As artificial intelligence (AI) increasingly becomes a part of our lives, it has also begun affecting our productivity and work. An issue with such new technologies, though, is that they are often difficult to understand, leaving us unable to use them to their full potential.
To address this issue and better prepare students to interact with AI, the University of Toronto will be partnering with Naver, a South Korean technology company, and its Toronto subsidiary, Wattpad. Back in early 2021, Naver purchased Wattpad from Allen Lau and Ivan Yuen, two alumni of the University of Toronto, for roughly 600 million USD.
The research partnership involves AI and improved interaction between computers and humans. U of T News reported that many graduate and postdoctoral students will receive training as part of this research project. U of T’s Frank Rudzicz, a researcher and associate professor of computer science, and Anastasia Kuzminykh, an associate professor in human-computer interaction, will act as two of the six leaders for the research projects. Rudzicz’s work will develop processes that will improve artificial neural networks to increase computer understanding of human language, and Kuzminykh’s research seeks to better understand how humans perceive and interact with conversational agents like chatbots and virtual assistants. Wattpad hopes these projects will improve its user-based services.
In an email to The Varsity, Kuzminykh wrote that collaboration between humans and computing systems could immensely improve human work, especially in situations where abundant or complicated data must be processed.
However, Kuzminykh highlighted that “the success of this collaboration is predicated on the ability of humans and AI to effectively communicate.”
A critical dimension of how users interpret AI systems is the structure and flow of conversation from the system, as enabled by user interfaces. A lack of understanding of what humans look for in conversation is a barrier in this area that the research project aims to address to allow AI models to enable smoother conversations.
According to Kuzminykh, current AI research methods cannot provide these AI models with the proper infrastructure to predict human needs in conversation, which is why collaboration is weak in its current state.
Kuzminykh wrote that a growing field, called “Explainable AI,” aims to help people understand AI. Users’ perceptions of conversation are affected by slight differences in how agents structure sentences in communication. For example, the impact of filler words such as ‘um’ and ‘like’ depends on the conversation. A user might perceive these words as less acceptable in task-oriented conversations as opposed to social-oriented ones. Similarly, in regular interactions between people, sentence structures can also change how people perceive the intentions of others. But Kuzminykh emphasized that, despite our progress in this field, “our understanding of the effects of conversation architecture in human-agent communication is still underdeveloped.”
Kuzminykh wrote that the research team hopes that their research and development will also influence people to pursue efficient and socially just models for AI.