Conversational AI Trends To Watch In 2021

Watch The Event Again

Event Q&A

Q: Can you talk about the ecosystem of fairness and explainability?

Karthik: So increasingly, as I mentioned, the usage of AI models will become more common. And the next time your insurance claim gets denied, you really wouldn't know if it was a human or if it was really a claims officer inspecting or if it was an algorithm that decided and the claims officer will tell you "no," then what do you do?

Karthik: So it's becoming very important as certain technologies become transparent to really look, look inside the box, right, and see how that decision was made. What criteria did you use? Did you use things like someone's if you use someone's driving record and risk profile? Sure, but how did that risk profile come? Was it before? of the race. Was it because of their ethnicity? Was it because of just their past driving record? Which could be fair, but anything else would be unfair? Right? Was it because of their age? Some that could be fair, some which could be unfair? So it's become very, very important to question those decisions. And I don't know how many of you know that today explainability in AI is a major challenge. So while these algorithms provide incredible results and provide very accurate results, they're also learning from the same biases and the history that we humans have created. They're learning from the data we generated, which means they're carrying forward those biases, right. So in the past, at least, you could find the human, and you know, potentially hold them liable for decisions they make. Whereas when it's these transparent algorithms making those decisions, it's no longer cool. Right? At that point, you have to be able to question what the criteria are. And as a technology, deep learning isn't the most explainable. If any of you have used it, it's, in many cases, difficult to almost impossible to be able to explain how that algorithm made a decision, right. And there's a field called explainable AI that's emergent that's really trying to, in some ways, reverse engineer these models and trying to understand, you know, what variables impact a certain decision and how did the algorithms I tried to build some reasoning around the model, and that in turn helps drive explainability which in turn brings accountability to the to these decisions being made.

How should people in the enterprise feel about the adoption of Conversational UI?

Karthik: So at a certain point, just like, you know, having a website wasn't an option anymore, right? For a lot of enterprises, having a conversational UI wouldn't be an option anymore. They would have to do it, right. That's the next frontier for automation. You know, good or bad, you know, that's judgment. But that is the next frontier. A great example is all of those one 800 numbers that we love to hate to call, right. And those incessant wait times, you know, potentially all of that can be automated COVID was a great example of what happens when that system collapses, right? People are on hold for hours together, you know, why couldn't you automate a lot of those calls? So that poses that question, and like I mentioned, based on the industry, for a lot of companies having that will be a capability. It's no longer an option of should we do it? Is there a business case? Right, you pretty much have no option, you have to, you know, I think there's a funny meme that was going around at the peak of COVID, which was, you know, who, who accelerated your digital transformation, your CDO CIO or COVID. And, you know, but the check around next to COVID. And that's kind of what's happening, right, like, people realize that some of these, they have to have it's table stakes. Now, having said that, how would companies go about this, the focus, and via seeing this on the field, right, and be talked to these enterprises, like I mentioned, a lot of my discussions as the CTO, you know, up until a year back would all be about the algorithms and talk me through it, walk me through it, show me as they are, and you would have these kinds of discussions, which would really create a lot of friction in the adoption process. Today, it's very different. And then through, you know, through most of 2020, you know, every single deal we've sold, or we've grown, you know, companies don't really care how you use it, they want to make sure what KPIs are you are addressing and make sure you have the ability or you know, the capacity to address it, right. So, you know, as companies as enterprises, the focus would be more around, what is the product we are rolling out? What is the customer experience? Because as technology gets commoditized, the battleground is customer experience. The battleground is no longer the technology because the capability is available. Even terms to everyone. So, as enterprises, heading into 2021 will be very much about looking at AI as a capability. Now, what will be a new ecosystem emerging is built by decision, you know, do we outsource that do we do it, how do you train and maintain the AI? Right? So getting AI capabilities will be very easy snap of a finger. The question is how do you maintain it? How do you make it better? The question will be how do you design around this new capability, right? So, it will be a lot of design hires, it will be a lot of optimization hires, potentially setting up ml Ops, those would be some of the capabilities and again, you based on the size of the company will either choose to outsource it to companies like wisdom, or you choose to take it in the house if you're big enough, but you wouldn't be hiring data scientists to write new algorithms like companies were doing up until today.

I was wondering to which degree using speech markup in your content in your online content influences the voice SEO versus the local markup and schema? And to what extent it is imperative to use it or not?

Georgie: I think because it is so new, exactly. Like you've said, It is definitely worth testing out. Because I think that Google is definitely testing a lot of things out. And it's also worth noting that at current, we can't, from an SEO perspective, we can't actually retrieve what data is being pulled in from voice assistance, which is extremely frustrating we can kind of make, we can make some judgments within our keyword research tools. And for example, if it says like, hey, Google, or if it's like, extremely local, however, I would recommend, it's a bit of a like a layered strategy. So having the technical aspect thereof the structured data, but then also having the layered content strategy of having a look at the questions that are being raised within the feature snippets that people also ask boxes, and then have kind of like trying to beat the customers that are currently ranking within those positions, and then hopefully being pulled through into voice assistance in that way to try and to get like every single angle to tobio capacitors. 

So Facebook had a recent release. They will be having an assistant that summarize news for people, any opinions about how that might impact SEO might be specific to an industry? And the second question is, should we invest any energy in BING? 

Georgie: So I probably say it's, it's really interesting how the search engine results page changes with the query. So if you can see it, like the top a little bit, it will kind of change between shopping news images. And so it'll be interesting to see how Facebook is being pulled into that sort of section. Google probably does prefer Google stories, because that's just the way it is. And more authoritative websites. If we think about it from a UK perspective, does BBC or those sorts of really, really authoritative sites are probably more likely to rank them? 

With Google versus BING it is an extremely interesting one from an agency perspective, why them for Google, just because that is where the largest market share from an SEO perspective has been. But there's definitely as a use case, for BING we think about consumers who go there, they're more likely to convert at the bottom of the user funnel. They know what they want a lot of the time, I would definitely have a use case for BING as well, if we think about it from a voice SEO perspective, as shown, the market share is pretty much 5050 and only expect that to grow as well. So yeah, I'd say exciting things to come with. I want to split it between both.

You had mentioned, multilingual, multi-person, as well as a voice for good. I was wondering if you'd had experience doing that with reaching out to the public or specific, like an example of techniques that you do, because we're always looking for new ways to engage non-technical people with voice design, and being a part of the practice to help us.

Maaike: Yes, one of the things that I do, multilingual, multi-capable, is the voice acts design sprint, which is a framework that I designed around the Google action sprint but incorporating more serious play, more data and user research upfront, and specifically tailoring also to the fact that you can be multi-capable, multilingual and so on and so forth. It's specifically made for that and has a lot of roleplay in it, roleplay for me, and serious games. Those serious games have really in my opinion more than anything else, more than a lightning talk more than explaining anything. It's having people live and walk in the shoes of another person. And that empathy that has sparked the most significant aha moments. And I remember one brand where I did a voice actor design sprint. And it was really funny because they had very strict thoughts about what they needed to be doing. And, and it was about using voice for an appliance use case. So a bit of IoT. And they thought, yeah, this is going to be great. For our IoT devices, and evil people that have difficulty seeing will be able to interact with it, the only thing that they hadn't thought of, and that they then knew, because we put wood, we gave them some, like fairy fuzzy classes that you can have that you can find in kids games, actually. And, and they said, Okay, well, yes, I can hear what I need to hear. But actually, now I don't know where the rest of my things are and actually do what the speaker is telling me to do. And that was a real aha moment that they would have never had if they hadn't engaged in that Roper engaged in those games, those conversational games. So yeah, I think walking a mile in a user's shoes is always a very good idea. Sometimes even if you have the user research, even have you have done the user interviews, and you're reporting on those things. You're saying, well, we interviewed users, they told us this and this and this and that. But as long as I always say, you need to understand it with your mind when you do a workshop. So someone is explaining what is going on and the analysis of user research report, for example. And then you also have to feel it in your body to really incorporate the knowledge that you're getting to say then, Yes, I fully understand what the problem is that I'm engaging with.