News Release

Social intelligence is the next frontier for AI, researchers say

Peer-Reviewed Publication

Tsinghua University Press

Artificial social intelligence

image: Three inextricably linked aspects of social intelligence — social perception, Theory of Mind and social interaction — are the cognitive tools that will help computational science to advance artificial intelligence beyond contemporary models. view more 

Credit: CAAI Artificial Intelligence Research, Tsinghua University Press

Siri and Google Assistant may be able to schedule meetings on request, but they don’t have the social understanding to independently prioritize the appointments — yet. According to researchers based in China, artificial intelligence (AI) may be smart, but it is stunted by a lack of social skills.  


They published their review of the current state and call for future directions on March 10 in CAAI Artificial Intelligence Research.


“Artificial intelligence has changed our society and our daily life,” said first author Lifeng Fan, National Key Laboratory of General Artificial Intelligence, Beijing Institute for General Artificial Intelligence (BIGAI). “What is the next important challenge for AI in the future? We argue that Artificial Social Intelligence (ASI) is the next big frontier.”


ASI, the researchers said, comprises multiple siloed subfields, including social perception, Theory of Mind — the understanding that others think from their own point of view — and social interaction. By using cognitive science and computational modeling to identify the gap between AI systems and human social intelligence, as well as current issues and future directions, Fan said the field will be better equipped to advance.


“ASI is distinct and challenging compared to our physical understanding of the work; it is highly context-dependent,” Fan said. “Here, context could be as large as culture and common sense or as little as two friends’ shared experience. This unique challenge prohibits standard algorithms from tackling ASI problems in real-world environments, which are frequently complex, ambiguous, dynamic, stochastic, partially observable and multi-agent.”


As such, Fan said, ASI requires a comprehensive approach, since improving specific components of an ASI system may not always result in improved performance — unlike contemporary AI systems. Rather, ASI requires the ability to interpret latent social cues, such as eye-rolling or yawning, to understand other agents’ mental states, such as belief and intent, and to cooperate in a shared task.


“Multidisciplinary research informs and inspires the study of ASI: Studying human social intelligence provides insight into the foundation, curriculum, points of comparison and benchmarks required to develop ASI with human-like characteristics,” Fan said. “We concentrate on the three most important and inextricably linked aspects of social intelligence: social perception, Theory of Mind and social interaction, because they are grounded in well-established cognitive science theories and are readily available tools for developing computational models in these areas.”


According to Fan, the best approach is a more holistic one, mimicking how humans interface with one another and the world around them. This requires an open-ended and interactive environment, as well as consideration for how to introduce better human-like biases into ASI models.


“To accelerate the future progress of ASI, we recommend taking a more holistic approach just as humans do, to utilize different learning methods such as lifelong learning, multi-task learning, one-/few-shot learning, meta-learning, etc.,” Fan said. “We need to define new problems, create new environments and datasets, set up new evaluation protocols, and build new computational models. The ultimate goal is to equip AI with high-level ASI and lift human well-being with the help of Artificial Social Intelligence.” 


Other contributors include Manjie Xu, Zhihao Cao and Song-Chun Zhu, with BIGAI; and Yixin Zhu, Peking University’s Institute for Artificial Intelligence. Xu is also affiliated with the Beijing Institute of Technology’s School of Computer Science and Technology; Cao and Zhu with Tsinghua University’s Department of Automation; and Zhu with Peking University’s Institute for Artificial Intelligence.


The National Key R&D Program of China and the Beijing Nova Program supported this work.




About CAAI Artificial Intelligence Reseach 


CAAI Artificial Intelligence Reseach is a peer-reviewed journal jointly sponsored by Chinese Association for Artificial Intelligence (CAAI) and Tsinghua University. The journal aims to reflect the state-of-the-art achievement in the field of artificial intelligence and its application, including knowledge intelligence, perceptual intelligence, machine learning, behavioral intelligence, brain and cognition, and AI chips and applications, etc. Original research and review articles from all over the world are welcome for rigorous peer-review and professional publishing support.


About SciOpen 


SciOpen is a professional open access resource for discovery of scientific and technical content published by the Tsinghua University Press and its publishing partners, providing the scholarly publishing community with innovative technology and market-leading capabilities. SciOpen provides end-to-end services across manuscript submission, peer review, content hosting, analytics, and identity management and expert advice to ensure each journal’s development by offering a range of options across all functions as Journal Layout, Production Services, Editorial Services, Marketing and Promotions, Online Functionality, etc. By digitalizing the publishing process, SciOpen widens the reach, deepens the impact, and accelerates the exchange of ideas.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.