The project, supported by a $20 million NSF grant, will guide the development of a new generation of AI assistants for use in mental and behavioral health.
The $20 million grant from the National Science Foundation will help institutional experts across the country develop artificial intelligence labs aimed at developing a new generation of AI assistants capable of trustworthy, sensitive, context-conscious interactions with people. AI chatbots are already deployed by businesses in mental and behavioral health environments. It is utmost important to ensure that the next generation of AI assistants can respond in a respectful, human-oriented way.
Professor Melanie Moses of the Faculty of Computer Science
The AI Institute on Interactions for AI Assistants (ARIA) Team, led by Brown University and includes experts from leading research institutes across the country, including the University of New Mexico. Santa Fe Institute, Colby College. Dartmouth College; New York University; Carnegie Mellon University; University of California, Berkeley. University of California, San Diego. Data and Society, Civil Society Organizations in New York.
Melanie Moses, professor of computer science in the Faculty of Engineering, and Sonia Gipson Rankin, of the Faculty of Law, will lead UNM's contributions to the project. The team builds and evaluates AI systems that understand human reasoning, respect community standards, and adhere to principles of justice.
According to Moses, the social impact of computing on society is accelerating with new forms of AI.
“While law is a way of dealing with conflict in society, it is difficult for law to keep up with the pace of rapid change in computing and AI. This project has the opportunity to use calculation methods to design reliable AI, taking into account the social and legal implications from the start,” Moses said.
Collaboration in specialties during the design and evaluation stages of AI development helps in creating safeguards that protect users. UNM contributes to justice-centric innovation, human-centric values, and the strengths that have established the interdisciplinary research relationships of projects.
Professor Sonia Gipson Rankin, UNM School of Law
“Integrating legal, computer science and various other disciplines is not only innovative, but also reliable, right reputation, and is essential to developing AI systems tailored to the public interest,” says Gipson Rankin. “In areas like mental health where trust, privacy and ethical responsibility are the most important, there is a great opportunity to design AI that truly serves people. To move forward, you need creativity, conscience, and collaboration.
Earlier this year, UNM launched a new Level 1 Grand Challenge Team It focuses on developing empirically reliable AI systems. Moses and Gipson Rankin lead the team alongside Stephanie Moore, an associate professor in organization, information and learning science. Working with faculty members across the UNM and UNM Health Sciences Center, the group is trying to fill the gap between trusted AI in theoretical models and trusted AI when trained with noisy data and deployed in real-world scenarios. Participation in the NSF-funded ARIA project will help drive UNM's work with trusted AI forwards.
Creating AI systems that can operate safely in sensitive areas like mental health care requires much more capabilities than today's most sophisticated chatbots and language models, according to Ellie Pavlick, an associate professor of computer science at Brown University, who leads ARIA collaborations.
“AI systems that interact with people who may be in a state of pain or other vulnerable situations, especially need to have a strong understanding of the world, and how the actions of the system affect that world,” Public said. “At the same time, systems need to be transparent about why they create recommendations they do to build trust with their users. Mental health is a high-stakes setting that embodies the most challenging issues facing AI today. So we're excited to tackle this and understand what it takes to make these things absolutely right.”
The work will require deep collaboration in the institution, expertise and academic fields, Pavlick said.
Aria's work includes a robust education and workforce development programme that spans K-12 students through working professionals. The ARIA team will work together Bootstrap Programthe Computer Science Curriculum developed at Brown supports evidence-based practices to build new AI curricula and training for K-12 teachers. The initiative, called the Building Bridges Summer Program, brings college and high school students across the country to ARIA campus to tackle cutting-edge AI research. These educational opportunities and materials will also be deployed in New Mexico.
According to Pavlick, the need for this work is urgent. New startups and existing companies are already developing AI apps and chatbots to support their mental health. Evidence suggests that people often rely on relationship advice and other chatbots to seek other information related to mental well-being.
“We hope that work we do based on trust, safety and responsible AI will address immediate safety concerns about these systems. For example, we will develop safety measures for responses that reinforce delusions and empathetic responses that can increase someone's distress,” Pavlick said. “A short-term solution is needed to avoid harm from already widely used systems, and we are combining it with long-term research to fix these issues that are experiencing.”
New, smarter AI systems are needed to provide the reliable, contextual feedback needed for safe and effective mental health interventions. Today's large-scale linguistic models generate text through statistical inference. This predicts the next word to use based on previous words or user input. Unlike humans, they do not have a mental model for the world around them. They do not understand the chains of causes and effects, and have little intuition about the internal states of the people they interact with.
At the same time, the team will attract legal scholars, philosophers, education experts and more to better understand how such systems fit into existing social and cultural infrastructure.
“I just don't take it for granted that there should be a system that can be built because not all systems will make a net profit,” Pavlick says. “So we'll be working on questions about which systems should be built and which should not be.”
Ultimately, Pavlick said that developing smarter and more responsible AI benefits not only in the mental health field, but also in the general AI development process.
“We're working on this important alignment question about how to ultimately build technology that is good for society,” she said. “These are very difficult issues with AIs that generally have particularly sharp use cases for mental health. By working towards answering these questions, we work to create AIs that are beneficial to everyone.”
ARIA is one of five National AI Institutes that receive a total of $100 million in funding, and in a partnership between Capital One and Intel, the National Science Foundation was announced on Tuesday, July 29th.
“Artificial intelligence is key to strengthening our workforce and enhancing our competitiveness in the United States,” said Brian Stone, who performs the role of NSF director. “Through the National Institute of AI Research, we are turning cutting-edge ideas and research into real-world solutions, preparing Americans who will lead the technology and work of the future.”
