Podcast

Conversational AI: Do Chatbots Dream of the Human Experience?

Written by Intevity | Dec 6, 2023 4:24:31 PM

George Jagodzinski (00:00):

Today, we discuss conversational AI, the challenges of building a virtual friend, and what lessons about perseverance, people, and communication can be learned from that challenge. I'm joined by Bilal Zaheer, who's held many roles over more than two decades at organizations like Phillips and Ness. He spent the last decade continually improving conversational AI. Please, welcome Bilal.

(00:20):

Welcome to Evolving Industry, a no-BS podcast about business leaders who are successfully weaving technology into their company's DNA to forge a better path forward. If you're looking to actually move the ball forward rather than spinning around in a tornado of buzzwords, you're in the right place. I'm your host, George Jagodzinski.

(01:02):

Bilal, thanks so much for being here.

Bilal Zaheer (01:05):

Hi, George. My pleasure.

George Jagodzinski (01:07):

So there's been a lot of buzz around AI these days, and last I spoke with you, you've been working on conversational AI for over a decade. I'd love to hear a little bit about what motivated you to do that.

Bilal Zaheer (01:23):

I started programming when I was 12, back in the mid-80s, on a Commodore 64 8-bit computer. And right from the beginning, the kind of children's programming books that I went through, it always intrigued me how when you're programming a computer, sometimes you actually have to think like a normal person would break down different steps of how a computer program should execute some instructions, and I always used to think about how computers could be made to behave more like people. And I got that opportunity in the late 90s when I used to write different scripts on MIRC, write different bot there. And it was very interesting how new users would come to a channel not knowing that they're actually interacting with the bot. It used to take them a few minutes to figure out that they had been actually getting responses from a bot there.

(02:15):

And tweaking those scripts, changing their behavior, randomizing the messages that a chatbot would give to a person, it made me understand, okay, it's not so much about the programming aspect of it, which is making it more real like a person would behave, but more about understanding how people behave with a computer, how people interact with a computer. If you are able to understand that, you are able to write better scripts, better programs, that are able to converse, so to say, with a person. It's not so much about the technology itself.

(02:51):

So eventually, a decade onwards, several years down the road, I was having a conversation with a former journalist from Reuters, and he started talking about how every year we publish a horoscope in the newspaper, we shuffle the daily horoscope, and yet, when people read through it, it's completely random, and still, they say, "Wow, this sounds so real. This sounds so much like me. This is accurate." When it's not. It's more about the perception of the people.

(03:20):

Again, it's about the kind of script that they're given, and it relates to them. And I started thinking if we were to build an app which could relate to people in a similar way, it could basically respond to people based on their messages in a similar way. It's more about understanding what people think about, what people desire, what sort of answers people expect. If an application could emulate that, you could actually end up building a virtual friend, not just silly chatbots. We have been experimenting with chatbot around that time, and a lot of them were actually just disasters. Even the award-winning frameworks that we tried to use, they would just break apart literally the second or third message at most. So the more we try to develop and develop such a system, the more we realize we have to understand people better. We have to understand how people interact or converse if you want to have more realistic virtual friends.

George Jagodzinski (04:24):

And there's no shortage of failures out there and horrible experiences, and like any interesting problem, it's rarely about the technology. And so, how do you find the right nuance of conversation because it's a very fine line of the uncanny valley where you can really latch onto what resonates with people? How do you find that balance?

Bilal Zaheer (04:44):

Right. Again, going back to the concept of a virtual friend, what does make a person your friend, whether it's an acquaintance or somebody you've known for a very long time versus a very intelligent virtual assistant even, who has massive amount of information with them, who will speak immaculate English? You still do not feel like you're talking to a friend. What you are getting canned answers, very well composed, very articulate, very accurate answers, but you don't feel any connection with them. And of course, the information that they can relay to you is limited by the amount of information which has been provided to them. They cannot carry on a conversation with you.

(05:27):

Whereas with a friend, who actually may not know anything at all about the subject matter that you want to converse about, a question that you might ask, a friend that you are talking to, it may or may not… that person may or may not have the right information, but you still feel like you can talk to that person. You can have a connection with them. You can build a relationship with them. It's more about the interaction aspect of it. It's more about the personality of that person that you need to develop. It's not just a massive amount of information that the person can keep on relaying to you and articulate in English, or any language, for that matter.

George Jagodzinski (06:05):

Yeah, I find with my friends or my teammates or anyone that I'm conversing with on a regular basis, what really helps me have a connection is having that common language. There's a short language that we can move through conversations rather rapidly. How do you build a platform or technology that's able to learn what that common language is or the shortcuts in their conversation?

Bilal Zaheer (06:26):

See, that is a tricky question, and that is why nobody has been able to build one, so far. The system has to be able to understand the person that they're relating to, and obviously, that would require a lot of information accumulated from the person themselves through various interactions or other people that such a system is interacting with on a real-time basis. On a daily basis, it may require tens of thousands of users.

(06:52):

Now, you cannot hire tens or hundreds of thousands of users to converse with that system. Obviously, you need a massive network of such virtual friends interacting with your system, which the system itself becomes your friend. Like Sun Microsystems used to say that the network is a friend. I'd say the network is the computer. My concept is the network is the friend. By interacting with so many people, every instance, 24/7 from around the world, you start to get a better know-how, better feeling, of what kind of people, what kind of background or what kind of demographic a certain person, if they belong to that demographic, how would you converse to them? What sort of slang, what sort of vernacular are you going to use to them? How are they behaving or interacting with the system? What sort of shortcuts, like you said, that they would be using?

(07:47):

And putting everything into context, you would be able to give them answers, which may not be accurate, by the way, but there would be something that the person would be able to relate to. The accuracy of facts can be improved, but it's the personality part which is more important in establishing a rapport, establishing a relationship, with that person. Knowledge packs can be integrated with that system regarding any field, I mean it could be health and fitness, it could be makeup and beauty, it could be fashion and apparel or something more technical, and you can add those knowledge packs to that system. But the overall layer, the personality part, is the more challenging part and actually more lucrative part for any business, in fact, to be able to build that.

George Jagodzinski (08:31):

That makes a lot of sense. Obviously, there's a whole host of concerns with providing the wrong information, but there's a better balance to be found. I find that most organizations that they're rolling out chatbots or other types of technology. I don't know. I feel like 80 to 90% of the focus and the effort is just on building the knowledge base, on building better answers, and not so much on that human experience and how people will connect. So maybe the secret is rather than 80/20, I don't know, maybe it's more 50/50 on the knowledge base and the human experience. I don't know if you have any perspective on what that right balance is.

Bilal Zaheer (09:06):

I mean, there's no fixed formula for it. Obviously, we still have a long way to go and coming up building such a system, which would be, say, more or less foolproof, which would not be able to start giving replies, which may be offensive to a lot of people, which may be very spreading misinformation for instance. But when you have a large enough population interacting with that system, obviously, there are already systems, already frameworks, available, not a hundred percent foolproof, which are able to figure out whether the information which is being provided to it is accurate enough or not, especially when you put related to the context, related to the person that you're interacting with, depending on the region, their demographics, their age groups, their interests, whether that information may be correct or not. I mean, for instance, “zupe” may mean one thing in Paris. It means something else in Singapore. It may mean something else in Pakistan.

(10:06):

So the more information you have regarding the person, and you can anonymize that information, in fact, the better you can interact with that person. And again, when you say you are trying to build a virtual friend, which is what we are trying to build, we're not talking about mission-critical systems. We're not talking about some… you're not advising people about their healthcare. We're talking about building a friend.

(10:32):

What sort of a problem would a virtual friend solve? You would basically be solving one of the greatest, possibly, I would say, the worst pandemics, which we are going to be facing in the coming years and a lot of people have already faced. I have faced that during the COVID pandemic – loneliness. And it's one of the number one reasons in the world for depression and leading up to suicide even. A lot of people are going to be needing, relying, on these virtual friends, and they are going to be helpful much in much more ways than just entertainment or keeping people company.

(11:07):

So yes, they may not be solving mission-critical problems, but they're already solving a much larger problem that the society is already facing, and it's only going to be increasing in the coming years.

George Jagodzinski (11:19):

And I would imagine solving these challenges that you're learning so much about the human dynamic. We work with global organizations, and I'm always fascinated with the inter-team dynamics across different cultures and different subcultures depending on what country the teams are in or what city that the teams are in. And I'm curious, 10 years pushing through this, what have you learned about those inter-team inter-human dynamics, and has it really just opened up your eyes a little bit on the human side?

Bilal Zaheer (11:51):

I spent a lot of time in Singapore. 10, 11 years of my career was spent in Singapore, and we were fortunate enough to work with people from around the globe. Singapore is the regional hub. Pretty much every role that I worked in Singapore covered at least 15 different countries in the Asia-Pacific regions. So you had Arabs, you had Malays, you had Chinese, you had Australians, British, American, all kinds of people. Even among the English-speaking world, you have so many different cultures, so many different behaviors, vernaculars. So again, this is one of the reasons why building a virtual friend fascinated me, and think about how such a virtual person would be able to interact with people from different cultures, from different backgrounds.

(12:39):

And this is, again, the reason why I figured out all the frameworks which are currently available and LP-based or LLM-based like ChatGPT right now, that's not the answer. That's not going to solve the problem. We tried, even IBM Watson, several years back, and one of the earliest things that we realized was no single framework or LLM system is going to solve that problem. And what you are going to do is, I mean, the dirty secret that has been held for a long time in the AI or conversational interface world is a hybrid system. People and machines have to work together, and like I said earlier, you can't really keep on hiring tens of thousands of people, hundreds and thousands of people across the globe, to solve this problem. You’ve got to build a whole worldwide network of people, of all backgrounds, all cultures. Sure, you're going to be starting out with a very specific kind of demographics and gradually grow from there, but the system is always going to be evolving while having live interactions with people all the time.

George Jagodzinski (13:46):

And I'm excited for these technologies to mature so that we can truly just have ensemble learning across all of these, both the hybrid and then all the multiple different platforms that are out there and point solutions. You've been doing this for, it sounds like, over a decade. You must've learned something about perseverance in pushing forward with this challenge. Tell me a little bit about what's helped you persevere and keep pushing forward.

Bilal Zaheer (14:13):

Yeah, of course. I mean, a lot of people came up to me and said, "You're crazy. The big five, I have not been able to solve that. IBM hasn't done it. Microsoft failed. Facebook couldn't do it. What makes you think that you are going to be able to solve this problem that companies with billions of billions of dollars could not solve this problem?" And that only made me more positive that this is something where you should keep on doing. This is something that we should keep on trying. This is still an open problem, still anybody's game, anybody could do it. It's just that we have to keep on looking for the right approach. And one of the things that early on I realized was that a lot of investors, a lot of tech companies are so hung up on the technology aspect of it, on the buzz of AI-ML, they're completely, pretty much completely ignoring the most significant part of it, which is the people, like I was saying earlier, is how people behave with systems, how people communicate with each other, how people even learn to speak a certain language.

(15:19):

I'm not speaking perfectly English, obviously, now. I did not learn this English through the education system. If I had learned English through my education system, my English would be much worse right now. I learned English by interacting with people, but way before that, I learned while watching literally TV every single day. Growing up in Pakistan, we used to have only one government-run channel. We would have only one English program per day, half an hour program. We would have one English movie per week, and I would be sitting till late night waiting for the movie to come out and spend half an hour listening to it carefully, and that's the only way I learn English.

(16:00):

So I mean, even babies, how do babies learn a language only by interacting with adults? Nobody's teaching them grammar or semantic or spellings or whatever. They only learn because they're absorbing whatever they hear, whatever they watch, whatever they absorb from their environment, the systems that we need to build has to emulate that, not an LP-based system or LLM-based system. That's not going to solve the problem.

George Jagodzinski (16:30):

It's interesting. Yeah, and the immersion makes complete sense to me, but do you worry at all about the kind of dumbing down of communication? I mean, maybe someone could have made this argument 30 years ago about TV, but I would be a little worried if someone's learning to speak from, let's say, TikTok or Instagram or something like that. Do you worry about that challenge? Do you put any thought around that

Bilal Zaheer (16:54):

A lot, yes. I just reshared a post today from one of the former product directors of McAfee, George [inaudible 00:16:59], a great friend based out of California right now. And he was talking about the same thing. And everybody's now just copy-pasting stuff from ChatGPT. And I've been observing the same thing in this younger generation. My own son, who's 15 now, they have stopped reading, and they have literally stopped applying their own mind. They're just simply copy-pasting stuff from other places. They're just watching videos. They're not reading.

(17:25):

So yeah, sure, I learned a lot of things from watching TV programs, movies. I mean, I'm extremely a huge fan of movies. I mean, for 10, 11 years, I used to watch at least one movie in the theater for 10, 11 years, every single week. And before that, I would read articles from at least four or five different sources, including Roger Ebert, before watching a movie and then go back and read the articles again if a movie is so good.

(17:52):

So you simply cannot give up on reading. And it's really unfortunate that I don't think the education system is really helping a lot. I don't think the social media, obviously, is helping a lot. It is dumbing down our younger generations a lot. So perhaps such a system could actually help improve that. But I don't see that happening right now that such a platform has come up so far, which would actually reduce this problem of dumbing down and rather encourage younger generations to read and not just watch and hear things.

George Jagodzinski (18:26):

Yeah, yeah. I hope so, as well. And so, as someone who learned through movies, I mean, I think we're not too far from an age perspective. I feel like you also, the level of movies was that much better in some of those decades versus what you're dealing with today.

Bilal Zaheer (18:44):

A lot of content in the movies right now is just being regurgitated and regurgitated. It's just reboots and reboots and sequels and sequels. Not a lot of thought, not a lot of writing is going through it. And now, Hollywood is facing two strikes from WGA, the Writer's Guild, SAG-AFTRA Strike. And one of the hated topics that is being debated is AI involvement in movie-making, having background actors which have been scanned once, so they're going to be reused, their digital versions are going to be reused over and over. And people are, obviously, already worrying about the scripts being written by AI, which means you are using a copy version of a copy version of a copy version.

(19:25):

My belief is that, hopefully, it's just going to be one dumb, silly genre that might still make money in Hollywood, but people still want to look for quality. People still want to look for depth. I hope that turns out to be true. People still want something which is more thought-provoking, something that not just entertains them but makes them think about things. There's always got to be some learning aspect to it.

George Jagodzinski (19:54):

And it is funny, your full circle with films and learning English from that, and now building this type of technology, it's been represented out in films. There's Ex Machina. There's Her. I was going to make a joke that, “Why do you have to build this product because Scarlett Johanssen already filled that role?” And I'm always drawn to dystopian future type of sci-fi, both in books and in film. I love it. I wonder what your response would be to people that are worried that, if we do have pervasive virtual friends, does that now start to just get in the way of real human interactions? Does it replace the human interactions?

Bilal Zaheer (20:34):

Isaac Asimov started writing about that a hundred years ago, and he was writing about the same things. I mean, nobody knew about robots back then. There was no AI back then, and yet, he was already writing about it. So this was always going to happen. People are always going to... I mean, are so fascinated about machines which have human-like interactions, human-like behavior. That's not going to go away, but hopefully, we just are going to come up with machine systems which are going to help improve our lives, which is actually going to be needed by more and more individuals, the aging population, but as well as the younger audiences, younger population.

(21:12):

I have observed so many teens online, my own friends, I would say… Why would I have teen friends? I used to be a photographer in Singapore, as well. There's so much loneliness among young students as well as people crossing 70 years, 80 years of age, post-retirement age. They really need company, much-needed company, which they are not able to get from their own family members, acquaintances, friends, colleagues. They just don't have anybody. It is driving people crazy. And again, why have that kind of an entity, a machine, a system? I mean, it's pretty obvious, and I really hope that we are able to come up with such a system within the next three to five years.

George Jagodzinski (21:57):

And there's a lot of people when they get worried about these topics, they're thinking that it's going to be completely pervasive everywhere. And there's a time and a place for different technologies and tools. VR, as an example, I don't know that we're all going to be walking around with VR goggles all the time, but people who have disabilities, they can now explore places in the world that they could never explore, and people who are maybe homebound, they can have those types of interactions with people and not be as lonely. So I think it's all about a balance, right?

Bilal Zaheer (22:27):

It is. And I would not want to use the word disorders, something that I don't know much about, but I don't think we should call this a disorder where people are so shy, so antisocial, they're not able to interact with real people, but they feel a lot more confident, comfortable, interacting with virtual machines who are very much people to them. There are extreme cases already where a person in Japan, for instance, married a hologram, Hatsune [Miku], and he's still married to her, by the way. And he found more comfort in marrying a hologram than a real person. I wouldn't want to call it the shape of things to come, and I hope that is not the shape of things to come, but there is a real need for it in certain markets already, certain worlds already.

George Jagodzinski (23:24):

That's fascinating. So interesting how things are going to shape up. Coming back to the films, maybe there'll be yet another Terminator film, and Skynet will be, it'll be like a buddy flick. Well, Skynet will be your friend rather than the evil overlord.

Bilal Zaheer (23:37):

Well, come on, T2, you already had a Terminator who was your buddy friend.

George Jagodzinski (23:41):

That's right. Yeah, he kind of was. Well, Bilal, I really enjoyed exploring this topic with you. I always like to finish with the fun question, which is, in your work or in your life, what's the best advice that you've ever received?

Bilal Zaheer (23:52):

Persistence. The one thing that you had mentioned about me 10 years on, still working on the same thing, still hoping, still sure, in fact, not hoping, that it is going to be building. It is just that we have to change the way, the way we address this problem, the way we approach it. Persistence, that's all it takes for any kind of success. I'm not talking about business, I'm not talking about technology. Whatever you do want to achieve, you just got to be persistent and not listening to people, "That is not going to happen. It's not..." You won't be able to do it. You just have to keep on improving yourself and, yes, change your approach if you have to.

George Jagodzinski (24:27):

I love it. Simple to say, hard to implement, but very powerful when you do. Bilal, thank you so much. I appreciate it.

Bilal Zaheer (24:34):

My pleasure. Thank you, George, for having me.

George Jagodzinski (24:38):

Thanks for listening to Evolving Industry. For more, subscribe and follow us on your favorite podcast platform, and pretty please, drop us a review. We'd really appreciate it. If you're watching or listening on YouTube, hit that subscribe button and smash the bell button for notifications. If you know someone who's pushing the limits to evolve their business, reach out to the show at Evolvingindustry@Intevity.com. Reach out to me, George Jagodzinski, on LinkedIn. I love speaking with people getting the hard work done. The business environment's always changing, and you're either keeping up or going extinct. We'll catch you next time, and until then, keep evolving.