Are you considering chatbots as one of the building blocks of your business and marketing strategy? Here are valuable insights into the future of chatbots after the COVID-19 crisis collected from industry experts and professionals!
Chatbots had a rocky start. Despite the initial hype, the adoption has been slow. Though, once we let the idea of making bots indistinguishable from humans go, their true potential came into the spotlight.
In the past few years, chatbots have been gaining on acceptance, popularity, and efficacy mainly because people started to reimagine the ways chatbots could be applied in daily personal and business scenarios, focusing on making them useful rather than “indistinguishably human”. This shift became ever more apparent as no-code tools freed bots from the shackles of hardcore software development and made it accessible to markers and communication specialists alike.
As one of the players in the field of no-code conversational marketing, we are always excited about the new waves of innovations in chatbot development and application. Therefore, under the quarantine-inspired “Antipodcast” initiative, I compiled opinions, experiences, and predictions of chatbot and AI technology experts and professionals on the future of chatbots and challenges that await the chatbot market and industry in years to come:
Start reading or jump to the section that interests you most:
- Conversational Experiences: Notable Projects & Initiatives
- The Dark Side of Chatbot Experiences: What Can We Learn?
- Impact of COVID-19 on Chatbot Industry
- The Main Challenges in the Way of Chatbot Success
- The Future of Chatbots
- Where Does Landbot Stand?
- Listen to the Full “Anti-podcast” Interviews on YouTube
What is a Chatbot, Today?
Before we get started, it’s important to clear up what chatbots are, or better said, what they represent. Today, chatbots are far more than a simple messaging service. Sure, they can be as simple as message exchange or become a rich fusion of interactive visual and conversational (text or voice-based) elements, designed to not only collect data, understand the natural language but also perform advanced functions such as, for example:
- Speech to text
- Language recognition & translation
- Speaker recognition
- Content moderation
- Text analytics
It’s looking at chatbot technology in a broader context of conversational experiences and language understanding that helps us see chatbots for what they are – smart conduits for all aspects of your business.
So to help you expand your idea and understanding of chatbots and conversational interfaces in the context of today, we asked some of the leaders and experts in the area to share their latest conversational projects.
Chad Oda, partner and head of consulting at Chat Mode, points out a surfacing trend following a simple, so-far largely overlooked use case. He mentions a project where Chat Mode helped a multinational mid-market company share key information instantly across continents. Previously, the company got in trouble for breaching a non-compete clause with their existing client because executives from Singapore were not able to get in touch with the US branch and obtain the information on whether a non-compete clause applied quickly enough. Chat Mode helped the company by building a chatbot layer on top of its knowledge base allowing the executives to retrieve essential data on demand, without delay. He further elaborates:
“… if you think about any digital transformation strategy a mid-market enterprise company has executed probably within the last 5 to 10 years, almost exclusively they’ve been centered around customer experience, right? Companies have made really, really good customer experiences but what they’ve done… they’ve, unfortunately, neglected employee experience. They haven’t taken those same principles and deployed them inside of their own enterprises. So, things as simple as knowledge management that we’re deploying for mid-market organizations… they are finding them extremely valuable […].
Chat Mode’s solution is simple: “Essentially, we build a chatbot layer on top of existing knowledge bases to extract that information much more rapidly in a way that is easy for the end-user. So, even simple use cases like that are seeing a tremendous value gain…”
Amir Shevat, a product developer with over 20 years of experience and companies like Microsoft, Google, Slack and Amazon (Twitch) on his resume, shared a COVID-inspired conversational project to simplify the recruitment process:
“…We had the COVID and a lot of people lost their jobs. So, I was thinking… how can we create a new experience around hiring that will… That is [based] around conversational interfaces that will help people hire all these people back. How do we rethink hiring and interviewing people? […]”
The conversational solution Amir came up with is called Interviewsly and can be accessed in the Slack app directory. In the simplest of terms, the app creates a conversational channel for the interview. It displays the candidate information at the top of the channel and offers a list of position-specific interview types each of which can be assigned to a designated team member. Once, you assign an interview to someone that person will receive a Slack message featuring interviewee information and a list of suggested interview questions specific to that position. Better yet, by clicking on questions related to a specific proficiency you the interviewer can write down their feedback in real-time. After the interview is done, the interviewer can submit his or her feedback by clicking the Submit button. This will then update in the main interview channel, however, the individual feedbacks won’t be visible until all of the interviews are completed in order to avoid creating preconceptions and biases.
Personalized Shopping Experience
Roger Kibbe’s recruitment bot is not the only COVID-inspired project.
Brandon Fluharty, currently working as a strategic account solutions lead on global conversational AI at Liveperson, shares his experience consulting on a project that draws on chatbot’s capacity for personalizing and improving customer experience.
Working with one of the top big-box retailers in the US, Liveperson built a conversational solution to support social distancing and keep customers informed in the most natural way possible. : “…with COVID-19 there’s a massive shift to curbside pickup and delivery which is very very difficult for these types of brands that don’t have a lot of people in their stores. So, what we are working on here is taking some of those one way SMS notifications that are coming to consumers to instruct them on how to pick up their products after they purchase them… […] [and] helping to give the customer more information when that SMS notification [arrives]. We are building bots and FAQ bots to make [getting more information] very easy.”
When asked about adapting this service to voice Brandon elaborated: “We could take that from messaging to voice assistants like Alexa or Google assistant… [though] the primary focus is these SMS communications [which] are going out to customers anyways. But, if you have questions, the problem is that when you try to respond to that SMS notification it goes nowhere or you don’t get a response. So, we are helping the brand to turn those SMS notifications with the solution we have into proactive messaging that allows you to actually engage in a two-way dialogue.”
While impressive, that’s not where Liveperson intends to leave the project: “…that’s easy stuff, anybody can really do that. What we wanna transition to is becoming more of a personalized shopper so, over time, the bot is gonna be able to collect those intents from those consumers, understand what product they purchased and come the busy holiday season in Q4 later this year they will be able to sort of deliver a personalized shopping experience: “You bought a smart speaker in May. Here are some great products that might complement that”…”
Customer Service & Intent Recommendation
Hanlin Fang currently manages the conversational AI interfaces product team at ServiceNow which focuses on taking chat from conversation to resolution, in other words, designing and end to end natural-language-processing-driven chat experiences across different channels. She sheds some light on how the company is working to improve customer service by analyzing and correlating conversational data:
“… The most recent AI project we are exploring is related to conversational intent recommendation because, you know, ServiceNow… ServiceNow is an IT management system of records so we have incidence data and we also have this agent data so we can analyze and correlate different data sets to see what are the intents or topics we should introduce as part of the bot conversation…”
Their idea is to increase the efficiency of issue resolution delivered by the bot (decrease the deflection rate from bot to agent) to save agent time allowing them to focus on more complex tasks. Access to extensive data sets allows them to identify relevant intents and keep their bot and natural language models domain-specific thus, avoiding the mistake many made in the early days of bot development of trying to create an “all-knowing” generic chatbot.
Conversational 3rd-Party Ecosystem
Adam Cheyer (the cofounder of Siri which sold to Apple and Viv which sold to Samsung to power Bixbi) is set on creating an open conversational AI ecosystem:
“I believe that to create conversational assistants can be as powerful or more powerful than the web or the mobile if done correctly. Today, the systems are not that, you know. They do a few things. Even though there is under a thousand skills in Alexa, nobody uses them so, what I am trying to do is open up a 3rd-party ecosystem that is equitable and fair for all members; Get in on hundreds of millions or billion devices and make sure that 3rd parties can monetize their services and also give them the best tools and the most advanced AI available […]
Apple has kept Siri closed to 3rd-party developers. Samsung, however, is taking another approach with Bixbi, working with Adam to empower developers worldwide:
“So, Bixbi is on hundreds of thousands of devices… So, if you are a developer you want to learn a completely different way of making assistants, conversational assistants, go to bixbidevelopers.com and you can get tools and platforms like no one has ever done before.”
Speaking of Bixbi… Seasoned tech architect and strategist, Roger Kibbe, currently working on developer evangelism for Viv highlighted the challenge of multi-modality that comes hand in hand with device variety:
So, one of the things that we’ve built internally for Bixbi is the Yelp capsule. [… ] …if I wanted to find a vegetarian restaurant that’s open at 10 pm next Saturday I would have to do a lot of typing and swiping but then I could see the results would be… that worked really well… so if you ad a voice interface to that then you start thinking about “Hey voice is really good at input!”. [… ] I can say that one sentence. That’s much faster than anything I could type, swipe or tap. On the other hand, the output is naturally a kind of scrolling list of restaurants [… ] So, one of the things we do with that capsule is you can all control it by voice but you can also view the results and kind of click it and see restaurant details via touch. So it’s an interesting combination of using voice at what it’s really good at – the input modality and then using the UI at what it’s really good at – the output modality and king of slicing and combining those together.
Since Samsung smart product offering goes from refrigerators through phones to TV’s, its Bixbi is inherently multi-modal. So, as Roger emphasizes, the question they are asking when developing virtual assistants is how do you mix voice and screen together in a multi-modal way?
“I think it’s a really interesting challenge because… you want a… You want to be able to uprate it with voice and in some sense you want the screen to be an adjunct, or kind of support what’s happening. On the other hand, if I am looking at a screen and can see something and some voice is drowning on and on about the options you know that’s not optimal. So there’s really kind of a design perspective, kind of thinking around… Hey! Is the user viewing the screen… Are they not viewing the screen? How do I design my voice application… my multi-modal application with voice and UI differently… what can I do from an AI perspective to understand the way the user is responding… We are doing a lot of thinking around that, not only on the technology side but also on the design side to kind of optimize the experience.”
Fighting the Bias
However, notable chatbot projects don’t always revolve around bot development.
For instance, Chris Messina (probably best known for the invention of the #hashtag) is a seasoned Silicon Valley product designer and technologist who has landed his skills to brands such as Uber and Google, emphasizing the importance of looking at the bigger picture:
“Last year, 2019, I actually spent a bunch of time going around the world talking about product design, strategy, culture… really with the recognition… I guess where the insight I suppose having worked in this industry for a long time and recognizing that artificial intelligence is going to build in a lot of biases that present in our culture… and that unless startup founders and creators and makers start to have a better sense of themselves, a better sense of emotional intelligence and social intelligence we’re just gonna build in a lot of those negative behavioral biases into our artificial intelligence and so… I was encouraging people to basically think, you know, harder about that problem and to spend more time working on it as opposed to only developing hard or technical skills for programming and so on.”
Caryn Lusinchi, a former Googler with extensive experiences launching hardware and software apps and IoT products in over 43 different languages, currently working as a contractor supporting global scalability for the WhatsApp Business API, too, was involved in a project fighting bias in AI:
“There’s a website I recently launched called biasinai.com and it’s an evolving global directory of companies, startups, academia and NGOs essentially creating products, data sets, research and services [showing] how some organization are helping the commercial marketplace navigate different types of bias in AI products be it racial, cultural, algorithmic biases…”
These and many more conversational projects are underway all around the globe. Despite the general notion that AI chatbots have faded away after the initial hype, the industry is alive and kicking, undertaking often even mind-bending projects.
However, there is truth in every stereotype. Assessing the future of chatbots without taking into account its current shortcomings and fails wouldn’t yield a realistic picture. So, we asked the experts to share their worst experiences with conversational interfaces and AI.
For Caio Calado, conversational product designer at Microsoft, one of the worst chatbot experiences centers around misunderstood context:
“I have faced many interesting and not so great experiences when it comes to conversational UI. Coming from the product view… design view… one day I designed a bot for a huge music festival – it’s called Rock in Rio – It’s a rock festival, huge in Brazil. The bot was supposed to help folks [with getting around] and help everybody with information about the space, the concerts and everything. And then, one day, one disabled person using a wheelchair asked for help and the way that she asked for it was kind of angry. And the bot was supposed to be happy and friendly and all… but the way I read the message [the context] was just: “Oh my gosh That’s so bad… the bot shouldn’t say those things in that tone…”
In the context, the end-user had a serious reason to be angry. However, the bot was designed to diffuse and lighten up situations when users used harsh language with good humor. Caio continues to explain: “It was trying to be friendly but it wasn’t appropriate for that situation and that’s the challenge of the industry. The bot was made in a different context and that’s a struggle for everybody. Because we never know what people are going to say… the bot answered right but it was wrong for the context.”
Cristina Santamarina, product manager at SVSG, and founder of community “Chatbots en Español” brings up another common issue that has given conversational experiences a bad rep over the years, pretending to be human:
“I wouldn’t mention one specifically but there’s been something in common [when it comes to] my worst experiences and it’s the fact that they were customer service chatbots, so chatbot that were meant to be helpful, pretending to be humans… […] I think this brings a very bad experience because they will never be good enough to answer all of my questions so I will get trapped in the usual issues with chatbots but also it feels like the chatbots and brands behind them play with the user’s trust and I think they are wasting resources on working on small talk when they should be focusing on delivering a brilliant feature experience.”
Andres Pulgarin, CEO and Co-Founder of BotsLovers, draws attention to the lack of “conversational intelligence” in chatbot design that makes bots seem too aggressive: “We call these kinds of bots, the impolite bots… it’s not natural […] on the websites, these bots don’t even let you ask something first and they [already] demand your email or name.” He also continues to point out another experience-breaker which he calls “lazy bots” referring to those cases where not the bot but its trainer didn’t bother to train the bot appropriately, creating friction when unnecessary.
Omar Pera, CEO of Reply.ai a conversational AI platform recently acquired by Kustomer, also mentions a fairly common and serious setback in customer service bots: the lack of bot-to-human handoff: “I think there are more bad experiences than good ones on the market today… I think the biggest problem for having a very bad bot experience in customer service is that there is a lot of friction to be able to talk with someone when the bot doesn’t understand.” He believes that phrases such as “Sorry, I don’t understand” shouldn’t be part of the bot’s vocabulary.
Continuing to dig deeper into the subject of the future of chatbots, we can’t ignore the current world crisis. COVID-19 represents a turning point. Though, the question is… is the impact of the global pandemic on the chatbot industry positive or negative?
While the individual experiences of COVID impact differ, the general notion remains positive: COVID helped chatbots and conversational AI break into the mainstream consciousness.
Surge in Conversations An Opportunity to Advocate for the Solution
“Taking out all the bad stuff of the COVID situation, […] it is really great for chatbots because many companies are moving to digital space. Some of them are creating websites, some others are creating apps and things like that… but there is one thing everyone needs to do and that is talk to someone. Most people… they don’t want to go to a website, they want to talk to someone right away and companies now have the opportunity to scale up that conversation and either help or provide guidance… […] Down here in Brazil, we noticed that there are some industries that are […] doubling the size, the number of bots and also the number of conversations they are having with their clients. Companies down here, most of them, noticed an increase in demand for providers to develop conversational experiences […] Everyone needs to adjust to this particular moment but for chatbots, in a sense, it’s an opportunity to advocate for the solution.”
Chatbots Going Mainstream
“I think it’s actually been beneficial for us, so when this all started I saw a lot of movement of people trying to create chatbots to inform on the figures and to give advice. I actually went on and trained my chatbot for this, too. But I think it was very positive that not only chatbot creators and companies but official institutions like governments and city halls and big organizations WHO went on and created chatbots for different channels. This has benefited us because they’ve grown their user base, they’ve made chatbots mainstream. It’s only early adopters using them but millions and millions of people.”
AI Community Coming Together & Adoption Fast-track
“… So there are two trends. I think the first trend is one that’s really reassuring. I think for the first time, we are really seeing the conversational AI community to really band together and build bots, you know, and really take it from the approach of trying to create value for people in a time of need and that’s something I’ve sort of seen across the board which has been really inspiring. So, number one is seeing the community really come together and drive really beneficial efforts. And, number two, we are actually beginning to see sort of a higher adoption of messaging channels, unsurprising as a result of quarantine. I think Microsoft Teams, for the first time, have essentially hit, you know, some of the highest monthly active users to date. What I have been hearing is actually:
- #1 – companies that already had Microsoft Teams on the road map to essentially adopt it, those timetables have been expedited so more organizations are adopting these things quicker.
- #2 – companies that already had Microsoft Teams, companies that have been already evaluating some sort of chatbot integrations, those projects also have been expedited.
So we are actually seeing an increased demand of not only internal messaging channels but also those bots that would sort of enable extended business functionality.”
Increased Use of Voice Apps
“One of the things we noticed in the conversational AI and voice is that actually voice application usage has gone way up during COVID-19. I think […] there’s an obvious explanation. People are home more, they’ve got their voice assistant, they’re like what I do with this beyond asking for the weather or to play my music? Right? So they are just using and experimenting more. But also I have a theory, it goes back to that ‘Hey, voice is an inherently human thing’ ergo voice is kind of the most social way of interacting with our technology and I think with us all being socially isolated, we are looking for social outlets… Now, I don’t pretend to think that your voice assistant is your friend, I’m not going that far but […] I think there is something to voice about being an inherently social medium that many people can get involved with which is attractive and maybe leading to some of that increased usage.”
Spotlight on Outdated Practices
“So with the pandemic, I think it’s more matter of sheller and place and work from home and this shift for so many more people that had – let’s say – been hold outs or who had believed that they can continue to do their work in a conventional way, needing to find new alternatives that didn’t require face to face human interaction.
And so in addition to that, I think we are also seeing the ways in which the existing infrastructure and systems one weren’t really set up for this type of spike in let’s say moving people online and two, I think perhaps more importantly that the way in which let’s say – I’m thinking specifically of filing for unemployment and things like that – the way in which the government has rolled out different types of support services that rely on conventional web applications and web infrastructure… you know, one, it’s a bit too slow, two it’s pretty user-hostile, three it doesn’t really meet people where they are anymore.
… just rolling out a webpage, you know, 2006-style is one way to do things but I don’t think it’s how people expect to use computing technology. They expect to be able to access services through SMS or through social media or through Facebook or Instagram or whatever and they are used to a different level of experience. And, I think that that also translates into a more conversational paradigm. Specifically, if anyone has ever purchased an item on Instagram, you know. They are able to go through the entire flow within the context of Instagram and then get follow up messages within the messaging experience in Instagram… and the fact that so many other systems are still relying on, you know, other platforms that people just aren’t really, I think, using as commonly… which, you know. could be email and things like that. Obviously, email is still huge but it’s not people’s first preferred way of communicating. It’s starting to show that more and more companies need to think critically on how to meet consumers where they are at in terms of their evolution and their use of technology. And that I think it’s exposing a huge hole and gap in the marketplace where what the leading tech companies are doing is so far and way ahead of any other companies and organizations are, are just trying to limp along and offer.”
“I do think there is an innovation explosion on how AI is contributing to fight essentially the first worldwide global health crisis. I think the most obvious applications that a lot of people know are symptom checkers or contact tracing applications through using conversational AI, through using various messaging applications worldwide… However, there’s a lot of other healthcare applications like, for example, one is called the Butterfly Network. It essentially uses AI and an iPhone to take thermal imaging of lung ultrasounds in areas where hospitals may not have medical equipment and allow medical professionals to communicate to one another based on that body of thermal imaging data. There are, like, other applications such as Rokid which creates AI fever testing glasses for security guards to wear to scan symptomatic individuals in crowds. There are drones which use AI for contactless delivery of medical supplies and food both in China and Japan. There is even a company that is essentially using an AI and conversational NLP to enhance patient experience and ER department flows to reduce waiting times…”
Opportunity to Re-envision Overall Customer Experience
“We operate a lot of contact centers around the world. So, these call centers when COVID hit were basically shut down. And so, it was basically… it was essentially a supply shock throughout the entire industry. What it has done was open the doors, essentially, to all things digital. We already had a pre-study pipeline in chatbots, everyone interested in ‘how can I just make things a little bit better?’… But now, because voice is essentially shut down, there is this huge opportunity to re-envision what the overall customer experience ecosystem looks like…. There’s a lot of the assumptions of what voice or analog phone can bring to CSAT scores. Those have been basically dispelled… and everyone is saying: ‘I need to save money… I can’t have people come to the office… folks working from home don’t have quiet space to even talk so we can’t do voice at all…’ But then we also face these other interesting challenges like security so it’s a dual challenge where we are trying to provide a new solution opportunity and also adjust the data access in those pipelines.”
Great Time to Showcase Good Automation & Shift to Conversational AI
“So with a lot of the brands I work with, personally, we are seeing a huge spike in demand. Obviously, this is, I think, for our industry – anyone who is building animation, building chatbots – this is a great time to showcase good automation. And so, we’ve seen a huge spike in demand on our platform because when contact centers are shut down across the world, you know… Trying to connect with the brand over the phone just isn’t gonna happen and brands can’t go forever with a notification at the top of their website that says ‘You are going to experience long hold times’. Essentially, what you are telling your customers is ‘Don’t interact with us because we can’t stop..’. So, the shift to conversational AI has just exploded and so we’ve done some interesting things… ”
Broadening Technology Comfort Zone
“…for the first time, many people are getting comfortable speaking to machines often to people, like we are doing now… Many, many people. And I don’t think it’s much of a leap now that they are comfortable talking to a computer, now, to start to talk to an automated assistant, not just another human. So, I do think it is broadening the technology comfort of many people, grandmothers, and midwest. First time they’ve used this kind of medium to communicate.
Despite the increased visibility, adoption and innovation due to CIVID, chatbots still face serious challenges on their way to success. Some of them are as fundamental as bad reputation, others as complex as the danger of AI algorithm bias.
Lack of Understanding
“I think that as companies are starting to be more digital… So, there is a digital challenge over there. When I say digital, it’s not only companies becoming digital but… [creating] a culture of people understanding the tools, what they can do with these particular tools, what are the things they can leverage.”
“…When it comes to the strategy and product there are many things to face. We are not in the hype that we used to be like 2 years ago but we… but people still don’t get what is a chatbot. Is it a cognitive service? Is it like an AI? Can I use it to do whatever at my house? It’s a challenge to educate people about that and to show them chatbot is just a tool they can use in the right way [or] in a bad way. There are some practices and there are some things, small things [to look out for]… Sometimes it’s not about technology, it’s about human interactions like conversations, content writing, language. And most of the time, the providers need to have this sort of knowledge to help the clients with that. So, if the provider doesn’t educate their clients it’s probably… this company is going to face some misalignment of the expectations in whatever they are trying to deploy.”
Lack of Standards
“As we are kind of a new industry, I think we are still defining the best practices, I joke we still don’t know for sure which is our common sense equivalent and there is also a lack of awareness and of standards so it’s very hard to benchmark our chatbots and know if they are good or bad. Also for companies to know if their chatbots are good or bad but I see this as a challenge and as an opportunity as well.”
Making Things Easy Might not Always Be For the Best
According to Andres Pulgarin, making bot creation to easy often results in unsatisfactory results, bots that don’t deliver:
“It’s too easy right now to create a bot. It’s too easy. There’re a lot of tools, there’re a lot of platforms with which you can easily do a simple bot fast but this POC, this proof of concept is good enough but the problem is when you really try to put it into the company and it’s part of the real world process… And the problems begin there. They cross the line from POC to the production bot. The real bot is a challenge and not a lot of people are doing well in this part. That’s one of the reasons the bots disappoint a lot because they don’t really pass through the test of the real production.”
He highlights that one part of the problem is the speed of production, while the other, a more complex of the two, is the lack of data that is inherent to newly built chatbots. “There is not enough data that you can use to begin a bot… you have to expose the bot without this knowledge and [let it] learn from the users and this is the problem.” Many times, even if the companies do possess large amounts of customer data, the data is not adapted to the bot and so falls flat. The bot needs to learn from scratch and so is bound to create friction and unsatisfactory experiences before being able to evolve. But maybe by that time the damage has been done?
Omar Pera draws attention to the simple truth of bots having a bad reputation. Many people cringe just hearing the word and so, from the start, are closed to the opportunities it can offer:
“I think it’s a bad reputation and I touched that before but it really has a bad reputation. Whenever you talk [about a] chatbot, some customers will confuse it with a Twitter chatbot or with Facebook chatbot because they remember the word ‘bot’. So I think, overall, [chatbot] has a bad reputation. These days, actually, I call it more ‘fun assistant’ or ‘automation’ that is helping agents to deliver their work in a more efficient way. And, I will say that we need to work as a whole as an industry to focus on making it very easy for users to not get frustrated. I will say that reduce the friction as much as possible from your chatbot in case it doesn’t work.”
Poor Conversational Design
According to Muhammad Siddique, one of the key challenges is learning how to design conversation in a way that feels natural and, in fact, conversational.
“…there are challenges because a lot of people […] they have email marketing or other marketing… copywriting background… they just copy the same message into the chatbot. That doesn’t work. That creates a bad user experience because every environment is different, the medium is different. Chatbot has to be small bites, small messages. We, as an industry, we need to grow, we need to learn and adapt to the medium and what are the best practices for the medium…”
Indeed, still today, there are a lot of people who go about bot creation by copy-pasting copies and texts from websites and landings, making the conversational delivery awkward rather than helpful.
Forgetting that Conversation is Not Always the Right Medium
Amir Shevat talks about the old hype-trap and going conversational ust for the sake of it:
“I think there are places where the conversation is not the right medium. So, for example, if I am selling you a shoe I can talk from here to whenever and you will not be convinced. But, by showing you something really changes your opinion. So, sometimes a conversation is not the right interface. The way I look at it is through a conversational funnel, and I have that in my book… talking about like if I am doing a web interface… How many steps does it need to take until you reach your target? If you wanna do… you wanna buy a shoe… On amazon.com there’re probably five, six, seven steps and then you need to compare that with the conversational funnel… If I talk to you, when I do the same thing, how long does it take you to choose and buy the shoe? If the conversational funnel is simpler and easier, then you should probably check the conversational interface; And if the web interface or the mobile interface is simple then you should probably go with that.”
Social and Demographic Bias
Caryn Lusinchi sees the greatest challenge of conversational AI is ensuring that the data sets AI works with are sufficiently balanced:
“I do think that, I think Gartner predicted that… about 85% of AI projects will deliver erroneous outcomes due to bias in either data algorithms or the teams responsible for managing them. And I do think COVD is putting certain economic competitive pressures to force companies to rely on AI to automate decision making that humans used to do… and I think that presents a lot of moral, legal and social consequences […]”
“I can give a few examples how this might play out across industries… namely healthcare. […] …with the rise of public surveillance initiatives happening globally, there is collection of biometrics and location data and, you know, feedback from the public on how infections are progressing. It’s really biased to the technical savvy who have smartphones and data plans globally and for instance, a 75-year-old grandmother who doesn’t really know how to use a phone for anything other than to talk to someone else you know might not be included in some of those data sets although that population is the highest risk group […]”
“…a lot of times there is an IoT voice assistant, as well, that is engaging in conversation. There are Siri and Alexa and a lot of them are trained to give responses based on huge databases of recorded speech that essentially is dominated by white upper-class Americans. So, sometimes, it might be challenging for the technology to understand commands and conversational… nuanced inputs whether they be emotional or very complex from people outside of that demographic, so I definitely think there definitely needs to be some more diversity in AI data sets that are training a lot of these AI applications.”
Adam Chayer points out another challenge standing in the way of full-scale adoption: fragmentation. He believes that in order to create the true conversational experience, we need to work together towards a single assistant that can work and move across services, brands and products seamlessly:
“Well, conversational AI, the industry today, we are too partitioned, too fragmented. Alexa and Google, third-party skills are all independent of each other, built by different people and they feel very different. Language is horizontal so if I should be able to stay ‘Near my house, three Thursdays from now, do something’, I should be able to do that everywhere. If the assistant knows where I live, it knows how to do three Thursdays from now… that’s not the case with most of today’s technologies. Bixby, I think, is trying to overcome this to make it feel like one assistant that can do everything using the brands and services rather than 10 000 assistants […] …in case of Bixbi the knowledge is shared between apps.”
If this is accomplished, Chayer claims: “The [conversational] assistant can be more powerful than web or mobile because it can integrate information across different services which is important for any task. But today our technology is not there yet.”
So, the question stands: What is the future of chatbots? Where will the conversational interface be five, ten or twenty years from now?
To spice things up and highlight the variety of thought processes shaping the industry, I asked my guests to not simply share their opinion about the future of chatbots, but share a contrarian view. One, they believe is against mainstream beliefs.
Caio Calado believes that rather than bots becoming the solution to everything, we will eventually accept them for what they are – a way to get to the solution:
“There are some people that say that chatbot is the future, it’s going to be the medium for everything and it’s going to solve all problems. I would say… honestly, I don’t think that chatbot is going to be THE experience. It’s not going to be THE way, THE solution. I think that chatbot, in the sense, is going to be more the way to the solution. It’s going to facilitate entries to the particular solution. So, today, most of the bots, they are not integrated, they don’t do many transactions and things like that. They are just like a way for other mediums and there is the idea of multi-modal conversations and the idea of omnichannel, as well. So, I think that chatbot is going to be the way, not the solution. It’s going to be the way to the solution.”
Cristina Santamarina, while agreeing with the general notion of NLP becoming a commodity, does have a couple of contrarian opinions. The first prediction concerns WhatsApp, the hottest conversational channel on the market today, which she believes will not be the main conversational channel of the future. The second prediction is about the real value in the future not being in the tech but in the soft-skill of conversational design:
“My bet is that the NLP is gonna become a commodity. I also believe that WhatsApp will not be the main conversational channel, in the long term. And, I believe that the value in the long term is going to be in the conversational and agent design and not on the technical side.”
Why doesn’t she bet on WhatsApp?
“So, I believe that chatbot, that chatbots in WhatsApp are gonna overload the channel and we will look for other channels for our personal communication for our important communication and that is gonna reduce dramatically the open rate of the WhatsApp messages. So right now, everyone wants to have a chatbot on WhatsApp because that’s the channel that everyone uses but Imagine what will happen when we have a hundred chatbots writing us there…”
Indeed, this scenario is not unlikely as many users do strongly oppose the marketization of personal messaging platforms and messaging apps. However, it might be just a question of right execution where convenience overpowers the less pleasant aspects as WeChat achieved in China.
When it comes to conversational design, agreeing with Santamarina is Omar Pera (Kustomer): “I believe it [the future] is not about the best technology, who wins for a good, good experience. I think technology matters but what matters more is to have a good conversation design […] to reduce the friction as soon as possible in order to contact agents.”
Roger Kibbe paints a different picture of the short-term and long-term expectations about the future of chatbots:
“From an AI perspective, long term, I’m bullish. I think AI is amazing, I think it’s gonna unlock amazing capabilities, it’s gonna help humanity, but I will say in the short term, I could be a little bearish about some of the things going on… I really think the hype has kind of outlived the reality. You know one of the things I […] AI isn’t very smart. I don’t mean to insult anyone now but really it’s just grabbing swaths and swaths of data and looking at patterns. A lot of what AI is, really, machine learning. Right, behind the scenes there. So it’s really good in pattern recognition and understanding what’s happening in the past and then saying OK that’s what will happen in the future.
But sometimes what happened in the past isn’t necessarily what will happen in the future. Sometimes, what we call… it might be like an anti-pattern […] Humans, we have an amazing ability to kind of synthesize our past experience with the present and go and say ‘yeah that’s applicable here’ because XY and Z or ‘No, things have changed, so even if this happened in the past, it’s not gonna happen now’. The machine on the other hand, the AI, the ML particularly, it’s gonna be like ‘OK, 99% of the time it happened this way in the past, ergo it’s a good way to proceed in the future’. That’s not necessarily true. There is this contextual ‘where we are’… ‘what we’ve learned’. There’s past errors and things to be done differently. So, I think when I hear about AI and you know being this like an amazing tool that’s gonna revolutionize things. Absolutely! But I think, in the short term, we gotta be pretty careful… We’ve seen some examples right with kind of biased training sets […] So, I think, from an AI perspective, I’m bearish on the hype bullish on the tech.”
Speaking of bias, Hanlin Fang puts the spotlight on the importance of ethics in AI development:
“You know, technology is very powerful. The ways the computing power, ways the data, the deep learning, the technology can empower us to do so much more. So, for me, it’s not about what we can do it’s about what we should do and what we should not do specifically, right? So, I believe it’s not just… within the conversational AI, it’s also applied to general AI or machine learning strategy… So I feel like it is a moral issue more than a technology issue. So for me that’s… I’m very cautious about that.”
Social and racial bias in AI is not a new topic in the AI industry. Businesses and people building AI solutions need to approach the algorithms and machine learning to avoid bias. So for Fang, the future lies in the ethics of it all: “How you think about the social factors; How you try to avoid the bias during the machine processing… how can we… make machines more trustworthy, right, because we tend to trust machines more than before…”
And she is not the only one thinking about the negative consequences of inconsequentially trained AI. When asked about the future of chatbots, Caryn Lusinchi shared:
“I tend to agree with a lot of AI scientists such as Hawking and Musk that kind of foretell the end of the human race with AI. There are definitely sci-fi visions of how this machiavellian machine world kind of plays out whether it’s Westworld or Terminator or Battlestar Galactica… I think humans have very good intentions when going into things and think it’s very easy to blame technology but I think it’s important to realize it’s us. And that machines really mimic and receive inputs from humans, you know, we are kind of the flawed evil ones. So, we just have to think very carefully [about] how we are approaching this new technology ‘cause I think it’s a fine balance.”
Last but not least, contrary to the popular opinion of making conversational bots and assistants more niche-focused, Adam Cheyer argues that the future is one assistant that will act as an access channel to all the services and tasks we need and use:
“…I think we are still very much at the beginning. There’s so much that needs to be done but I think the potential is huge, maybe that’s contrarian, or not… But what is contrarian is that we are going after ‘I want one assistant that can do everything’. Many people in the industry think ‘I’ll have a million apps, each app with their own voice interface’. Those are two very different views. I would say most people in the industry today go to the ‘million different voice assistants’ because it’s easier to build. But every system that I have ever built… if you go back over… this is 27 years I’ve been working in this space… Every assistant is… I want a user to have one assistant that I can say ‘I wanna know this or do that’ and it will help me coordinate tasks across all of the services that I want, that I trust, that I use. So customized and personalized for me but it really is a one assistant as a tool to help interact with everything, not a million different assistants and I think that’s still contrarian.”
Wolf Paulus’ contrarian opinion centers around the rise of the voice interface as the ultimate conversational experience:
“I think that chatbot experience will fade away. For me, chatbots are blackberries of our time. So… remember when blackberries were the predecessors of our phones? They were very limited. They were basically machines that allowed you to read and write email. At the time, we could not really imagine that people would want to touch their screens… when the suggestion first came out that Apple is building a phone and the whole screen is a touch screen and there are no keyboards, we thought these guys are nuts. So, now I think chabots are…. we are in this phase where we are transitioning from traditional user interfaces to more conversational user interfaces. But, to me, conversation means something we currently do, a conversation means a face to face voice interaction…. you don’t really have a good idea of how this conversation continues and this is what makes a lively conversation. If i just put two or three buttons on the screen, this replicates, for me, an IVR. So, I think, chatbots as we currently see them, is a transitional phase towards a much more open interface…”
He further argues for the cause of tackling the question of privacy and security which is one of the biggest concerns when it comes to voice interfaces. He highlights the opportunity to use voice to increase security, not just in the digital world but also physical:
“Privacy and security always comes up with voice and we as an industry absolutely have to address this issue. One of my hopes is that, out of the current situation, voice biometrics becomes better and more popular. So, it’s… It’s very accepted now to use your fingerprint to unlock your phone. It’s also very accepted now to use your face ID through face recognition to unlock your phone. What… where we currently do not have good success with, is using your voice. And there are companies that are good at that. Companies that have solutions for that and I could see that as a very good solution to address privacy and security issues on these smart speakers. So, for instance, we already have user identification with smart speakers that’s just… user identification just means […] the system can differentiate between family members. So, it knows it’s the mother, the father, the kids are in the room and talking to the device… it’s not voice authentication, where we really want to authenticate a user with an account. So out of all that… That currently with the face masks, face ID is not that good… Touch ID might also not work everywhere, we are discouraged to touch things… So that could be the big chance for voice biometrics to come in. So that could help us make systems more secure only allowing you to access information to certain accounts if the user has been authenticated.”
Since we are talking about contrarian opinions, Adam Cheyer and Chad Oda both express their belief the future of chatbots being about multimodal conversational experiences.
Chayer explains: “When I worked with Steve Jobs here is how I explained it to him: If it’s on the screen the best way is to touch it… if there is a button… or… just click it. And designers work really hard to put the right things on the screen. But if it’s off the screen, the best way is to ask for it and the ultimate interface is multimodal where you can be asking and clicking and using the screen and using your language without thinking about it which is what humans do every day. So it’s not just about voice, it’s about context, conversational context, and it’s about using the best modality for the current circumstance. If I am jogging with a watch or driving in a car voice is very important because I can’t use my eyes or my hands. But if I can use my eyes or my hands, I should use them cause it’s very quick and easy. So conversational, multimodal, contextual… that’s what’s important, not the voice so much…”
He further elaborates that the advantage of multimodality is not just about having the right mode circumstance but also having the right mode for the user’s level of knowledge regarding a certain interface/concept:
“[…] Voice has strengths and weaknesses and touch and visual interfaces have strengths and weaknesses. If you are a new user and you don’t know what you can say you wanna browse menus to discover you can’t do that with voice. Once you are an expert user and you know what you can do and say voice is probably the fastest way to get it done. You need a very… you need to transition from new to expert to intermediate. And we are all new in some domains, expert in other domains and intermediate in some. So we need a seamless interface that lets you transition through visual and voice, you can’t just have one.”
Similarly to Cheyer, Chad Oda expressed a similar opinion warning against a myopic view of the new technology:
“So, I think, there’s a tendency by pretty much anyone in any new emerging technology to begin to have this certain myopic view sometimes of like how we sort of look at the technology and our expectation of experiences. So, essentially what I mean is, I think that we shouldn’t try and force total use cases […] We should sort of step back from looking at the interfaces with the modalities whether that be voice, whether that be chatbots and not try to force everything on one specific modality. Because I think what we have to see is that chatbots and voice are not a panacea for every single customer touchpoint on the customer journey and not every single use case. […]
Essentially, if we think about each interface individually, each interface has pros and cons in regards to what can be successfully executed on that one single interface at that point. […] So, I think… the future lies in thinking about how we can integrate multiple different interfaces at the same time, essentially balancing the pros and cons. Because, if you can leverage multimodal interfaces, you can effectively smooth out the overarching user experience by sort of taking the best attributes from each interface… and that’s where I see the future actually morphing into. It’s just a matter of like if we are thinking about that today, so we can position ourselves for tomorrow.”
It’s clear that COVID-19 has accelerated digitalization at an unprecedented level. The demand for communication automation is skyrocketing across all industries… However, I feel the existing chatbot solutions have failed to meet the expectation by only focusing on automating the conversations without adding any deeper layers… At Landbot, we believe that the key to a frictionless conversational experience is about offering to the customer relevant information on the right channel in real-time…
NLP and AI are important and the Landbot platform isn’t forgetting about them. However, instead of heavily relying on complex NLP technologies, with the rise of no-code solutions businesses today demand an easy and flexible way to iterate the experience based on actual customer interactions. That’s why we are proposing conversational apps as a natural evolution of chatbot… they combine the benefits of conversational interface with rich UI elements and are able to automate advanced data workflows like a business application. So, in a way, they are able to address business needs more than just on a communication level but actually help automate entire business processes.