How Cognitive Intelligence will Reshape Chatbots with Dr. Michelle Zhou

We have heard for years that Chatbots are the future and although we have seen some significant progress in the concept and application, they are still impersonal and not as helpful as we had hoped they would be. Dr. Michelle Zhou, co-founder and CEO of Juji, Inc, as well as inventor of IBM Watson Personality Insights, joins us to talk about her belief that in order for Chatbots to be effective, they need to include cognitive intelligence that is both empathetic and responsible.

Learn more about Dr. Zhou’s work here :

https://juji.io/publications/
https://juji.io/blog/

Here is the entire transcript of the show (please excuse any transcription errors) :

Brent Csutoras:

Hello, everyone and welcome obviously to the Search Engine Journal Show. I’m Brent Csutoras. I’m joined today by a very special guest, Dr. Michelle Zhou, who’s the co-founder and CEO of Juji, Inc, as well as being the inventor of IBM Watson Personality Insights, which I think most of us will agree that we’ve heard of historically and I’m very, very, very excited to have you on the show and talk with you today. So thank you so much for joining me.

Dr. Michelle Zhou:

Thank you, Brent for the invitation. It’s just an honor to be the show. And love to talk to you, the topics… You are interesting.

Brent Csutoras:

Absolutely. So one of the things that I want to dive into is … Because, as I mentioned, we have a little bit of understanding of IBM Watson, but I don’t know how much of our audience really, fully, understands how important that was to the progression of AI and what exactly it was. So can I ask if you can just briefly, we don’t have to go too far into it, but I’d love to know what is Watson and how did you become a part of the project? And what did you specifically do on the project?

Dr. Michelle Zhou:

So Watson actually started with a very focused project. I think many Americans, probably most Americans, have watched the show called Jeopardy. It’s a game show, Jeopardy, almost on every day. So Watson actually created a AI system that can beat the human Jeopardy champions. So that’s our idea, right? So basically you got question, answering, think about this, really [inaudible 00:01:40] the questions that the Watson was built to answer.

Dr. Michelle Zhou:

So then, the project that we started to… At the beginning of the project it starts to expand to cover many aspects of the AI we call the [inaudible 00:01:55] computing which means gave the AI computers more human intelligence. So the project that I worked on, now it’s called IBM Watson Personality Insights, the idea is if we can actually take someone’s communication texts, let’s say your emails, your Twitter tweets, your Facebook posts, then the computer can automatically analyze the texts to infer your unique characteristics. For example, your values, your psychological needs, or maybe your personality traits. So that’s the project that actually I started and then became, actually, the product, the IBM Watson Personality Insights.

Brent Csutoras:

That’s really interesting. I mean, to give some background on some of the aspects of being able to do that so other people might … If you didn’t know that already, you might know that Facebook also did a really big study where they were able to calculate how accurately they could determine what you would like to see and how you would feel about it. They basically went through a whole grid where they were after 50 likes we know that you like a colleague, after 100 we know you like a friend. And it was like after 300 we know you better than your spouse, right? And it was really actually quite accurate. So there is a lot of different work that is being done to determine personality insights. So where do you see the most use of that today, if you were thinking of the personality insights, where is that used at today that people might recognize? You made the example with Jeopardy earlier.

Dr. Michelle Zhou:

Right. So Brent, you mentioned that actually many of your audiences are marketers, right?

Brent Csutoras:

Yes.

Dr. Michelle Zhou:

Who are all small business owners, so those kind of… We call the deep understanding of the each individual, this insights, the personality insights can really be used for many types of personalize the engagement and personalize the services. So, for example, when you actually ask the question to say, “Tell me more about Charles Darwin,” and you could have a very standard answer, right? But if different people have different personalities or different interests, you could actually phrase the answer in many different types of ways to personalize that answer to each individual. So that’s the personality insights can really be used then. For example, in our current work, if the conversational agent can really understand a user’s personality, it can offer us very personalized guidance. If somebody who is very ambitious and really wants to impress other people and the conversational agent, AI assistant in this case, might say this one, “I know that you’re very ambitious, you want to achieve more,” sometimes stick to the facts really can help you go further away, can go far away, right?

Dr. Michelle Zhou:

And similarly, on opposite, if the AI finds the person who is very careful, thoughtful and always afraid to make decisions or speak up, you can actually encourage that person to speak up more, to basically be more decisive. So you can see that once you knew more kind of about the person more deeper insights about the individual, the computers can help more, can give more personalized guidance or advice.

Brent Csutoras:

One of the things that i definitely want to dive into with this conversation next is kind of like the implementation of that today is a lot through chatbots, right? I mean, we see a lot of people first trying to implement this kind of bridge the gap kind of solution. I think the idea at some point is there would be a more of an entity, some sort of AI entity that would participate more. But I wonder if you’ve seen anybody… You talk about small businesses from a website standpoint being able to do that. But is there any implementation outside of chatbots today for that? I mean it makes me wonder, could we start seeing websites… We already show people information on a website based on geo targeting and based on you know cookied interest. Could we at some point see websites being rewritten and presented to people, data being presented to people in different ways based on their knowledge graph in a way?

Dr. Michelle Zhou:

So some of our clients actually working with us on this particular aspect even though I’m not sure it’s on their websites yet. So it’s based on the chat based, on the conversation, and then the chatbot can actually deliver personalized information. Just for example, let’s say if you go to the book website and then after the person chats with the chat of a certain amount of time, let’s say five or 10 minutes, the chatbot can really recommend books which are tailored to the person’s interests, passion and even personality interests, right? So that’s kind of work we’ve be doing right now. And as you mentioned something interesting, because most of the chatbot today, people really… That’s why I said we’re trying to avoid the word chatbots because it sounds like really, how do you say the bad connotation. There’s a kind of three reasons for that. First one is most of the chatbots are really the bots, they can’t chat, right? They often leave their users a very frustrating experience to be able to find anything, due to a lack of intelligence. Second part of it is that they’re impersonal. If you like the topic, we’ve been talking about this, they always give one-size-fits-all answer, right? If you go to the book website, the normal chatbot, if you’re lucky, I could understand what you’re saying.

Dr. Michelle Zhou:

Even they understanding the way they just gave you maybe all the popular books versus give the one maybe you are interesting the most. And the third part of it is this is actually a… We are very passionate about it is most of the AI, if you really want a very powerful AI solutions, it requires organization to really have, what you call it, AI expertise or a lot of financial resources. So that’s why we’ve been working on this area what we call the democratizing AI, which means to make a really powerful AI accessible to every business no matter how small the business it is. So they don’t need to hire, let’s say a world-class super AI expert team or hiring top-notch programming experts in order to actually install. So I think gradually in the next, I would say a couple of years, we will see that very powerful chatbot, very powerful AI assistants will be sitting on businesses including small businesses website and perform functions as we just did.

Brent Csutoras:

So you mentioned like, and there’s a couple points in there I’d like to kind of dive into. I personally think that it’s going to be extremely helpful because I know that for me, I have a hard time finding TV shows. I watch a TV show and I’m like, “I love this TV show.” I go and I read a review on five other TV shows I should love and I go and start all five and I don’t love them. And I’m like, “Well, what is it about this show that everybody else loved,” and it’s just based on people’s personal recommendations and it’s so few. What we can gather from people today, and this is where I think wearable glasses really excited me when Google launched their glasses, was that we literally have millions of micro data interpretations that we make. And yet so few of them are dispersed to anyone who could receive that, right? We have to take the time to say, “Oh yeah, I thought the restaurant was good but we didn’t see the fact that oh, the floors were a little dirty and made me think and the one person didn’t wear a glove,” and it’s not enough for me to go and make a review that’s like I’m angry, but we lose all of that, right?

Brent Csutoras:

And so as we start getting more data in, I’m very interested to see if that can get aggregated into these things. And you mentioned that like chatbots isn’t a great term and I think that unfortunately we’ve dealt with that a lot, right? When social media came out it got kind of skewed because everybody was looking at it as a traffic generator for digital marketing and not for actual communities, right? So do you see AI as an AI assistant as the the term that you’re trying to see more branded?

Dr. Michelle Zhou:

Definitely. Because we like this term very much it is, when you’re thinking about doesn’t matter chatbots including the functions, right? So why would you need a chatbot? So we need a chatbot is really to scale automated certain business tasks. This is extremely important for small businesses, right? Because you don’t have the resource to hire let’s say a gazillions of people. I mean especially if you have the people you don’t want to overwork them, right? You don’t want them to work 24/7 and also talking to maybe hundreds of people at the same time which were[inaudible 00:11:13] humans were incapable of doing. So in this case it is… AI is really as the, I would call it augmenter, augmenting your workforce, working with your existing workforce to do the job better, to scale your operations. That’s why assistant sounds like a very natural name to it, so it assists your business, assist your existing workforce versus replacing them.

Brent Csutoras:

And this is an interesting thing because it really starts to get into the satisfaction level of it, right? So my problem is, I don’t have a lot of faith in the systems, right? So every time I call a system if, I’m talking or if I’m chatting, I’ve got to answer a bunch of questions that I don’t want to answer because I’m already past that point, right? And there’s so many flaws in the system today, I’ll give an example. I was just with Comcast and they were going to bury a line outside because my internet was messing up, right? So they sent an email to me, a message to me that said, “We are going to bury this line and this is the information we have,” and it’s like, there is no fence. I’m like, “But I have a fence.” They’re like, there’s um watering system, but I have a watering system. So then it said, “If you need to update this information go ahead,” and I click it and it takes me to their chat system with no option to update anything. And then I chat to a tech who comes on it’s like, “Hey, can I help you,” a person, they’re like, “Yeah, I don’t know what you want to do.” And I’m like, “Well, this system doesn’t work,” right?

Brent Csutoras:

So I think I mentioned to you in a previous conversation but we were saying that this was really supposed to change the world as far as we knew it, right? Five years ago chatbots was going to be just the next thing everybody doubled down on it, it was like the same time it was chatbots and voice skills, right? Amazon skills and chatbots and FAQs and then Google rolled out a lot of the search engine having FAQs and certain type of quick answers, and so it was all about getting people what they needed faster. But what we found was, it really didn’t jump as far forward as we thought. And even the rhetoric and kind of excitement around AI assistance, when they’re called chatbots, as you mentioned, has kind of been diminished because of the experience. What do you think, and I mean I make assumptions as to why we didn’t go as fast as we wanted to go, but you being on the inside of this and having access to many other professionals like yourself who have those inside conversations, what happened with the excitement, the implementation, and why did it not quite solve that issue? Why did it not quite work, and why is it still kind of at the point where today it barely works and so forth?

Dr. Michelle Zhou:

I think the biggest issue people maybe didn’t realize, even the people who are trying to develop, or people who are using, it adopting it, there is a gap between what the humans want the chatbots to do, or AI to do, versus what this AI can do, right? So for example, you’re just talk about the example, I love to use. You go to the website, you chat. Actually in this case, you chatted with a human, right? But if you chat with a chatbot, often they force the users to go down a fixed the path and it happens, you said it.

Brent Csutoras:

Exactly.

Dr. Michelle Zhou:

They’ll take you where… And you said it, that’s not my problems. My problems are different, and you can’t interrupt that. If you interrupt it, and the chatbot is completely clueless, right? In this case, it’s really a lack of intelligence. This is why we call it our… We basically brand our chatbots with cognitive intelligence and you know why? One is called cognition [inaudible 00:15:05]. Cognitive intelligence in the human world it is our abilities to understand, to remember, to actually learn. So that’s exactly, we want to teach our AI assistants those skills. It’s a human life skills. So for example, if you said, just use your example, to understand. So now you say they ask you other things. You said, “Sorry i don’t have a fence.” So ask that AI should understand that there’s a difference and it should also then go to a different path to help you, not adapt a chatflow to you versus to force you to adapt…

Brent Csutoras:

To adapt to them, yeah exactly.

Dr. Michelle Zhou:

Second part, what do i say to remember. So let’s say you go down the path now you said that you have no fence for the next one they still need to ask you for example what the site looks like it’s a very narrow place which people can get in the repair worker or maybe it’s a white space that they can drive the trucks or maybe some equipment come in. So they still should remember the flow what other things that haven’t been asked kind of being collected, this is called remember the memory, right, which humans have naturally but many chatbots are completely clueless once it gets interrupted they can’t come back, they can’t complete the tasks. To learn. Learning is if it interacts with you, like we’re going back to the books with the one information you’re talking about, then I should know, “Oh actually Brent is very open-minded, always looking forward to new things,” then I wanted to recommend something to him, he said, I know that you always look forward to new ideas, a new trend. Here are some materials that you might be interested in, right? So that’s very different than somebody who is very careful, thoughtful and might say, “Oh no, this is a fair information for you I even made a comparison table so you can compare and contrast the different pieces of information.” So that’s the key concept between [crosstalk 00:17:06].

Brent Csutoras:

So how does that practically… So when you dive into that, how does that get accomplished? I mean obviously I think that most of our listeners can understand data flows, right? You’re basically saying, and this is part of the learning process with the existing chatbot solutions out there is that like, “Hey, you have 20 questions that people ask that you weren’t able to answer,” so you fill them in. So you’re just entering fields of data and you’re trying to chain them together in a way that makes sense. But as we’ve kind of said, even with the best intentions, there’s few examples of this really working and the only time they really work are when there are pretty standard flows. If I want to order a pizza, right? Some of the pizza apps do a good job because there’s a very structured flow for ordering a pizza, you’re just not going to deviate from what options they have for toppings and the size and where you live and processing a payment. But how are you approaching this solution to add more cognitive elements to the AI’s interactions?

Dr. Michelle Zhou:

Very interesting question, that’s also my favorite question as well. So training AI with cognitive intelligence is to teaching AI human skills which is very difficult, right? It’s a very tall order. That’s why we should leave it to professionals like us to do it. So what happened here is you actually touched upon two points which is very interesting, so that’s why we wanted to democratize the process so the people who don’t have a resource or don’t bother want to learning how to teach AI cognitive skills because still have a chatbot for AI assistance which is cognitive intelligence. So if you said that first one, as any other workflow, let’s say the pizza caller, right? So they can actually go there, for example in our system, they can define the main workflow so the way to look at it is when somebody orders the pizza, you need to ask them what type of a pizza, what the size of a pizza, what toppings, right?

Dr. Michelle Zhou:

But in the meantime, they also need to enable the chatbot or these AI assistants answer questions. So for example, people might ask, “I have a coupon, can I still use it?” You have to [crosstalk 00:19:24] the coupon. “My birthday’s coming, do you have any birthday perks? If I buy let’s say five of them, do you have any discounts,” right? That’s a people question. So the trick or maybe if you think about the secret our [inaudible 00:19:40] but what we’re doing here it is the users or maybe the businesses don’t need to worry about how nitty-gritty details to be connected to be threaded together. But they would need to provide the business content. So our AI behind the scene, will work together to figure out, “Oh now, this is the main chat flow still there, I have to remember why I got interrupted. But if I got a question, I should go to check out my knowledge base to see how my actually knowledge base will provide the appropriate answers.

Dr. Michelle Zhou:

So actually it puts almost like a precisely you just articulated it, people just can go ahead and define the workflow they want and they can define the basic FAQS but then leave the rest to the experts which means, in our case, it is that we have already trained using tons of the data using our recorded actually neurosymbolic approaches to train, pre-train all these AI assistants with those basic, what I call, the essential human conversation skills.

Brent Csutoras:

And you mentioned this kind of like, and I’ve heard a lot of people in the AI space talk about, Nick Bostrom, a lot of people have talked about the need for a centralized controlled database, not only for the ease case but to avoid people going too far with AI and so forth. And so a controlled element like that, and where that would also be very interesting and very beneficial to most people and where there should be a huge adoption thing, is it’s very similar to established coding principles that we have today. When you have that you can attach in an API that’s like here’s all of our coupons. And you don’t have to even worry about do they expire because you’re it’s connecting to your existing database and so when people know, hey this ai system connects through these different connecting points so I need to have these databases and I can pull that from my existing systems, then all of the interconnections allow it to really be plugged in and take over in a sense without having a lot of input.

Brent Csutoras:

So who is this that’s doing this right now? I mean obviously you’re interested in this, is this something that your company is working with or spearheading and how far along is this and what are the resistance? I don’t think a lot of people really think about this idea of an open AI that everyone can utilize either for free or at a low cost price to standardize the experience.

Dr. Michelle Zhou:

So your question actually it’s a very important, but it’s also a big question, right? So there are different parts to it. So at a low level, there are open source software like from a Google, like from Facebook to basically help you to train one part of the AI called the natural language processing. You can think about it in a conversation, it’s not just about natural language processing, it’s also about the empathy, the responsible manner, right? So in our case, it is, besides our company, I am not aware of it people have been standardized that process and we call it conversational policies. So you can actually define different types of a conversational policy, for example, I wanted the conversational agent in this case, AI assistant, to be more persistent on its workflow versus can be flexibly digressed to different type of topics, right?

Dr. Michelle Zhou:

So another topic there’s a low level to basically to think about if you get dissect how the conversation occurs. There are many levels of skills that AI needs. Bottom level basically natural language processing, many companies are doing that. So many companies actually open source that we also leverage that as well, right? So then going on top, there are different types of skills like for example, even you understand, [inaudible 00:23:39] in your conversation you might understand everything you said and I might not respond to your questions as you expected maybe, right? So I may choose another way to respond it or maybe I choose not to respond it, so this is actually called a next level of conversational skills, we call it a conversational logic.

Dr. Michelle Zhou:

And then that one, we are trying to work very similar to just to discuss them modulize it, so other people can just take different parts almost like a lego system, right? Let’s put them together, we call it conversational topics. So for example, if you want to create am aI assistant, you said, okay, I want this AI assistant to talk about the three topics. First topic is asking users about their roles their challenges. Then another topic talking about it and maybe what’s are solutions you have found so far. Last one it is talking about, what’s your current plan of adopting a new solution? You see these are different topics? They don’t need to worry about those topics is very similar to just say that underneath their guidance knowledge base they all connect to it, so they just need to high level to specify I want my AI assistant to be capable of speaking this topic, this topic, this topic, then leave the rest of them to the underlying system to figure out how the topic can be discussed and how the topic can be even, how to say, thread it together because people may come from a topic to topic. [inaudible 00:25:10].

Brent Csutoras:

You mentioned there’s a lot of layers and and I’m very interested in the concept of the scope of what is needed to solve the issue, right? We talked about how when we talk like this, you can see the facial expressions, you can see when that changes, right? People start out a conversation a certain way and then they can progress that conversation, you can start to see cues of frustration, right? So some of the chatbots will identify that, right, some of the voice assistant kind of the voice interaction AI can really get that right, if you start using… I know some companies if you use cuss words they’ll go to a representative quicker or if you… But you could start to notice a chat has a certain flow and then it gets shorter. Like people stop saying, “Well, I don’t really want that and they start saying no, no, no,” and it’s like okay they’re getting frustrated, right? So there’s visual, there’s tone, and there’s change in pattern. How much of that is required for us to get to the point, because ultimately, we’re still trying to get to a point where we feel like we’re getting a human’s engagement without a human. How much of that is going to be required for us to truly solve this problem completely? Do we need all of it or do you think we’ll be able to get it through text only at some point?

Dr. Michelle Zhou:

It really depends on your task, right? So it’s almost like think about a company when they hire a person, it really depends on the job role, right? If you want this person to be a CEO of the company versus you want this person to be the salesperson of the company, they have very different roles, they have different skills. So it’s very similar to what you are talking about, the task of the AI assistant, what their roles are. So if their role is to serve as. let’s say, product marketing assistant, right, just actually try to understand the users or customers needs and want and elicit the information and provides the product information as in return, and then currently I think the technology are pretty much capable for actually supporting such scenarios. But if you… Have you watched a movie called Her? It’s a movie.

Brent Csutoras:

Oh yeah, absolutely.

Dr. Michelle Zhou:

So if you wanted to create somebody like AI assistant like Her, yeah, there’s a long way to go, right> So probably the next five or 10 years and maybe we’re going to reach there. Because in that case, you really need what we call the sensory information and gesture and facial expression, gaze, even body movement, body poses. So all these depending really on the tasks the roles you want your AI assistant, and right now I would say for operational especially business operational reasons the technology is already there and they just, with the right technology, when the businesses can really deploy those AI assistants to scale their operations in many areas already.

Brent Csutoras:

How are we learning today that outcome, right? So if somebody comes in as a company and they’re like, “We’re going to launch a chatbot,” is there a system like… I mean, I guess I would speak to your [inaudible 00:28:42] is there a system in there where they can see a human interaction but yet mirror it with a chat interaction and then when it goes off, it’s like a QA testing, is there a testing period, how do we actually get there? Because I still feel like anybody listening to this today, right, and maybe you can compare with what you guys are doing now and maybe what you’re going to do in the future, but right now.

Brent Csutoras:

I want a chatbot, in my head I’m going to go and put an AI assistant on my website or I’m going to put it in my app or I’m going to put it somewhere, I’m going to sit down and think out what are all the questions that people are going to ask and how do I answer it, so I’ve got the FAQ side. I’m going to look at what’s my process flow and how am I going to guide people through that. What is there that’s more than that and how is that taught? So somebody who’s looking at maybe they want to use your company or something and they’re like, “Hey, we want to work with you right now to create this.” What does that actually look like timeline and workflow to get something that’s functionally working?

Dr. Michelle Zhou:

Great question. Actually, I just wrote the blog on this one. The factors that people… There are four factors that people should really consider when they are considering adopting a chatbot solution or AI assistant on their website. Because in the future, it’s almost like every business right now must have a website, right? In the next couple of years, every website must be powered with a chatbot or conversational ai because that just gives you much more visibility, much more relation building opportunity with your customers. So how should the business go about it especially if they don’t have the AI expertise or AI teams in-house, right? Very simple. So I have this slogan called the U2C2, which means the two Us and Cs. First U. Because you want to adopt the AI assistant, first one you want to be useful. Useful which means it is you can complete the tasks that you want the assistant to accomplish. So you go about to do this one, if you are evaluating a chatbot company or AI platform, go to their website, try their chatbox on the website. But if they don’t even have a web chat box on their website, that’s a very big sign, stay away from it because you don’t even have a [inaudible 00:31:16] to put on one because this is their best show, right?

Dr. Michelle Zhou:

Then try it, say if you like that kind of a tone you like kind of a conversation, if you like it. Then second U, usability, which means it is then ask them, can I create this one on my own or I have this scenario, can you create it for me? Basically to see because after all, the AI assistant that will be sitting on your website that represents your brand to interact with your customers, you want them inside of it, right? So you want to do that that’s why we call it we want them to be empathetic to be really feeling base there to understanding your users.

Dr. Michelle Zhou:

Third one is the first C in the C2. First C is cost. To cost and speed to develop and then ask and say this one now I have this scenario so can I do it myself because that’s the cost effective, right? Or if i couldn’t do myself I would use your platform if you have to do it for me, how long will take it, how much cost you don’t want to go with let’s say for example hundreds of thousands of dollars taking 18 months that’s just not practical for many small businesses, right, with the cost and speed of development.

Dr. Michelle Zhou:

Last C is collaborative. Because the AI is never perfect, it always needs to work with your existing workforce. How does this AI work with your existing workforce? Can they report problems? Like you said, how do I know if it’s not working even? Can the AI actually actively real-time monitoring ah I couldn’t answer these questions, cannot be surfaced to the human assistance and also how fast for the human really collaborators to fix the problem, upgrade the teaching, this is learning. So for example, we hold this bar very high, we want all this to be real time. Which means it is let’s say, you go to talk the chatbot it’s powered by us if there’s a answer the questions that I’m detected immediately by the human. The human can immediately actually submit the answer and then all chatbots who are still chatting and without being taken offline to immediately learn that new actually knowledge.

Brent Csutoras:

And also to learn what didn’t work right? Our answer didn’t solve it because they asked the same question in a different way and things like that so it’s really, really interesting. So I like that two Us, two Cs, that’s really cool. I like the way that it kind of flows as well. For anybody who’s listening, there’s two parts to engaging with the company, right, that are beyond where we’re at right now, right? And that’s, what do I do to make the platform and the project successful. And then the second one is, how do I know and measure if it’s working successfully, right?

Brent Csutoras:

So the first part I’d love to get your idea because I think a lot of people don’t understand the mechanics and don’t understand all of the moving parts and they can be their own detriment. So if there’s somebody who’s thinking, I absolutely want to get into this and going with the same two Us, two Cs, how do they approach it? Do they need to hire somebody that’s dedicated? Do they need to put together data? How can they be successful in putting together their support for the two Us, two Cs?

Dr. Michelle Zhou:

Great question. Actually after answer you, I’m going to write another blog on this one. This is a great question. So first one, we are working with many customers, clients of ours, have asked these questions too. So first one, just ask the business owner or the marketer, right? So think about it is I need assistance, what I want this assistant to do? Almost like you hire, I’m giving example about, let’s say you’re a sales person right. So I teach you, let’s say a entry level sales person, what I want this entry-level sales person doing is that when this person is starting to talking to a potential customer, I wanted to prepare a set of questions to ask the sales person to ask the customers, right? So because you want to qualify the customers to say that, why are you interested in our product, right? So what challenges have you faced in your own work, right? What solutions have you tried? So that’s how you prepare.

Dr. Michelle Zhou:

Second one, ask if you prepare the salesperson then thinking about it is anticipated, what type of questions the customers might ask, right? Customers might ask it is so tell me about your product? What’s the cost of your product? How long will it take you to ship it to me? Do I have any uh special discount if I buy more? So anticipate, right? Anticipate the questions, that’s it. So ask the owner of the business if they’re actually working with a platform like ours they should really need to remember two things. What are main tasks this AI assistant needs to accomplish, right? What questions they need to ask your potential customers and what questions they need answer to the customers, that’s it.

Dr. Michelle Zhou:

So I want to go to your second question, how do you evaluate it? Very easy. So we’re actually just doing this one right now with the other customers. First one there are absolutely several metrics, KPIs, if you remember. First one is the quality of the information you got. So for example, if your assistant really does what you wish to do, for example, do they give you the quality of information of the sales leads you have, right, to compare with whatever you had before on your website with the static form with even human sales person. Second, does it answer the questions your customers have?

Dr. Michelle Zhou:

So we have this actually live dashboard to show you the percentage of the questions being answered. So our bar it is after two weeks your AI assistant should answer automatically, fully automatically, about let’s say 75% of the customer questions without any human intervention, right? [inaudible 00:37:42]. Third one, which is also very important, customer satisfaction. It’s almost like you have a sales person who is very capable, who actually complete all these goals, but really doesn’t leave a good impression wouldn’t do a very good job for your brand, right? You want to measure the customer satisfaction, customer engagement, [inaudible 00:38:02], very similar to what they measure web engagement, how many people who actually chatted actually browse the web? How long they stay, right? And how many people are converting the conversion rate as well?

Dr. Michelle Zhou:

And the people are… Because we just actually helped another company does this are using our chatbot and they said, “Oh, I want to know how many people who chatted with my chatbot are actually converted?” Either they signed up for certain events or actually applied for their program, right? So it’s KPIs, it’s absolutely important for them to measure and to basically to even to see, to compute the ROI, their investment and the return on investment on this particular AI assistant.

Brent Csutoras:

That’s awesome. I know that we realized from… We were just playing around with searching the journal when we launched our chatbot because we were just curious, we were kind of playing with a little bit and I don’t think we put a ton of effort into it and we just got a bunch of people that would just start kind of cussing at us or yelling at us or just screwing with the chatbot, and that’s kind of when we realized it wasn’t really doing anything. And then we were like maybe, this isn’t the best brand exposure for us to have something out there that people hate so much that they’re telling us off in the chat so we ended up kind of turning it off or making it very simple. But I do think a lot of people, they want to do well but I think that a lot of people are swamped and marketing becomes that 100 different directions they have to go in and they’re trying to figure out how to do all that they just know that somebody told them they need a virtual AI assistant on their website so they’re putting it on there and they’re hoping that it works.

Brent Csutoras:

What do you think is a general time period and cost expectation people should have to not just throw something on their website but to actually have something that they can benefit and see sales from? Is it a three month or a six month cycle? Is it something that’s like a couple thousand dollars a month or is it five figures a month? What is the range and the [crosstalk 00:40:10]?

Dr. Michelle Zhou:

[inaudible 00:40:12]. Be realistically, so using the current state of our technologies maybe setting up will just requires about a few days to a week. That’s actually people should keep in mind that if it takes longer than that and then unless you have some special needs normally for a typical meeting. But it requires a little bit of adaptation [inaudible 00:40:35] what people and you will be very surprised as a business owner once the chatbot can answer certain questions and people actually have more questions, which is a good thing [crosstalk 00:40:45]. So I would say expect a couple weeks to a month to actually gradually stabilize, which means it is you actually get to the I said get it to the 75 or even to 80 percent of the one. That actually helps the employees. So because we’re working with our clients employees that be much happier right now with this AI assistant because they said I hate to answer those repetitive questions, answer the same question literally he said a thousand times a year, I hate that. Right but they answer once and this chatbot can answer hundreds of thousands of times so they feel very achieved accomplished. That’s a good sign actually to elevate actually the working environment atmosphere, workplace morale.

Dr. Michelle Zhou:

So it’s basically it’s really about the couple of days to set it up but it takes a month or so to really to get used to, get accustomed to the customer base, right? But I wanted to mention that having an assistant it’s almost having a child at this stage. You cannot just ignore it, almost like your website very similar maybe otherwise good analogies kind of works out if your website always have a sales content, it’s not good for your business, it’s not just your brand. So similarly, when you have an AI assistant, you want to refresh the knowledge of this assistant. So for example, once you have a new product, once you have a new campaign, you have a new announcement, it’s great to teach the AI assistant as well right, so because then they can actually help you more to tell your customers about new products maybe even comparison a new product versus existing products.

Dr. Michelle Zhou:

So we always also kind of share with our customers and clients it is, just remember your AI just like your website, just like your kid, you wanted to continuously actually instill new knowledge and new content to it, [crosstalk 00:42:50]

Brent Csutoras:

For sure, and even as times evolve things change, right? The buying cycle, how people purchase, what’s important to them, what influences them, what kind of information they want? Those things can really kind of need to be continuously taught. So I think this is really amazing, I’ve really enjoyed kind of diving in. I think we’ve gone through some really fun parts to this. I want to ask you one kind of one more question and then I want to ask you another kind of follow-up thing. But the one thing I’m interested in is from again, you have a complete career in this space, right? So i mean you have a long career in AI, you’re looking at it from a very current practical standpoint for adaption, which I think is huge, right.

Brent Csutoras:

I think that’s a really, really positive thing, I think it’s going to help us get forward faster because I think so many people are focused on a very clinical or a very, not clinical that’s not the right word, but very experimental kind of can we break certain records or get faster or do something else in our industry for the accomplishment versus really focusing on the end user of this product and how do we evolve adaption right, so that when we do have that product, that product is useful to people. So I think that’s a very commendable thing and I don’t think a lot of people are doing it so I love that you’re focusing on that. Where do you see this going in 2022, 2023. What do you think is not only just from what you’re doing, but what do you think we’ll see in that transition, what kind of interesting ideas do you expect to see in the next two to three years?

Dr. Michelle Zhou:

Great question again. So this is also the things I have been thinking every day as well, right? So what’s also keep me and stay up at night. So one is very important and also near dear to my heart is this ethical AI topic. So what does it mean here is because especially [inaudible 00:44:51], our AI is very powerful now. It can really understand a person after maybe about 15-20 minutes of a chat, understand the way you handle stresses is, understand your passions and interests, right? So now next question is that’s why we also keep saying responsible AI, ethical AI, once you have that power and we say that with great power comes with great responsibility, I truly, truly believe that especially today. Because of that, you don’t want the AI to take advantage of people because [crosstalk 00:45:30].

Brent Csutoras:

I was just going to say that. What if they know that they always buy more at a late hour and so they always send them impulse things and they take advantage of it, yeah exactly, I was thinking that same thing, that’s very interesting.

Dr. Michelle Zhou:

If I knew that somebody’s an impulsive buyer than my people encouraging them to buy, encourage them to actually… And also if I knew somebody who is addicting to play games every day, every hour, then encourage them to have that addiction because it benefits some businesses, right? So this is a huge one coming up because in the meantime I’m also the editor-in-chief for our academic journal scientific journal called the SCM Transactions and Interactive Intelligence Systems. So we just published an article about the breaking the gap between ethics and practices how should you install instill this type of guidelines when you actually develop or deploy AI systems. I think that’s a there’s a long way to go but as a community, we have already started talking about that, discussing about that.

Brent Csutoras:

Well, you added another question to my mind and that’s that right now across the world we’re looking at data privacy, right? More and more, you can’t make the assumption that it’s okay to use the data that you might have on somebody without their explicit permission, right, and their ability to say, “I want you to erase this.” Have you seen that start to play into this space at all?

Dr. Michelle Zhou:

Actually let me revert to your question back to a little bit. So when the people actually ask to opt-in, this is one of our question, and then there’s some people don’t even ask you the opt-in, actually pry you on your privacy, right? So you can think about it is some of the, I would say bad behave chatbox, especially I have been in this area for a long time, I knew how people will behave with chatbots, with AI. Most people are very kind, they will adapt to it, they will be themselves, they will reveal personal details, they will reveal. So it is in our case actually we put what we call the watermark in a chatbot. So even when any other companies the business is using our AI assistant to elicit let’s say potentially sensitive information like social security number, we actually our water mark will be automatically fired in a way to warn the user to say this may contain sensitive information to make sure you only reveal such a personal sensitive information to a party you trust sure, right?

Dr. Michelle Zhou:

[crosstalk 00:48:09] especially if every child bodies on every website that’s going could happen, right? So we trying to this one to protect the businesses as well as to protect the users of the chatbot. Now going back to your saying obtaining information, so then once you do the information, so in our case, a big area we also share with our clients that is, be transparent, especially with that kind of intelligent cognitive [inaudible 00:48:38]. Sometimes it’s harder for the people to realize, wow this is actually, AI is not a human. Right? So we wanted the business to be very transparent and tell them this is not a real human which is the sake of expectation, bit in time, that transparency also helps. [crosstalk 00:48:56]

Brent Csutoras:

Yeah, that actually drives me nuts because every time I call, which I’ve had to do a lot lately because of my internet issues with Comcast, they always say let me check while I look at your record and then all of a sudden I hear click click click click click on a keyboard and I’m like, you’re trying to convince me that there’s a real person here looking up your information. I was like, I don’t know how I feel about that. The part of me and my… Obviously, most people don’t even notice that but I immediately am like, not quite sure how I feel about the attempt to make me feel that you’re real in this particular chat environment, right, in this engagement. So I’m probably super over sensitive to it just because I’m looking for it but I absolutely find all of this fascinating.

Brent Csutoras:

I can only guess that you know myself included, I’m going to definitely want to continue reading what you’re writing what you’re putting out and learning more about your company. Can you just take a minute or two to tell me a little bit about Juji tell me a little bit about you know the publications you have where you’re writing how people can find you how they can contact you.

Dr. Michelle Zhou:

Yeah thank you. So finding us is easy it’s juji.io J-U-J-I.io, that’s our website. And so mission of Juji is we really want to what we call the augmented humanity, advance humanity by unifying human and machine intelligence. That’s our belief. So towards this goal, for the past five and a half years now, so Juji has been working on two tasks first task is to power AI assistance with what we call cognitive intelligence, right, in the form of a chat box. People have to use the word chatbot [crosstalk 00:50:39] AI. Which means those agents, this AI assistant, really can understand, can remember, can learn, and then they can be responsible and empathetic in their tasks. That’s the one first task we’re doing. So the second task that we’re doing is I also touched upon is democratizing the adoption and the creation of such powerful agents. So businesses without any coding experience, without any AI expertise, can come to our website, sign up and put in there as I said earlier your question is, put in their main task flow, workflow, put in their questions they want their chatbot answered to their customers, that’s it. Then the AI assistants will accomplish built-in company changes will be automatically generated.

Dr. Michelle Zhou:

So in the meantime, we also publish in the top scientific journals like most of you will probably see ACM which is the world’s largest organization for computing professionals or also practitioners in this area so like Tokai this is the transactions and human computer, human interaction and the various of large conferences like Recommender systems like also the what you call it, intelligent user interaction conferences. So you publish there. And then our publications talking about it is how did you actually create AI so two types of publication, one it is to publish on state of the art technologies that we have developed. They share with people how we develop technology many people can benefit. Second part of it is to really share with people users experiences with those AI. So for example, do they trust those AI. If the AI has a different personas how their trust how human trust may change, right, and how the AI actually help the users accomplish certain tasks.

Dr. Michelle Zhou:

So for example, one of the universities is the published researcher uses our AI to interview students [crosstalk 00:52:46] effort, right? So which means they can really predict which team can do well in real world with the interview they assess every team members personality traits, interests, the passion, and then figure out which team would do the best actually it’s really predicted a ways…

Brent Csutoras:

That could be a really great just job hiring in general. How much you have to go through to take 300 resumes and put them down. If you had a bot they could send out an invite and then they didn’t have to talk but they had to write out and interact with that bot and answer questions and they could read the resume and ask specific questions about specific things and get that information and then come back and say, we think these are your 10 best candidates, that would be huge. I mean and time saving for companies. So yeah, it’s beautiful.

Dr. Michelle Zhou:

Yeah, we’re using that for our own hiring and also other clients have used for their hiring as well. And also it’s very hard to say and also this is a help so because right now it’s an inclusion of diversity too, right? Because you don’t look at the resume, you don’t get this, you don’t have this human biases for example, it’s a certain school, you bias certain people you knew, you bias a certain references. In this case everybody has a very fair opportunity to chat with the AI. And AI really tries to elicit the information you want and figures out what their strengths are. In this case you may actually find the person who is not the person you want to hire for this role but you’re not perfect for another role you have never saw that.

Brent Csutoras:

Exactly right yeah.

Dr. Michelle Zhou:

Basically give yourself the business a chance to give everybody a chance as well. So I grew up in China, I have this firm belief in Chinese saying that everyone is born with the talent, so we want to use AI to discover that talent to maximize that talent.

Brent Csutoras:

Well, Dr. Zhou, I appreciate you joining us, I appreciate the information, I look forward to being able to chat with you again, I look forward to reading some of your work and checking out the product. Thank you so much for sharing with us, thank you so much for joining me today.

Dr. Michelle Zhou:

Thank you for having me Brett, it’s been great.

Brent Csutoras:

Thank you.

Brent Csutoras:

(music)