Humans were born to talk and communicate. Much of the brain is devoted to generating, sensing, processing, and understanding speech and communications in its various forms. When humans were building their first civilizations we were also expanding the spoken and written form of communication. We didn’t start communication with swipes and types, in fact for most of human history we did not have technology, so it’s no surprise that what we want from our interactions with machines is what we want with interactions with people: natural conversation.
In this podcast, we talk about what chatbots are, a brief history of chatbots to this date, and what they are good (and not good) for. We also touch on the idea that perhaps we don’t need human-level intelligence in our chatbots for many applications, and what level of intelligence is required for particular human-machine conversational interaction.
|For over 25 years, QS has been helping prospective MBA candidates just like you make informed decisions about choosing the right business school. At our upcoming Dallas event, you can meet face-to-face with admission directors from top ranked US and international business schools, including UT Austin, SMU, Rice, IE, Hult, and many more! You will also be able to participate in interactive GMAT sessions by GMAC – creators of the GMAT exam, apply for $7 million in scholarships, attend Panels, and network with alumni and your future peers. Learn more about exclusive opportunities on the day by registering now to claim your free ticket at https://goo.gl/iRF9PR!|
[00:00:01] Welcome to the AI Today podcast, produced by Cognilytica, cuts through the hype and noise to identify what is really happening now in the world of artificial intelligence. Learn about emerging AI trends, technologies, and use cases from Cognilytica analysts and guest experts.
[00:00:22] This podcast is sponsored by QS. For over 25 years, QS has been helping prospective MBA candidates make informed decisions about choosing the right business school. At our upcoming Dallas event, you can meet face-to-face with admissions directors from top-ranked U.S. and international business schools – including UT Austin, SMU, Rice, IE, Hult, and many more. Find out more at topmba.com.
[00:00:50] Kathleen: Hello, and welcome to the AI Today podcast. I’m your host Kathleen Walch.
[00:00:54] Ronald: And I’m your host Ronald Schmelzer.
[00:00:56] Kathleen: And today we’re discussing chatbots. What are they, what are they good for, and what’s their future? We as humans were born to talk and communicate, and much of the brain is devoted to generating, sensing, processing and understanding speech and communication in its various forms. So when humans were building their first civilizations, we were also expanding the spoken and written form of communication. We didn’t start communication with swipes and types. In fact, for most of human history, we didn’t even have technology. So it’s no surprise that what we want from our interactions with machines, is what we want from our interactions with people – and that’s natural conversation. Or to put it another way: the natural language of humans, is not the language of computers.
[00:01:46] Ronald: So at it’s simplest core, a chatbot is a software application or a computer program that accepts input in the form of a written or spoken text in a natural language, and provides as output text in written or spoken form, also in a natural language.
[00:02:00] Kathleen: And so for us a natural language is slang, it’s colloquialism. It’s not just very cut and dry English, the way that you would find it in a textbook.
[00:02:10] Ronald: Right. And of course, it’s the natural flow of language. When you’re interacting with someone, you have pauses, you have starts, you go back in the conversation.
[00:02:17] Kathleen: You get interrupted. [laughs]
[00:02:17] Ronald: You get interrupted [laughing]. You don’t necessarily even have a direction with your conversation, you let the conversation flow. So that’s basically the whole idea of natural language. And yeah, it takes a lot of computing power and complexity to really understand natural language. But what’s different about chatbots and other programs, unlike these other programs, the whole trick of a chatbot is to make it seem like you’re conversing with a real human being. And how real this trick is, depends on the sophistication of the system, but also the extent to which the human wants to know if it’s talking to a machine or not. So that’s our overview of chatbots at a high level.
[00:02:51] Kathleen: Right. So now that we know what chatbots are, what are chatbots good for? So chatbots are best suited for applications where back-and-forth interactions with humans are required. So these scenarios would include customer support and customer service, information acquisition – especially over the phone. Interaction with devices where physical input is not possible or convenient, such as if you’re driving a car, you’re flying a plane, operating heavy equipment. Or a new class of intelligent personal assistants, where hands-free interaction is preferred.
[00:03:26] Ronald: So there are many possible applications of chatbots, where we want to have this conversational style of interaction – whether we’re interacting with our fingers and typing, or we’re talking, communicating. But simply using conversation for conversation’s sake, when you could more easily directly click or swipe or type, can actually be kind of annoying. If you remember from the ’90s Clippy, the paper clip from Microsoft Office, do you remember that? That was particularly annoying. So I think rather than thinking of chatbots as artificial intelligence per se, we should think of them as possibly intelligent conversational interfaces, and really think about it as the interface. So then you can think about when and how chatbots can be applied.
[00:04:05] Kathleen: Right. And I think that in order to understand chatbots, why they’re around and what they’re good for, you need to understand the history of chatbots. So the concept of the chatbot came very early in the origination of computing, and even before the invention of digital computing technology. So in 1950 Alan Turing proposed his Turing test, and we have previously talked about this in our ‘weak vs strong AI’ podcast. So if you’d like to learn more about the Turing Test, we suggest that you listen to this podcast, as we do a good job at summing it up. So indeed, the whole aim of the Turing test is to prove that the system is intelligent enough, that you cannot discern whether their responses are coming from a machine or a human. So basically a smart chatbot will pass the Turing test.
[00:04:53] Ronald: Right. And a lot of the goals of these early chatbots were to sort of prove that… The Turing Test is all about trying to hide the differences between a human and a machine: if we can hide those differences behind some system, then we can prove that it’s intelligent. So the first real chatbot to test this conversational mode of interaction was Joseph Weizenbaum’s Eliza program, developed in 1966. And for those of you who are not familiar with it: you would type commands and it would display interactions in this conversational mode. Eliza was modelled after a so-called Rogerian-style therapist; it would ask you questions and lead the conversation into particular directions. And Eliza was actually surprisingly good at fooling people into thinking that they were actually interacting with a real person, a real therapist. But the interesting thing about it, is that Weizenbaum wasn’t trying to prove intelligence with Eliza. He was actually in part trying to debunk some of the hype around artificial intelligence. And so what he did, is he created a unique way for the program to generate these open-ended questions, and remember answers from the respondents, and then turn those answers back to the person – as if it was responding in a way that a person would. And he really wrote Eliza as a way to basically reveal the magic of how it works, to say: ‘there’s no magic here, here’s how the system works: this couldn’t possibly be intelligent, if all I’m doing is sort of mimicking about it’.
I think one of the important notable things about Eliza is that all the chatbots that have since fallen – there have been many – have been designed using the same basic pattern as Eliza: recognizing key words in the input, have various different templated responses that can be used in different ways; use previous answers for future responses, and have a basic flow. So we could talk – if you’re familiar with PARRY, which was the paranoid schizophrenic alternative to Eliza… There was A.L.I.C.E, Jabberwacky, Cleverbot, and a whole bunch. And just long story short: over these few years, these chatbots really competed with each other: not only to see how close they can mimic human interaction, but also for notoriety. And the Loebner prize came out, it’s an annual competition in AI that awards prizes to the computer programs that the judges felt to be most human -like. So, you know, this is kind of where we are with the state of the art with chatbots.
[00:07:01] Kathleen: Right. But there are some examples of chat that’s gone wrong, and I think that probably the most famous example is Tay. Tay was released by Microsoft in 2016, and it ended up lasting for only about 24 hours before it needed to be pulled down. Tay was programmed to imitate those that interacted with it. But the problem with this, was that due to some trolling users, Tay ended up evolving to spout off Nazi rhetoric and feminist-hating banter, within about only 16 hours of it being released. So this was obviously not what Microsoft wanted and it was quickly pulled down – but the damage had already been done. So the moral of this, is that you need to be careful with this emerging technology, and always prepare critically for the worst way that consumers may try and use it. And as we have brought up in a previous podcast, this is a really good example of bad actors doing bad things. So this isn’t necessarily some lifechanging, life-ending wargame example, but it still wasn’t good. It wasn’t good for Microsoft, and some of the things that Tay said unintentionally were very hurtful and offensive.
[00:08:11] Ronald: So where are we now with the whole chatbot market? We see chatbots used in a variety of different contexts. Some of these so-called traditional or less intelligent chatbots, are focused on a single purpose mode of interaction, such as customer support or phone interaction, or assisting with online commerce or tech support. And odds are that you’ve probably interacted with many chatbots. If you call into most airline reservation systems, it’s going to say: please tell me what you’re trying to do or your reservation number. If you talk to a bank or any sort of customer support now, you’re having a conversational chat experience. The thing is that the context of the chat is narrowly restricted to just that thing. So if you start talking about your personal feelings, or trying to spout off some rhetoric, it’s just not going understand – nor should it.
[00:08:52] Kathleen: Right. It probably has about you know 20, 30, 40 keywords that it can catch and understand. But like Ron said, if you start talking about things that it doesn’t understand; if it’s an airline, for example, and you start talking about a boat or a car, that’s off topic and that’s not what it’s used for. So another thing is interaction with intelligent personal assistant chatbots. We’re really starting to see chatbots take off within the context of intelligent personal assistants. And these are Alexa, Google Home, Siri, Cortana. Those are the really most popular examples of it, and we’ll spend some time digging into these assistants in an upcoming podcast, so we won’t go into it too much here. But let’s just say that there’s a lot of chatbot in personal assistants, and the challenges of chatbots become very obvious when you start talking to these devices a lot.
[00:09:42] Ronald: So in addition to the narrow applications of chatbots within those purpose-specific tools, and in those intelligent personal assistant chatbots, we’re certain to see them used more and more on instant messenger. Instant messaging applications as Facebook Messenger, or Kik or Slack, Telegram, Discord, are certain to add bots to their technology. And these bots are helping these applications provide greater functionality, such as e-commerce or a greater search capability, or allow them to integrate with other applications. In the case of Slack, there are many bots that will allow you to interact with other applications, in the context of the conversation you’re having. So if I’m having a conversation with Kathleen on Slack and I say “oh, I need something”, I could say “hey Slack bot, go get me this thing” – and it’ll go and it’ll get the thing, and it’ll return the answer within the context of that chat. I don’t have to launch a new application, I don’t have to open up a new chat window. And I think that’s kind of an interesting way to use chatbots in these instant messenger-style applications. And you know, we have a long way to go to see what the ultimate benefit of instant messenger chatbots are, but we’re starting to see a lot more of them.
[00:10:42] Kathleen: Yeah. I mean, right now, like Ron said, I don’t see much benefit from these instant messenger chatbots, other than maybe it might save me a step or two. Like he said, we asked Slack bot to get us something; to take notes for us, do something like that. I haven’t – at least personally – gotten much benefit from these, but I’d be interested to see where they go.
[00:11:00] Ronald: So let’s talk about the future then. So where do we see chatbots? I think this is the kind of area that Kathleen and I have had some interesting conversations about. When you look at science fiction movies, you look at Star Trek, you look at Star Wars, you look at 2001 Space Odyssey. What do you see people doing with their computers? They’re talking to them. They’re not clicking and swiping. Okay, yeah, on the enterprise they got buttons. I’ve seen that too – from Uhura in the original series, to data in the Next Generation. Yeah, they’re using interfaces, but they’re spending a lot of time talking to computers. But what’s interesting about this, is that they’re not having conversations with those computers as if they’re humans. They’re having conversations with most of the machines as if they’re machines. And the whole idea about that is that these systems are very complicated. The Enterprise probably has billions of moving parts, it’s extremely complicated to go into warp drive. The last thing you want to be doing is fiddling with knobs, getting it wrong and warping yourself into a star. So it’s easier just to say, talk to the computer, have it do warp nine or whatever you need it to do. And so even in that situation in the science fiction world, even though I know there are movies that you see that where you have real life robots and people are talking to them, the computers that we seem to like to interact with – whether it’s R2-D2 or C-3PO or the Enterprise – are machines that are doing something useful. I think that’s, to me, the kicker here. How intelligent do we really want our chatbots to be? Do we really want to be fooled that they’re humans?
[00:12:20] Kathleen: But I just want to interject here. The one thing that I do think is very different, is that in the movies, you’re talking to a machine and you know it’s a machine. Nobody at any point in that interaction is trying to pretend that the machine is a human. I think where we’re moving to – and we’re starting to see companies move to – is that they are using these machines to replace humans, but don’t want people to know that they’re being replaced. I think that that’s where the difference is, and I think that that’s why people want these intelligent assistants to pretend that they’re humans. Because they don’t want you to know that their call center has now been replaced with all bots, and that there’s only maybe two people in there to handle really complex or really irate customers.
[00:12:59] Ronald: Okay. Yeah, that makes a lot of sense then. In the context where we are using machines to replace people, and where people are used to having a conversational mode of interaction, I can totally see that. But I think for a lot of these other applications, we really want our chatbots to be useful. Not necessarily intelligent, we want them to be useful. And so I think this is the question for your particular application. I think the answer to this is it depends right, for your particular application. Does a chatbot really need to pass the Turing Test or win the Loebner prize, to be useful? Or maybe for their particular application, like the calling into a call center and dealing with a reservation, we don’t need it to be so intelligent. It just doesn’t need to be that intelligent to be useful.
[00:13:35] Kathleen: But also, it’s not pretending to be a human. So I think that the level of intelligence is different. It needs to be intelligent enough to understand what I’m saying, it needs to be intelligent enough to answer my questions; it needs to be intelligent enough to help me change my flight or let me know what gate I need to go to – but it doesn’t need to be intelligent enough that it can help me also book a taxi when I land at the airport, or help me book my hotel. That’s not what I’m looking for. And I think that that’s the question that the enterprise and vendor listeners need to answer: what exactly are you trying to solve? And I think that that gets down to how intelligent your system needs to be.
So if we can go back a little and wrap this up for our enterprise and vendor listeners, with how useful are chatbots, we need to go back to where we started. And chatbots are best suited for applications where back-and-forth interaction with humans in a natural language is required. So we’ve given a bunch of scenarios and in these cases chatbots are absolutely useful, and being used more and more every day. But we don’t necessarily need these chatbots to fool the users so that they think that they’re human, and that they’re having a human-to-human interaction. In order for these to be useful, they just need to be intelligent and useful enough, without frustrating the user or a customer. And they need to provide enough value to the organization.
[00:14:59] Ronald: Alright. So listeners, we hope you enjoyed that back-and-forth. We will provide some more articles and details in our shownotes, and some links to some interesting videos about interactions with some of our chatbots. I hope you enjoyed that.
[00:15:09] Kathleen: Thanks for joining us and we’ll catch you at the next podcast. And that’s a wrap for today. To download this episode, find additional episodes and transcripts, subscribe to our newsletter and more, please visit our website at cognilytica.com. Join the discussion in between podcasts on the AI Today Facebook group, and make sure to join the Cognilytica Facebook page for updates on this and future podcasts. Also subscribe to our podcast on iTunes, Google Play and elsewhere to get notified of future episodes.
[00:15:39] Ronald: Want to support this podcast and get your message out to our listeners? Then become a sponsor. We offer significant benefits for AI Today sponsors, including promotion in the podcast and landing page, and the opportunity to be a guest on AI Today Show. For more information on sponsorship, visit the Cognilytica website and click on the podcast link.
[00:15:59] This sound recording and its contents is copyright 2017 by Cognilytica. All rights reserved. Music by Matsu Gravas. As always, thanks for listening to AI Today. We’ll catch you at the next podcast.