SILVIA Technology – Intelligence On Command Interview

Interview Continued

Conversational intelligence, learning through real-time experience, applying logic to solve solutions – all of which would do well for enterprise solutions. The one thing that appears to set SILVIA apart from prior and existing technologies of a similar goal – chat bots, automated response systems, “Siri”, Amazon's “Alexa” for instance, the one thing that seems set this apart is the focus on providing a rich user experience which learns through experience, as opposed to being a platform-specific.

AM: I’ve used Alexa pretty extensively, and I’ve played around with Siri, Cortana and Google Home as well. From my perspective, these devices are entirely focussed on utilitarian functions. They check the weather, they play music, they can turn the light on and off, maintain a shopping list, set a timer, and search the web. And they do these things quite adequately. But these devices are stiff and robotic. They mostly just respond to user input, and can’t really hold a back-and-forth conversation. They cannot lead the conversation, or act beyond their interrogative nature. They cannot synthesize their own responses, or vary their language.

I believe that as conversationally intelligent systems become commonplace, the engagement factor will not be about how well they can predict the weather or if they can find you a great recipe for pumpkin pie. It’s going to be about how the virtual agent interacts with you, how it evolves over time, how it personalizes itself based on your needs and desires, and how it ultimately makes you feel.

That being said, how and where could the technology benefit mobile devices which have adopted similar technology? These devices rely upon pre-scripted responses and search engine-dependencies – utilitarian functions as you said. SILVIA is an intelligence engine. Learning.

AM: SILVIA has the same capacity to act as a virtual assistant. The tools have C# built in along with a full kit of API functions, so developing those types of behaviours are not technically difficult. But where Siri and Cortana are specifically made to function as assistants, SILVIA was designed to do much more. Ultimately, we want to develop an experience that is closer to the movie “Her” than it is to Siri or Cortana. This puts our focus on the user experience, and designing our avatars to be smart, humorous, adaptive, and engaging.

We also have proprietary voice technology that allows us to create virtual agents who can speak in a real voice with significantly less cost and overhead than traditional voice acting. This enables us to create custom applications in all sorts of areas you’ll never see Siri or Alexa. For example, we once created a talking Orc that could tell you about World of Warcraft, the responses coming straight from the WoW new player guide. It even spoke a few lines of Orcish, and would occasionally call you a puny human.

Disrespectful Orcs and movie influences, I love it. The user experience and intuitive functionality of the technology appears to be of high importance, in order to differentiate from those such as Siri. As a virtual assistant, how or where do you see SILVIA benefiting a traditional computing environment – if any? Technologies such as touch screen displays in-combination with application such as Cortana attempt to establish some kind of relationship between the user and the OS, but more times than none, it's unintuitive, too slow, or simply ignored in favour of tradition.

AM: I see SILVIA acting as an intermediary and personal assistant who learns about you and understands your desires and needs. One of SILVIA’s strengths is her inference engine, which gives her an ability to understand a wide variety of user input without having to filter for every possible permutation of a question. You can talk to her in broken English, and so long as the appropriate linguistic concepts are present, then she will respond properly.

Often times with other conversational intelligence systems, they can come off as robotic and limited in their responses. If you don’t give it just the right input, then it doesn’t understand what you want. I’ve also seen rigidness in responses, and a tendency to provide the same canned output over and over. SILVIA is designed to be much more of a natural conversationalist. She can synthesize her own responses, and augment her language to vary her output. While it’s perfectly acceptable for an artificial intelligence to not understand every input that it’s given, how it responds when it doesn’t understand the user is just as important as how it responds when it does.

Even when the AI doesn’t possess a good answer, it should still seem intelligent in the way that it responds. When we combine SILVIA’s inference engine and varied responsiveness with her ability to learn about her user, we can develop a personal assistant that is more than just a robot in a box. We want SILVIA to feel like an extension of the user, and give her the ability to develop a relationship with the user that is similar to what we see between Iron Man and Jarvis.

You briefly drifted in to the topic of World of Warcraft – your career history, your passion resides within video games. Needless to say World of Warcraft isn’t exactly a small fish drowning in a lake. It’s one of the oldest and biggest titles – forever going on strong. The aspect of SILVIA which holds the most interest to me is how it can be used for the application of gaming.

When I, as do most, think of conversations and artificial intelligence in gaming we think of RPGs – branching conversational choices consisting of text boxes and dialogue selections, based on the limitations of what’s been set by character creation tools or developer intentions for the sake of a pre notion plot, narrative, theme, world and character. Intelligent computing however, which learns through experience. That holds potential to change the way we play these type of games. The way NPC respond to situations, conversations and player behaviour.

AM: RPGs are a great example of a game genre that SILVIA can push to the next level. With SILVIA, we can skip all the text boxes and all the dialogue trees. We can remove the interface part of the mechanic, and make it purely conversational. Imagine a character in an RPG who you need information from. You take him to the tavern and start feeding him drinks. When he gets drunk enough, he becomes susceptible to your suggestions. With enough coaxing, maybe you can get him to divulge the information that you need.

Or imagine a survival game where you’re in a hostile environment, and your only contact with the outside world is with someone on the other end of a radio. You have to speak with them, describe your surroundings and your obstacles along the way, and the character on the radio helps guide you out. The really nice thing about utilizing conversational intelligence in a game setting is that your characters don’t need to get the weather or tell you road conditions; they don’t need to have open-ended conversational ability.

Like any character in a game world, they only need to know what you want them to impart to the player when the conditions are right. And since this is what SILVIA has been designed to do, there’s no worry about maintaining dialogue trees and state machines. SILVIA handles it all under the hood, allowing developers to focus on their character design rather than heavy technical implementation.

In some ways, I see these experiences hearkening back to the old days of Infocom text adventures, when all you had in front of you was a black screen and a cursor, armed with nothing but your imagination. Those games offered a mechanic for discovery that we are bringing back with SILVIA. While a game character can tell you exactly what it needs to say, it can also have content that is deeply buried within its digital brain, which can only be accessed through ongoing interactions. Think of a game like Fallout 4, only every character you come across has his or her own story in its head. From a pure design perspective, this opens a whole new world of opportunity.