U.S. Senate AI Insight Forum: Innovation - Read written comments from Austin Carson, Founder & President, SeedAI
Leader Schumer, Senator Rounds, Senator Heinrich, Senator Young, and other distinguished attendees of the second U.S. Senate AI Insight Forum, thank you for the opportunity to present SeedAI’s views on innovation and artificial intelligence. In these comments, I outline why I created SeedAI, what SeedAI has learned by engaging directly with communities across the country, and how I think the United States can support responsible innovation in AI technology through shared infrastructure.
About SeedAI
SeedAI is a 501(c)(3) nonprofit organization that engages directly with communities to democratize access to AI technologies and job opportunities, as well as ensure that communities have a voice in how powerful new AI technology is developed and deployed. I founded SeedAI out of an early recognition of the dramatic transformational potential of AI, but also a concern that the benefits of that technology would be concentrated in the hands of a few, as we have seen with prior technological revolutions.
SeedAI works to ensure that everyone across the geographic and economic spectrum–not just a select few–can influence and benefit from transformative AI technologies. We believe that this will lead to a fairer society, better AI policy, and a more robust and diverse workforce, enabling the United States to maintain its leadership position in driving technological innovation.
SeedAI’s Programs and Activities
SeedAI’s activities revolve around high-impact, high-touch programming, and we have launched numerous programs in direct service of our mission. Some key initiatives include:
Co-organizing the Generative AI Red Team Challenge at DEFCON 2023, which drew thousands of people to test generative large language models from various companies for bias, potential harms, and security vulnerabilities. Working with the White House, SeedAI organized and funded the participation of hundreds of students from community colleges and historically black colleges and universities traditionally left out of the early stages of technological development.
Implemented our AI Across America program, which brought together experts across a variety of disciplines, geographic regions, and socioeconomic backgrounds to make AI education, training, and R&D available for underrepresented communities across the country. SeedAI partnered with the Congressional AI Caucus on this program and convened forums in Palo Alto, Chicago, Houston, and Washington, D.C. during the first year of the initiative. Hack the Future is our expanded initiative building on the work from AI Across America, with events planned for Tulsa, Houston, Cleveland, Chicago and other cities that directly build ecosystems to allow these underrepresented communities to prosper through economic growth enabled by AI.
Researching and convening experts to explore the potential of a comprehensive U.S. infrastructure that supports AI development and implementation from conception to end use and democratizes access to cutting-edge technology, in partnership with the Congressional AI Caucus and the National Artificial Intelligence Research Resource Task Force.
Creating a National AI Community College Consortium, including Houston Community College, Miami Dade College, and Maricopa County Community Colleges, to expand the diversity of students who have access to AI tools.
Hosting and organizing educational events that have informed hundreds of lawmakers and policymakers about the latest advances in AI technology.
Our Recommendations to Facilitate Innovation and Make it Work for Everyone
Based on these engagements with community leaders and feedback from our programming, I would like to offer three general recommendations that I believe are essential not only for the United States to fully benefit from the transformational potential of AI but also to ensure that as many Americans as possible benefit from these advances. These ideas are interconnected and complementary, and collectively they offer a blueprint for steps that national and local policymakers can take to ensure the development of novel and useful AI applications.
First, our programs reveal the importance of engaging directly with communities, including by bringing AI tools and training directly to communities that might not otherwise have access and by seeking out and listening to their perspectives. By exposing populations that are traditionally underrepresented in the technology workforce to powerful new AI tools along with partners and coaches to leverage them, we can build interest in and provide pathways into high-paying jobs, ultimately building out a more robust, more equitable workforce. In addition, AI has significant potential to help solve local as well as national problems, but only if we make efforts to involve community leaders in AI development, provide them with the knowledge and resources to develop potential local-serving applications, and connect them with the investment resources needed to implement and scale their ideas. SeedAI plans to operationalize this at our next Hack the Future event in the historic Greenwood neighborhood of Tulsa, where we will combine red-teaming exercises with experiential real-world use case exploration in key areas of Black life, and connect promising applications with local funders and business experts.
Second, we must reduce the geographic concentration of AI resources and talent, and the costs of making AI applications, by creating a federated national infrastructure to support AI research and development. SeedAI has long espoused the approach taken by the National AI Research Resource (NAIRR), and we hosted a launch event for the release of the NAIRR Task Force Final Report unanimously advocating for creation of “a shared research infrastructure that would provide AI researchers and students with significantly expanded access to computational resources, high-quality data, educational tools, and user support,” including “a federated mix of computational and data resources, testbeds, software and testing tools, and user support services via an integrated portal.”
I cannot overstate the importance of creating and fully funding a NAIRR infrastructure. Currently, AI talent is concentrated in a handful of companies primarily located in Silicon Valley, and many talented researchers and entrepreneurs do not have access to the basic infrastructure and resources they need to develop and test their ideas. The NAIRR can also act as a space to explore the myriad areas of research required for a robust AI governance regime.
As legislation emerges from these Insight Forums, open questions that persist in assessment can instead become opportunities – ways for communities around the country to explore the risks and benefits to their interests that can help inform larger policy discussions. Much of the critical work in understanding advanced AI systems will likely fall to the emerging third-party auditing ecosystem, and so having a place to aggregate and make useful the information that they find, transparency information from technology providers, and information discovered in government fact-finding such that it can be operationalized is critical. We envision an entity – a National AI Advisory Committee 2.0 – that could facilitate the development of this ecosystem by pulling in a broader group of community participants to help guide and direct the interplay between regulatory and research efforts, ensuring that the rules of the road are grounded in a robust fact pattern and practically implementable.
The NAIRR can also support structured access to AI models and novel secure information-sharing arrangements that can help drive advances in AI safety and governance. For example, it might facilitate confidential information sharing between government agencies, research institutions, and AI companies. Also, should the risk of public release of open-source model weights outweigh the benefits, the NAIRR can be a semi-public environment for research to proactively address concerns about safety and adversarial fine-tuning.
SeedAI supports full funding, as laid out in the NAIRR final report, of $2.6 billion over an initial six-year period, although we consider that likely to understate the final need given the rapid development of generative AI tools and significant increases in overall AI venture capital investment. We believe that full funding should be provided upfront and outside the normal appropriations process, in the same manner as semiconductor and other funding provided in the CHIPS and Science Act. We believe that a fully operationalized NAIRR, dollar for dollar, is likely to match the economic and national security benefits provided in the CHIPS Act, making this upfront investment more than worthwhile.
In the meantime, SeedAI is taking additional actions to support interest in the NAIRR. In particular, we are in early discussions about the possibility of opening an innovation lab that would pilot the benefits of a federated research infrastructure and provide tangible demonstrations of its value. We also encourage all government agencies, private corporations, and philanthropic organizations to support “grand challenges” to frontload the work of a NAIRR, either through existing statutory authority or coalition efforts, and begin supporting the critical research needs in AI in both mitigating risks and manifesting benefits. We also encourage policymakers to expand legal authorities and provide additional funding to support challenge implementation.
Third, we must create open, affordable, and impactful platforms to help expose more people from traditionally underrepresented backgrounds to AI technology, to broaden the base of potential entrepreneurs, expose students and others to new employment pathways, and provide the public with the educational background to productively engage on AI issues and policy. Again, SeedAI has taken important steps towards creating these tools, working with our partners at AI Village and Black Tech Street. In particular, we have now hosted multiple AI red-teaming events, which have allowed a broad range of participants to engage directly with advanced AI models and surface potential vulnerabilities that can directly inform safety work at AI labs. Our event at DEFCON featured thousands of participants, and via the Hack the Future effort, we are taking those learnings and creating events in many locations this coming year with a deeper focus on communities.
While we have partnered with a variety of leading AI companies, we believe this approach is highly scalable if it centers on open-source generative AI models, uses a cloud-agnostic backend, and is developed within an open-source framework. This will allow the development of online red-teaming experiences, experiential real-world use case exploration, crowdsourcing of educational modules built atop the platform, and the creation of uniquely tailored products at universities and communities across the country. Our partners and SeedAI are taking exactly this approach in building upon our work at DEFCON.
We believe that this program can also bring significant added value by focusing the open source community’s energy on competitively ensuring AI systems adhere to safety and value-system controls. We envision a scenario where the efforts to red team models nationwide drive the creation of leaderboards and communities that can more fully capture the range of Americans’ concerns around how safe or value-aligned these systems are. The open source community has tremendous value to add here; we need to make sure that we enable their nobler ambitions.
Conclusion
Over the past year, with the release of chat-based generative AI tools, artificial intelligence has burst into the public consciousness with a force and speed that has surprised even those of us who have worked on AI for many years. I am heartened to see novel efforts like these Insight Forums where policymakers, both in the United States and abroad, can concertedly engage on issues of AI safety, governance, and innovation.
However, I remain concerned that without appropriate interventions, AI will work like previous technologies and concentrate wealth and opportunity with a select few. That is why I have outlined three ways in which policymakers can help open up access to AI tools and therefore drive innovation deeper into communities across the United States. Broadening the set of entrepreneurs working with AI tools is important for ensuring AI develops in a maximally helpful way. By providing secure, easily accessible infrastructure, directing energy towards solving local as well as national problems, and allowing more diverse voices to influence how AI is deployed and used, we can fully leverage the pro-social potential of this transformative technology.