Our Vision for AI
Cornell is committed to building upon its existing leadership in Artificial Intelligence (AI) to shape and drive the development and practice of AI for a sustainable future. In particular, we envision an AI that is designed to engage with humans, and that sustainably serves humans and humanity.
Why Cornell
As a top-ranked leader in the growing field of AI, Cornell is uniquely positioned to lead the development and refinement of AI by leveraging its broad interdisciplinary collaborations.
We conceptualize our vision into the following three components:
AI Algorithmic Capabilities
We aim to develop new algorithmic capabilities in Learning, Reasoning, Perception, Language and Actuation, since they enable new applications for AI and new economic opportunities.
Human-AI Engagement
We aim to understand how humans and AI engage at all scales (e.g. individual, groups, society) to inform technology practice and design. This requires connecting the technical AI Algorithmic Capabilities with Ethics, Law & Policy, Social Science, Cognition and Design.
Application Impact Areas
We are bringing AI to important problems where we can draw on Cornell’s broad excellence (e.g., Society & Institutions, Autonomous Systems, Scientific Discovery, Health, Culture), ensuring impact and driving algorithm development. Sustainability (e.g. climate, society) serves as a cross-cutting set of goals and values.
A schematic overview is provided in this graphic, with more detailed descriptions of the three components in the following sections.
AI Algorithmic Capabilities
Most AI systems require all or at least multiple of the following AI Algorithmic Capabilities:
Learning
Supervised learning, unsupervised learning, reinforcement learning, causal inference, statistics, etc.
Reasoning
Planning, control, multi-agent interaction, knowledge representation, optimization, etc.
Perception
Images, video, depth cameras, sensors, bio-medical imaging, haptics, on-body sensors, etc.
Language
Discourse, sentiment, summarization, speech recognition, translation, question answering, generation, etc.
Actuation
Autonomous robot, medical device, smart city, deep fakes, recommender system, extended reality, etc.
Human-AI Engagement
AI engages with individual humans, groups, and society far beyond current computing technology. This means that AI systems do not succeed based on their algorithmic capabilities alone. We need to understand how humans interact with AI and sustainably design for humans and their needs. We posit that effective AI requires deep integration of the AI Algorithmic Capabilities with the following human-centered questions and methodologies:
Ethics, Law & Policy
AI raises new ethical and legal questions that have deep implications on the technology and its use (e.g., algorithmic bias, fairness, privacy, explanations, regulation, national security).
Social Science
AI systems act in the human world and we must understand human behavior in the context of AI (e.g., human bias, human decisions, social behavior, game theory, markets).
Design
We must carefully design AI systems in a complex space of affordances and requirements (e.g., human-robot interaction, interaction design, value-based design).
Cognition
Our understanding of natural intelligence will both inform and be informed by AI (e.g., bio-artificial analogies, cognitive psychology, brain-machine interfaces).
Application Impact Areas
The following application areas combine high potential for AI impact with existing research excellence at Cornell, which provides multiple avenues for recruiting faculty to deepen connections with AI technology:
Society & Institutions
AI is making small and increasingly big decisions affecting retail, political discourse, labor, law, education and many other areas of our society. How can AI be used to strengthen our institutions by improving effectiveness, fairness and transparency?
Autonomous Systems
AI is moving into the physical space that is shared with humans, with new applications in transportation, farming, climate, medicine, manufacturing, assistive technologies and others. How can we build highly reliable and effective physical systems that suitably share autonomy with humans?
Scientific Discovery
AI not only supports the scientific process (e.g., literature analysis), but has shown to be highly effective when optimizing complex objectives with expensive measurements. How can AI more generally accelerate the discovery of new climate interventions, materials, drugs, crops, chemoresponsive sensors, solvents and others?
Health
Newly available data (e.g., clinical, multi-omics, behavioral) and perception capabilities (e.g., imaging) can inform treatment and care. Tele-health provides new opportunities to support physicians, patients and caregivers. How can we bring AI to bear on both precision medicine and improved care?
Culture
Nascent questions have arisen around AI-augmented creativity, neural style transfer, aesthetic feature classification and others. Can AI (and its limitations in art, literature, poetry, etc.) shine new light on what may be uniquely human?
We further assert that Sustainability is an overarching goal that can inform the AI technology we build, direct the values that this technology embodies, and guide the application problems we tackle.