Making the internet accessible with AI web agents
Researchers at The Ohio State University are working with artificial
intelligence to try to make the Internet more accessible for people with
They’ve developed a “web agent” that helps people navigate complex
websites using simple language commands.
The team’s research was presented last month at the 37th Conference on
Neural Information Processing Systems.
Here to talk to me about their work is Yu Su, a co-author of the study
and the distinguished assistant professor of computer science and
engineering at Ohio State. Yu, thanks for speaking with me.
Matthew Rand: Okay, first of all, let's help us understand what this AI web agent is and what it's like to interact with it.
Yu Su: Sure, the Internet is such an integral part of our work and life right now, but it's not equally accessible to everyone.
So with these web agents, what we want to achieve is that you can just go to any website and just say whatever you want to do, like book a flight, order some delivery, or even book a car knowledge test on DMV. Then the agent will understand what you said, understand this website, and just do it for you. That's essentially the goal.
Matthew Rand: Why are such AI agents needed, and who could benefit the most from this technology?
Yu Su: Yeah. So on the one hand, people with disabilities, including visually impaired people with physical disabilities, can definitely benefit from this to make web access much easier, but also people who are less tech-savvy.
The Internet could be a bit scary for our parents, for example. And on the other hand, the modern Internet just becomes so complex. So even for the tech-savvy people like us, it could still be hard to find the information we need on some compact website. So this web agent can help us all.
Matthew Rand: It seems like there's been just an explosion in AI technology in the last several months. Why? Is right now really the right time to tackle this problem?
Yu Su: Yeah, absolutely. So we have been interested in doing something like this for years for the people at the Ohio State, especially the national language processing group that I'm co-directing.
But last year was the best time to do this because of these larger language models (LLMs) like Chat GPT, which I'm sure everyone has heard of.
So compared with the earlier AI technologies, these LLMs, they show like a broad understanding of language, so it can understand all the websites out there and also some more deeper understanding, like it can understand whatever the user is saying, as well as these websites, like the HTML code of this website. So that makes it the perfect timing to tackle something like this.
Matthew Rand: Finally, are there any ethical or safety risks associated with this technology?
Yu Su: Absolutely. More powerful technologies always brings with its own new safety concerns. Web agents are, I would say particularly so because now you have these autonomous agents that can just do all kinds of things on the Internet.
And we know that humans can do literally everything on the Internet because the Internet is such a rich space. So we have to be very careful when deploying these agents. We have taken a lot of measures to counter these potential safety risks, but we need to do more.
Matthew Rand: I've been speaking with OSU Professor Yu Su about his team's research into using AI to make the Internet more accessible for people with disabilities. Thanks again.
Yu Su: Thanks for having me.