Experts Discuss Human-AI Interaction at CAIT Research Showcase

An interdisciplinary group of experts in artificial intelligence (AI) gathered online May 8 to share their latest research findings and discuss how the rapidly advancing field is already affecting business, academia, and society. The annual Columbia Center of Artificial Intelligence Technology (CAIT) Research Showcase featured an expert panel and lightning talks given by recipients of CAIT Research Awards and PhD fellowships.

In the day’s main event — a panel discussion on the fast-moving interface between humans and artificial intelligence — a group of Columbia Engineering’s AI experts discussed the transformative role AI now plays in various sectors ranging from manufacturing to education. The discussion offered insightful perspectives on how AI is both changing the landscape of work and business while complementing human capabilities in unprecedented ways.

Humans and AI as collaborative partners

One major theme of the discussion was the importance of integrating AI into workflows where they can help humans expand their capabilities.

Lydia Chilton, assistant professor of computer science, told the audience of researchers and representatives from industry that “we’re not looking at AI as a substitute for human effort but as a complement to it.” 

“There’s a rich potential for AI and humans to collaborate on tasks that neither could achieve alone, which could lead to remarkable advancements in how we approach problems and tasks,” Chilton said.

This concept of collaboration rather than replacement could redefine job roles and lead to more effective ways of working. In her work focusing on combining human cognitive abilities with AI’s processing power, Chilton suggests a future where AI assists in amplifying human potential.

Rethinking conversational AI

Zhou Yu, associate professor of computer science, highlighted some key challenges the field must confront before AI systems are able to engage with humans in natural, useful, and wide-ranging conversations. Her focus on refining AI systems to better manage dialogues and assist in decision-making processes points to the complexity of human language and interaction that AI struggles to grasp fully. 

“The future of conversational AI lies in its ability to not just respond but to understand and initiate context-appropriate actions,” Yu said. Her work on conversational AI explores the limitations of current technologies and the potential for AI to become a more proactive participant in conversations, underscoring the need for systems that can effectively collaborate with humans.

“AI must evolve to anticipate needs and understand contexts—it's about creating systems that are not just reactive but truly interactive,” she said.

From concept to application: AI’s real-world impact

Vishal Misra, Columbia Engineering’s inaugural vice dean for computing and AI, described his approach to researching AI as “backward” from the traditional academic practice of doing fundamental research in some area before finding an interesting area and building a commercial application. 

“Three years ago, when GPT-3 was released, I built a commercial application that's been running on ESPN for the past two and a half years,” he said. “Now my research is modeling how and why these LLMs work.” 

Misra compared today’s AI chatbots to highly skilled research assistants “with early onset dementia.” 

“They live in this alternate reality where they don’t know that what they’re saying isn’t factual,” he said. “They can do reasoning, but they require a human in the loop. This idea that you will have this autonomous agent that will go and do everything for you? It’s quite a ways off.”

Countering bias

As these systems make increasingly important decisions, mitigating instances of bias and other forms of unfairness in these systems becomes paramount. 

“Bias is not one problem,” Chilton said. “It’s always creeping back in. You always have to be looking for it and fighting against it.” 

For Yu, this means a regulatory regime based on diverse “test cases” designed to reveal biases in AI systems.

“We have to make sure that a model passes all these test cases before it’s released,” she said. “This should be standardized across various industries.”

The discussion concluded with a consensus on the need for continued innovation in AI research and application, fostering a collaborative environment where AI and human intelligence coalesce to tackle the next frontier of technological challenges.

About CAIT

The mission of the Center is to better society through the development and adoption of advanced AI technology contributing to a more secure, connected, creative, sustainable, healthy and equitable humanity.

Connecting and channeling the broad strengths in Columbia Engineering and related disciplines through collaborative projects, student fellowships, and outreach programming, the Center will establish a world-class center of knowledge discovery and talent training for broad societal impact. Through a partnership with Amazon, the Center will leverage the deep insights and extensive experiences in developing robust, scalable AI systems meeting customer needs in critical applications and objectives. Such insights play an essential role in cultivating new research ideas and validating academic research results.