Chatbots Demystified – Michelle Zhou, Juji

This is Techstrong tv. Hi, I am Bonnie Schneider. Welcome to techstrong tv.

Joining me now is Dr. Michelle Zo, co-founder and c e o of jui here to debunk some common myths surrounding chatbot creation and management. Dr.

Zoe, so great to have you here. Hi, Bonnie. It's nice to have to be here.

Thank you for having me. Great. Well, I'd like to start with a little bit about your, your own background and if you could also tell us about jui and its role in AI and chatbot and chatbots.

Sure. Uh, I'm a co-founder and the c e o of Jui, uh, which is a AI company located in Silicon Valley. And before I started at Jui, I have worked at I B M Research, I b m Watson Group for almost 15 years.

So by training, I'm a computer scientist, so my area of expertise has always been in the area now as human-centered ai, which intersects artificial intelligence and human computer interaction. So my whole career has been trying to at least make machines smart to, uh, help human computer interaction to advance human computer interaction. So the GG today we have been focusing on teaching machines, human advanced skills, for example, active listening and people reading because we want to make machines the best partners, the best teammates of human beings.

That's fascinating. And, and so on point since everyone's talking about ai, uh, right now, but, um, you know, there's a lot of misconceptions about chatbots and how they work and how they respond and, and how they're managed. Can you debunk some of those myths for us?

Yeah, I think the two, um, most common myths will be number one is, um, everybody can make a chat bot. Making a chat bot is very easy, especially with, um, the power of a chat, g p t, it's so called a large language model, which is a really technology behind the chat G p D, but actually making a chat bott, it's much harder than everybody probably has thought of. Just for example, let's say there are typically three types of a chat bot.

So one type of a chat bot people have often interact with, we call the two a chat, which means the user ask the question and the chat bot will answer the question, the thinking about this way. So what do you fool? You just use chat g PT to help answer the questions about your company, about your organization.

But chat g PT is not gonna listen to you because the use the content and the material it was trained for and not going to use your material. So you're thinking about you have to retrain a chat bot to use your proprietary content. Then what about security?

So what if somebody hacked you into your chat bot, make your chat bot say something really bad about your organization? Right? Then what about performance?

Let's say you have hundreds, thousands of users are using your chat bot. How you gonna manage that? So it's not really easy.

And now we go into the enterprise. You wanna create a chat bot that's actually can perform multipurpose. So let's say in the educational world, you want the chat bot not just answer students questions, but actually ask students what they care about, how they feel about their programs.

Their progress now becomes multi-purpose chat bot. That will require even much more efforts, even sometimes what I call the cross organizational efforts, right? For example, marketing has to be there, maybe sales has to be there, recruitment team has to been there, maybe the IT team has to be there.

So it's not an easy task really for any organization. Just think about, yeah, creating a chat bot. It's just a piece of cake.

So it's number one, myth number two method is now with the chat, g p d, with the large language models, the chat bot can do everything for us. No, that's not true. Actually, in reality, a chat bot, uh, can we call it, uh, hallucinates, which means it is, I just saw it actually in a, on LinkedIn article talking about, and, uh, in a medical domain, right?

Asks a chat bot, uh, chat G b t to write, um, particular medical article, and it did, it did, uh, write article. But you know what, it made up all the references, which means the reference other, uh, medical use cases and, uh, medical discoveries actually nonexistent. Mm-hmm.

That's why we called, we called it, uh, the AI start to, uh, hallucinate, right? So that's for your organization, and then you really have the liability to give the accurate and the correct answers to your audience. So not easy, not easy task.

I like the way you said AI in the chat. Bott, hallucinated. That's, that's a nice way to say it.

Um, because I, I, that's actually come up in a lot of professions where people are, are leaning in and, and they find out that a case was cited like for an attorney that that didn't exist. So you're right, uh, you have to be on top of, of your chat bots, but can you differentiate, um, about chat bots? What are the different types and the different capabilities that they have?

Yeah, so, uh, I would say, um, uh, there's three major types. So the first type, as I mentioned earlier, we are all familiar with like a chat G p T. It's about a very straightforward question and answer, right?

A user's, humans ask a question, and the AI chat bot in this case answers the question. Second type is another way around it. The chat bot will ask the question and the humans will answer.

So the use case for this one is very interesting. So for example, like a conversational survey or maybe like the interview, job interview process, you can have the AI ask questions and the job applicants answer the questions, right? So the third type is what I call the multi-purpose chat bots, which means the chat bot just behaves like a human, like we are having a conversation, and the conversation needs to be two-way conversation, which means the chat bot needs to ask the questions as well as answer questions.

So that's, um, kind of a make it a more humanlike, make it really the conversational or, um, manner in a very conversational way, right? So to make the engagement, uh, uh, much more advanced and make the, uh, maybe the human interaction also much more engaging, those three types. Yeah, I mean, and that's key for, for anyone that's on the other side of the chatbot as well.

Um, you know, and I think people also think it's, it's an easy thing to operate and requires minimal resources. But, uh, that may not be the case. Or is it, can you explain the operating, um, operating chatbots?

What goes behind that? Right. Uh, thank you for asking this actually, uh, uh, just not very long ago, maybe a month ago, an organization and, um, uh, not just one organization, many organizations came to us to say that, you know, our C E O really wants a chat bot project because we want to leverage the AI set of art AI for our, our users, for example, employees or our customers.

So the smart CEO O would actually, CEOs would ask their personnel to do some research, do an investigation. So some people really smart, they went out to check what, what will be required to actually create build as well as maintain a chat bot, right? Because AI is not perfect.

You have to keep the maintenance, uh, there. So in that case, it is the comes sense. 5 million from scratch, right?

Because you need AI experts, machine learning experts. You also need system experts who can actually put up and actually make sure your system is robust, it's secure. You also need a user experience expert because your chat bots has to interact with other human beings.

The user experience side needs to be there, and of course you have to have the operating expense on hosting it, right? And make sure it's a security and it's robust and it's secure, it's robust, and also has the performance as well. So it's actually, it's a very, um, expensive operation.

And not to mention, as I always said, that, uh, when you adopt a chat bot or AI solution, it's like adopting a child. You can't just abandon it and somebody has to maintain it, right? And who is gonna maintain it?

It or your domain experts. So most likely your domain experts need to fade the chat bot with updating knowledge, but if you don't have the right tools to do it, they have the route to the IT and it has to help them to do that. And thinking about that across organizational effort and cost, right?

So that's why, uh, we actually, um, uh, uh, collaborating with our clients to help them analyze, uh, when you wanted to do it on your own, uh, maybe when do you want decide to partner with somebody else? So, which means that it's almost like, uh, my co-founder has a really great analogy. So it's thinking about the charge G P D or let's large language models.

It's a one type of a very powerful engine or CPU u, right? But, uh, even with this powerful CPU engine, not everybody really has a skill to create a computer to build a computer from scratch. So you want to think about it, whether you want to do it on your own or you want to partner with somebody to help you to create that very powerful computer.

Cause not everybody maybe wants to do that. In your experience, um, you know, you mentioned all the different aspects of human oversight, but I was interested in what you said about user experience, um, how much of that shapes the way, um, the chatbot is put together and adjusted, and has anything surprised you in that process? Absolutely.

Thank you so much for asking this question. It's a great question too. So, traditionally, because the chat bot, the conversational ai, Tim from the IT world, right?

So it, people always care about the models, the machine learning, but less so on the user experiences. Two types of user experiences. One type of user experience is the people who actually teach the chat bot.

So think about in the organization, if you wanna create a recruiting chat bot, do you think the it should do it? Or actually the recruiting experts should do it, right? But in the past, the recruiting experts normally cannot do it because they are not computer scientists, they're not AI experts.

They don't know how to write programs to do that, right? So they rely on it. That creates a level, another level of, uh, difficulty from, uh, organizational coordination and from even knowledge transfer, right?

So because it may not have the knowledge and more, as I mentioned the earlier, when you have a chat bot, you have to keep the knowledge up to date. Again, the domain experts have those know, have this knowledge in their mind. They are, they should be the people who are going to maintain, nurture and, uh, continuously update the chat bot.

If they don't have the right tools to do that. They couldn't. So ma in many cases, the chat bots got abandoned.

So I said, almost like you adopt somebody, a child that you abandon it and the chat bot just is not gonna function very well, right? And the, the users will abandon them. So another side of the equation is every chat bot is supposed to engage with the, what we call end users, but the end users are not, no, or not computer scientists have a PhD or have even degree in computer science.

So they may not know what chatbot can do and cannot do, right? So the designer has to really put this, uh, what I call the teaching the chatbot, teaching the users how to use it. That's why you've heard about, even with chat pt, such a powerful tool probably you heard about this word cause the prompt engineering, right?

Mm-hmm. Prompt engineering means it is, uh, you have to know how to issue your request. So the chat by chat G P T and gave you the correct response why it's called the prompt engineering not called the prompt experience because, uh, it does require the knowledge of the engineering, uh, knowledge of understanding how the generative AI actually works.

So that's not really good for the masses. That's why we are also very, very much caring about how to democratize this kind of technologies. So other people do not worry about prompt engineering.

That makes sense. You know, I think, um, if people have different experiences with chatbots, um, whether they're using them or, or creating them, but I'm just wondering if you think that, um, people get frustrated and give up when they realize this, this human centered element that you're talking about, isn't there? Um, how, how do you get people to kind of lean into chatbots and, and not reject it immediately?

I think we started out talking about some of the myths, but, but that seems to be one too, that there could be a negative connotation to it. How do, how do you flip the script on that? Thank you again for asking this one.

So my formula for that, it's very simple. I code it U square, U square, it's always this principle I use have used in my work to construct, to develop what I coded the ultimate human computer interaction systems, right? So think about chat bot is also just the, uh, a typical human computer interaction interface.

So in that case, it is the U square. One of the U is you have to make sure your chat bott is actually useful to your users, right? So for example, uh, like where are working with the healthcare companies so that if a chat bot is trying to supply information and provide the different types of, uh, encouragement to the patients, a better does that very well, otherwise the users would just go away because you're not helping me, right?

So that's why the first you will be usefulness. The second, uh, uh, you would be usability again, even though I knew this chat bot probably can help me, but doesn't matter how I say it, it doesn't matter how I interact with the chat bot doesn't understand me, that's going not going to be a complete turnoff. So in that case, it is.

So whenever our organization is going to start a chat bot, a project or solution really needs to think about it is how the chat bott will help the users solve a problem or give out the solution. And the second part of it, how the users are going to use it. And it's better to be very easy, very intuitive, and again, very helpful.

Well, um, for our final question, I was wondering if you could share your vision for the future of chatbots. Um, what, what do you see as their role going forward in business and society? I think in my, uh, view of this chat bots is just, uh, one type of a format, or shell, if you call right, with for intelligent, intelligent machines.

So could be in the form of a chat bot, could be in the form of a physical robot, and it could be even in the form of your, uh, your kids' toys or device, right? So just chat by just one type of form. So my ultimate vision, my really ultimate hope and my vision is, so I'm very, uh, a big believer with the human machines symbiosis, which means it is, I believe those machines doesn't matter which form they come in, will become our really companion, our teammates.

So they understand us very well, brought me better than we understand ourselves because of, uh, objective, because of the, uh, the empathetic side of it. And then use those information, those insights about ourselves to help us. So I really envision a world that not very far from now, probably five, 10 years down the road, that every one of us will have our own AI companion, will really understand our temperaments, our unspoken needs, and wants our personality and help us in every aspect.

So thinking about the movie her Right, but it's a much nicer, much kinder, much more helpful her for everyone. That is really fascinating. I don't know if that's gonna replace friends and spouses.

Hopefully not. Maybe just an, an extra in your life, an extra, uh, chatbot that understands you. Thank you so much, Dr.

Michelle Zo, co-founder and c e o of Jui. It was a fascinating conversation. Thank you so much for your time today.

Thank you, Bonnie. Great. Well stay with us on Techstrong tv.

We're going to have a lot more of these great interviews coming up.