Podcast: Download
Subscribe: Apple Podcast Google Podcasts Spotify
Visit our Marketing Nerds archive to listen to other Marketing Nerds podcasts!
In this episode of Marketing Nerds, SEJ’s Chief Social Media Strategist, Brent Csutoras, was joined by Heidi Young, the principle engineering lead at Oslo, an AI program, to discuss the future of conversational search and how AI could help create a true virtual assistant.
Here are a few excerpts from Brent and Heidi’s conversation, but be sure to listen to the podcast to hear everything.
How Far Are We From Having a Truly Virtual Assitant?
We very much are in the early stages of it, but I would say what makes today very different from even five years ago, is that technology has kind of been catching up and the market it ready for it. The fact that Google has worked so hard, over the years, to digitize the world’s information, organize the world’s content, and that Facebook has learned so much about you, and Amazon knows everything that there could be to know about products, you have this ocean of information that’s specialized and personalized, all at the same time.
We’re hitting this interesting point in time where now is the right time to move to the digital assistant, whereas, maybe five, ten years ago, certainly not, because accessing that information was just way, way too hard, whereas, now it’s very easy. You can connect to an API, or the information exists in Wikipedia, or in your Facebook feed, or something like that. I think now is the time and people are also ready.
Because people have moved off of the desktop and they’ve moved into their phones, they’re walking around with this almost digital assistant in their hand, now is the time to start pulling all of that information together and making it a natural product.
How Does Personalization Work? Is a Learning Algorithm That Adjusts Over Time?
You know, what I spend the majority of my time doing, is the actual interaction, measuring the interaction. I’m more on the metrics, analytics, measurement, ML, IR, training data side of things, so I’m looking a lot at the actual interaction itself and one place that I think we’ve failed in the actual measurement of how a conversation succeeds or flows. I think that piece is where these other companies are falling over. They have all the information about you, right?
They know all your in-session contacts, they know the time of day, they know your location. In some cases, they know your Facebook profile. They know all of your session history leading up to that point so that they can build this rich user profile on you, but the actual interaction itself just really fails.
I think where these companies are failing is, “Okay what is the next natural reaction?” How to anticipate where the user is going next. Using that to help you refine your experience and navigate that ocean of information, narrow it down, and then make sense of it, and then act.
Where Is the Future of Virtual Assistants Heading?
If you look at where digital assistants are going to take us, they’ll replace that ten pages of apps you have on your phone, and you’ll just interact with the digital assistant. Where I think places like Alexa, or Siri, Cortana, the Slack bot marketplaces, the Skype bot marketplaces, they’re offloading the tasks and what we’re trying to do is understand the world of information that exists and then help you execute on that task.
For example, let’s go back to my query about restaurants near by. Now, I’m able to refine and chain those queries together, and I come upon a set of results, and then I choose one.
At that point, the next question might be to request a ride there, or book a reservation or something. Let’s say it’s request a ride. I may want to offload the task of actually requesting a ride, but the difficulty in arriving at that decision was offloaded to the assistant. I didn’t have to query five to ten times on Google, navigate between Yelp, TripAdvisor, Facebook, Four Square, go back and forth with my friends to decide which one we wanted to eat at.
I was able to arrive at that decision quickly and then execute on that task naturally. We believe the place to hand off that task is probably a bit later and to help the user synthesize the world’s information is really what the digital assistant needs to be taking on.
What Are Some of the Issues with Conversational Search Right Now?
We’re focused on the “I’m hungry” problem. Somebody opens their phone, “I’m hungry.” That can take a completely huge set of different paths. Does this person want to cook and stay at home? Do they have a lot of money? Are they looking to go out? Is it now? Is it later? Are they looking to pick up take-out for their kids? “I’m hungry,” is a huge problem that can go in a variety of different ways. Our job, and what most digital assistants do, is they do this kind of slot filling or path navigation, where they’re trying to lead you through that path.
What we’re trying to do is understand what it is you’re saying, with the context we have about you, things like your location, time of day, the entry point, are you coming in through the app or some other entry point? We take all of that and then your immediate session context, which was the first thing you issued in your query, to try and navigate you to the right task.
The problem with this, and the reason it is so nuanced, is nobody wants to sit there and have a discussion with a chat bot for 12 back and forth turns, when they can just type two or three search words in Google, step back, look at the entire set of information, and then make a choice.
Our real success, and what I have to focus on, is how to do that in the most natural way possible. When you say, “I want to get an Uber,” at the end of the query we just discussed, just booking an Uber is one way to go or to ask them, maybe the next natural question is, “Did you want that for one or more people?” Right?
Usually, users don’t mind that one to two additional refinements, but they certainly don’t want to go down a customer service representative type of experience.
To listen to this Marketing Nerds Podcast with Brent and Heidi:
- Listen to the full episode at the top of this post
- Subscribe via iTunes
- Sign up on IFTTT to receive an email whenever the Marketing Nerds podcast RSS feed has a new episode
- Listen on Stitcher, Overcast, or Pocket Casts
Think you have what it takes to be a Marketing Nerd? If so, message Kelsey Jones on Twitter, or email her at kelsey [at] searchenginejournal.com.
Visit our Marketing Nerds archive to listen to other Marketing Nerds podcasts!
Image Sources:
Featured Image: Created by Paulo Bobita
In post Image: Deposit Photo
