OpenAI’s non-profit arm has revealed its plans to tackle “humanity’s hardest problems.”
The OpenAI Foundation has announced a sweeping range of investment and research goals, from building safeguards around how AI behaves in the wild to pushing for shared data ecosystems and funding disease research.
The stated roadmap could serve as a blueprint for how other organizations approach model development and safety.
“We are still at the beginning of what AI can make possible,” Bret Taylor, chair of the OpenAI Foundation’s board of directors, wrote in a blog post today. “The opportunity — and responsibility — is to make sure these technologies lead to real progress for people.”
Establishing ‘AI resiliency’
OpenAI completed its recapitalization last October, with the OpenAI Foundation now holding equity in the for-profit OpenAI business and pledging to invest $25 billion into AI research, $1 billion over the coming year.
The Foundation is focusing on “two dimensions”: AI’s potential significant benefits in the way people work, learn, and access medical care; and, on the flip side, the challenges that are already surfacing as AI becomes more and more capable.
The non-profit calls the latter ‘AI resilience.’
“What OpenAI means is, will AI continue to reach the goal that humans meant for it, or will it fail in some way?” explained Brian Jackson, principal research director at Info-Tech Research Group. “’Will there be errors that cause us to fall short of the goal? Or, even worse, will there be safety considerations that lead us in a bad direction and we end up causing harm when we intend to help?’”
In promoting AI resilience, the Foundation says it will target three areas where concerns are “already apparent”: Biosecurity, model safety, and impact on children and youth.
In biosecurity, the non-profit will work to improve detection, prevention, and mitigation of potential biological threats, both “naturally occurring and AI-enabled.” It also will support independent testing and evaluation, develop “new and stronger” industry standards, and fund research to help detect AI model safety issues early or avoid them altogether.
To address AI’s impact on children and youth, the Foundation says it will invest in data-driven research and evaluation, and identify and build safeguards to promote “beneficial interactions.”
One of “humanity’s hardest problems” that the Foundation is targeting first is life sciences and curing diseases.
“AI has enormous potential to speed up scientific and medical progress to save and improve lives,” Taylor writes. “Researchers are [already] using AI to better understand diseases, explore new ways to prevent and treat them, and move ideas from the lab to patients faster.”
Three initial focus areas will include:
- Public data for health: Underscoring the importance of sharing scientific data, the Foundation will help partners create and build out “open, high-quality datasets,” and potentially open ones that were previously closed, to support medical research projects.
- AI for Alzheimer’s: Alzheimer’s is “one of the toughest problems in medicine,” Taylor notes, but AI’s ability to reason across complex data could help researchers derive new insights. The Foundation will partner with research institutions to map disease pathways, detect biomarkers for clinical trials and care, and speed up personalized treatment. Work could potentially include repurposing existing FDA-approved molecules.
- Accelerating progress on high-mortality and high-burden diseases: AI can help lower the cost and risk of developing or repurposing therapies for diseases where research is under-funded. The Foundation will bring together AI researchers and disease experts to identify ways to support scientists with AI tools.
“What I like about this OpenAI statement is they gave us some really specific areas where they think they’re going to make a difference,” Jackson said.
These specific medical programs are “highly-complex and specialized,” he noted; there’s a ton of data from millions of the afflicted, but limited ability to reason about that data, parse it, and make sense of it to derive insights on different treatment options.
“They see a lot of data. They see that clear limitation,” Jackson noted. “That’s a good lesson that can be applied from an enterprise point of view: Where do you have a lot of data but not a lot of bandwidth to do the reasoning?”
Prioritizing AI projects, sharing data wisely
There are a few important takeaways for enterprises, Jackson said, notably, OpenAI is creating an example for other organizations struggling to prioritize AI projects. They should focus on the areas where they can make the greatest impact.
“A lot of enterprises are still saying, ‘Where do we get our return on investment in AI, and how do we organize to make sure that we’re focused on leveraging AI for the best possible value?’” he noted.
IT leaders can also view this as a framework for safe AI practices. Jackson pointed to the way the Foundation is looking to build a “data culture” and share domain expertise for “the most shareable and valuable data that you can get your hands on.”
“What’s the best way for all these different research efforts to surface their data so that AI can learn from it?” he asked.
It’s an important development that enterprises can learn from, particularly in legacy environments that still hold a ‘don’t share your data’ mindset, Jackson noted. Enterprises are still siloed or protective of their data, and often don’t look at it holistically, even across internal departments.
However, “when we put our different domain expertise together, we create value that’s greater than the sum of its parts,” he said. The Foundation’s mindset is, “we’re creating a safe ecosystem where we can share data and we can all benefit from it.”
