • Workflow creation and management
  • Dataset management, handling, and sources

The following were the most frequently mentioned participant concerns and topics, in descending order:

  1. Gateways could utilize AI techniques to provide conversational, interactive assistance and suggestions to gateway users. Topics highlighted include:
    1. Interactively guiding data preprocessing
    2. Guiding user selection of compute resources
    3. Providing a natural language chatbot/user assistant
    4. Guiding dataset selection
    5. Interactively guiding workflow composition and job formulation
  2. Gateway usability. Specific topics included:
    1. Caching previous computations
    2. Natural language to query, and natural language to API translation
    3. Guided, conversational data exploration and visualization
    4. Suggestions should utilize user history on the gateway
    5. User interface customization
    6. User support, documentation, and assistance
  3. Computation and (model) training. Specific topics included:
    1. Job scheduling assistance
    2. Guiding parameter selection
    3. Predicting results of simulations
    4. Comparing physical models with proxy models
    5. Guiding model training
    6. Interactive computing
  4. Data access and data management. Specific topics included:
    1. Data cleaning
    2. Data ethics
    3. Data provenance
    4. Protected Health Information, or other personally identifying data
    5. Access to training data
    6. Data from edge devices
    7. Data compression

Unveiling the Road Ahead

In conclusion, the Birds-of-a-Feather session at PEARC23 provided a platform for the research computing community to begin envisioning the future of AI-infused Science Gateways and computational resources. As the SGX3 Center of Excellence embarks on its mission to document these trends, a Blueprint Factory focused on AI enhancements will gauge the community on addressing challenges and identifying next steps. If you wish to be involved in this activity, contact us at