[Guy Bate and his colleagues at the University of Auckland are exploring the use of presence-evoking AI-based assistants to serve as supplementary teachers in the Graduate School of Management, as reported in this story from New Zealand’s The Post. See the original version of the story for five more pictures and a 1:24 minute video. –Matthew]

[Image: A University of Auckland teaching avatar. Credit: Simon Morrow / Stuff]

Forget imaginary friends, how about an imaginary teacher?

By Kevin Norquay
August 17, 2024

Wish you always had a friendly face close at hand, to help with tricky problems? Well, the University of Auckland has an impending deal for you, as it develops online “humans” to chat to and help students.

If only we could clone ourselves, we’d get so much more done, the lament goes. Now, with artificial Intelligence (AI), that day seems close at hand, with Dr Guy Bate and team having an array of AI assistants all but ready to go.

They are not AI replicas of lecturers or tutors, nor are they intended to replace humans; they’re there to offer “face-to-face” academic knowledge, the Graduate School of Management project director tells the Sunday Star-Times.

If the experiment works, a student will be able to converse with an avatar at any time of the day, in a way intended to enhance their learning experience.

It’s cutting edge, as the world heads swiftly in that direction.

“It’s not in any way to replace the time students might have with real people. It’s times when we literally cannot be with them,” Bate says.

“We have limited face time, so this provides a different platform that can test and challenge those ideas when they’re doing their own private study.

“They’re trying to challenge their own knowledge. It’s not a mobile textbook, that’s not what these things are designed for.”

Companies in China are creating digital avatars of real people using generative AI, while Instagram parent Meta allows users in the United States to create AI avatars or characters.

Creators and business owners will be able to use them to interact with their fans or customers online on their behalf, Meta says.

Companies now offer realistic chatbots of dead people, so you can talk to a deceased relative, as if they are alive. Whether that is a good thing is under debate, as there are fears it would disrupt natural grief.

Anyone online can interview AI models that look like former US president Abraham Lincoln, Anne Frank, the diary-keeping WW2 Jewish girl, or Greek philosopher Socrates.

But it’s rudimentary, so don’t expect the brilliance of the Gettysburg Address or human emotion above “it was scary when the Nazis found me”.

As the front end Bate’s team uses Soul Machines, a company that develops AI assistants you can interact with as if they are a real person. Bate chooses an avatar, then pours in their knowledge base.

“You’re basically saying ‘You are this person. Your role is to do the following. This is how you’ll behave.’ You tell the avatar what it is, what it’s an expert in, how it behaves, how it interacts,” Bate says.

“You’re not telling it what it knows, because it’s plugged into the generative AI platform – Open AI, in this case – to construct that knowledge.

“By giving the person a style of university lecturer, a style of a business expert, a style of a marketing communications expert, you hone the way the large language model is interrogated.

“You put a skew on it. So basically you can try to limit nonsensical answers, because you’ve already given it a persona almost which allows it to skew the way it uses the large language model.”

So a student could talk to an avatar who has – for example – worked for 20 years within the technology sector, leading strategy, new product planning, business transformation, and commercial operations.

“What are the crucial elements to get right in a bioscience company start-up?” a student might ask it at 1am, and the avatar will respond with no signs of being annoyingly woken up.

“So you’re actually having a discussion with a technology expert … I’ve got this idea, what do you think?

“[Students will be] getting push-back and trying to address questions … It’s coaching students to have a dialogue with a simulated external person that they want information out of.

“We’re way off that yet. All these things are research-informed, we’re doing test cases, trialling them, seeing what happens. There’s a whole host of opportunities to deploy them in the future.”

Instead of typing in an AI request and getting a typed response, you speak to an avatar and get a verbal response. The intent is for student learning to be enhanced through better engagement, though whether that will happen has not been tested.

“We don’t know yet,” Bate says.

“That’s one of the things we want to try to work out, and actually do a study … so we can determine in which cases does it help, in which cases is it just easier for someone to type into to do it?

“We’re not committed to using this at the moment. We’re testing it out and trialling it.”

The University of Auckland is in good company: Soul Machines assistants are used by the likes of Google, Microsoft, Amazon, Heineken and the World Health Organisation.

Originally trained in the biomedical sciences, Bate has a 1998 PhD in cellular and molecular physiology from the University of Liverpool.

He recently completed a second PhD, his thesis looking into the use of digital technologies to support strategy implementation across a multinational enterprise.

With that background, and those of others in the department, he can check his “assistants” are giving correct answers, and tinker so that they do a better job.

Bate has tested the avatars on students, and the validity of their answers on himself. It’s fair to say that so far, he’s been impressed.

“The answers I’m getting from the avatar are excellent, very engaging, very supportive,” he says.

“We’ve been testing it with each other as well, lecturers quizzing it. We wouldn’t even put it out there, unless it was tested by ourselves … Is it coming with answers that are correct? There’s always that caution.”

Open AI – large language models that learn as they go along – emerged only in the past two years. Now they are being hailed as a technological revolution, so universities intend to make the most of the opportunities, Bate says.

“It’s fascinating, It’s an emerging field, and we’re kind of working it out as we go along, just like everyone else is with ChatGPT, with any AI.

“It does things that we haven’t done before, so we’re working out how best to incorporate these things. The key thing … is making sure we have the right ethical frameworks in place to do this.

“We have ethical principles in place for teaching so we know how we want to frame these things, but there’s always caution. That’s why there’s no intent to replace formal teaching.

“So it’s thinking about the use cases and really, really carefully asking, how is this going to add value? How is it going to meet learning outcomes that are prescribed for the course and the programme?”