SILVIA – An Intelligent Assistant Platform Whose Time Has Come

SILVIASometimes you’re too early to market with an idea or a technology that’s ahead of its time. That might have been the case with Cognitive Code’s SILVIA platform. But time has been catching up and SILVIA is well positioned to meet growing market demand. The brainchild of Leslie Spring, former Principle Software Engineer at Sony Pictures Digital, SILVIA is a conversational agent platform that was developed years before Siri appeared on the iPhone. In fact, when SILVIA first saw the light of day, smartphones didn’t even exist.

SILVIA has matured since those early days. Now her timing couldn’t be better. That’s the sense I got during a recent phone interview with Mr. Spring. It’s not just that Siri, Google Now, and other intelligent assistant apps have made people aware of their compelling possibilities. It’s more about what Spring refers to as a convergence of technologies. More on that later. First, though, just what is SILVIA and what makes her special? (Note: I’ll refer to SILVIA and her technology as “she,” because that’s the pronoun Spring used during our discussions and it seems to fit).

SIlVIA is actually an acronym that stands for “symbolically isolated, linguistically variable, intelligence algorithm.”  That’s a mouthful for sure, so let’s break it down. “Symbolically isolated,” Spring explained, refers to the unique method SILIVA uses to process language input. SILVIA doesn’t work in the same way as chatbots built with AIML (Artificial Intelligence Mark-Up Language). The AIML runtime platform takes words as input, matches these words to patterns in the chatbot’s database, and issues predefined phrases as output. SILVIA takes words for input, but she transforms them into a mathematical language all her own. As speech or text enters her brain, her runtime system isolates those words into an intelligence core based on symbols and statistical relationships. This transformation is a bit of a challenge to imagine, but it plays a key role both in SILVIA’s capabilities as well as in her potential commercial applications.

“Linguistically variable” refers to the fact that SILVIA isn’t tied to any one language. Because the SILVIA brain thinks in terms of mathematical units, SILVIA is a polyglot who can easily switch from one language to another without requiring special programming.

The “intelligence algorithm” part of SILVIA’s name refers back to the statistical runtime of the platform. Because SILVIA doesn’t rely solely on simplistic pattern matching, she is able to conceptualize meaning from input and dynamically generate new output. With the ability to connect concepts and ideas, SILVIA can engage in something akin to a real conversation. She’s also easier to “train” than a normal chatbot, because you’re not required to pre-populate her database with every possible question and answer. You also don’t have to account for misspellings or different ways of asking the same question. SILVIA’s flexible intelligence core can sort those things out automatically.

The SILVIA platform consists of the runtime, a developer studio, extensible plug-ins to connect to data sources or applications, and brain files that control the app’s behavior.  The built-in database makes it possible for SILVIA to learn and remember the preferences of those she serves. She can use this stored knowledge in a number of ways, including to infer user intent and to proactively engage the user.

With all the power this technology offers, why haven’t Cognitive Code and the SILVIA platform taken the world by storm? This brings us back to Mr. Spring’s theory of the convergence of technologies. For years, SILVIA’s powerful brain and conversational processing capability didn’t have the benefit of today’s voice recognition software, the advances in natural language processing, and the advent of smartphones and lightweight apps. Now that this world of intelligent, speech-ready assistants is finally here, SILVIA may have a key advantage over potential rivals.

What’s the SILVIA advantage? Our future is likely to be filled with smart machines that simplify and enrich our lives in ways that are currently hard to imagine. The “Internet of Things” (IoT) is the buzz phrase used to describe a world of objects embedded with computing power and capable of communicating with other objects. In the world of IoT, home appliances are aware of their surroundings and their current state. They can communicate with each other, with service centers, and with the homeowner. In the IoT world, you can talk to your thermostat about what you’d like it to do while you’re away on vacation, chat with your crock pot about recipes and cooking times, and converse with your TV about the evening’s viewing selection. While all that is going on, your smart health wristband can communicate vital information to your physician.

What separates SILVIA from other intelligent assistant apps or virtual agent platforms is her ability to run natively within a small device. With her compact runtime environment, SILVIA has a minimal footprint and doesn’t even require a connection to the cloud. SILVIA technology could be embedded in your TV, for example, giving you access to her full range of capabilities. The same thing goes with SILVIA on your tablet, in your car, or embedded in appliances and devices throughout your home and office.

The time is ripe for smart virtual agent platforms. Technologies that were long in the making have matured and converged and it’s anybody’s guess what the intelligent assistant market will look like in the next few years. If Leslie Spring’s enthusiasm about the possibilities is any indication, SILIVA’s future looks very bright.

Yahoo Working With Robin Labs On Intelligent Assistant App?

RobinTechcrunch reported last week that Yahoo might be working with Robin Labs on a Yahoo-version of the white label Robin Labs intelligent assistant. According to the story, the Siri-like virtual assistant was supposed to stay under wraps for a while, but the story surfaced after a video with the Yahoo-branded app was leaked.

Robin Labs has developed a technology platform that incorporates speech recognition, natural language processing, and task-based building blocks that it refers to as task agents. I looked through the Robin Labs website to see if I could detect whether they use a third party solution for speech recognition and NLP, but I didn’t see any references in that regard. That leads me to surmise that Robin Labs has its own proprietary NLP technology, but I could be wrong.

The task agent platform is called Robin.AI and consists of granular components of functionality such as messaging capability, note taking, calendar administration, traffic and weather. Non-programmers can apparently build their own custom intelligent virtual assistants by starting with the core Robin.AI platform and then plugging in task agents to configure the kind of intelligent assistant they need. If a pizza delivery shop wants to create their own virtual agent, for example, they can assemble the traffic and messaging app to send alerts to drivers in case of traffic back-ups. That’s a simple example, but it illustrates the intended flexibility of the building block platform.

Another aspect of Robin.AI is that the platform tracks user data, including user preferences, so that it can learn to infer user intent. For example, an intelligent assistant based on Robin.AI can make note of the fact that you like music by the rock band Boston and that you are currently in the car listening to your iPod. So if you say “I feel like Boston,” the assistant has enough information to extrapolate that you want to hear a Boston tune (and not get driving directions to Boston, for example).

It remains to be seen if Yahoo really is building an intelligent assistant app based on the Robin.AI platform. It wouldn’t be an unexpected move, since the prevalence and importance of virtual agent technologies is undoubtedly on the upswing.

MindMeld Lets Your iPad Listen As You Talk

MindMeldExpect Labs issued a press release last week about the launch of their MindMeld app for the iPad. The press release describes Mindmeld as an anticipatory intelligenct assistant app. It can listen to you as you talk to it or to one or more friends on Facebook, and then go out and search for related content either within Facebook or on the web.

How does the app work? I watched the short app demonstration of MindMeld on the Expect Labs webpage. It appears that you have to be a Facebook user to take advantage of MindMeld, as the only way to log in is through your Facebook account. It seems that the idea is for you to join in a conversation with friends–either one that’s currently underway or one that you initiate. MindMeld then listens to what you are saying and it starts displaying what it believes to be relevant and helpful content on the MindMeld screen.

If you and your friends are planning a trip to the BCS Championship game to watch Auburn battle Florida State, for example, I suppose that MindMeld would show you Facebook feeds from anyone trying to sell tickets to the game or maybe even offering a place to stay in Pasadena. MindMeld would probably also show things like airline tickets, hotel specials, or driving directions.

I’m a bit confused as to how the conversations work. Are you actually carrying on a live conversation where you can hear all your friends talking, like in a Google circles call? Or are you and your friends each just talking to MindMeld separately and the app listens to each person and pulls out the things it hears that seem relevant or interesting? I’m guessing that it’s the former, and that you can actually hear your friends speaking.

It’ll be interesting to see how MindMeld functions in reality and whether people find it helpful to be bombarded with content that an app thinks you might like. If you say something like “I drank way too much last night,” will it show hangover remedies? Or  will it show you the news feeds of all your other friends that recently typed or said “I drank way too much last night?” Right now when you google that same phrase, the results are mostly in the latter category. Misery loves company, so that might be helpful. But it could just as well be annoying.

I’m confident that there are valid use cases for an app like MindMeld, though. Speech-based search will definitely be a part of our normal lives in the future. The question about how often and under what circumstances we want apps listening into our conversations remains open.

Falling in Love with Your Virtual Agent (Maybe)

Intelligent Virtual AgentThe New Yorker ran an article last month called “Can Humans Fall in Love with Bots?” It’s a rather sensational title intended, I suppose, to grab attention (and yes, I latched onto the hook for this post!). The New Yorker piece, though, covers the topic of virtual agents, chatter bots, and the universe of conversational virtual assistants broadly, not just from an attachment standpoint.

The article was written by Betsy Morals, a frequent contributor to the New Yorker on technology topics. The anchor for her post is the Spike Jonze movie “Her,” which I wrote about previously. Morals viewed a pre-release version of the film and she uses this scifi fantasy story as a jumping off point to explore the current state of virtual agent technology as it exists in the real world today.

Morals references discussions she’s had with Fred Brown, CEO of virtual agent technology company Next IT. Next IT created as virtual assistant called Jenn for Alaska Airlines. It turns out that Jenn attracts users who engage her in conversations that go beyond just asking about airline tickets and flight status. Brown told Morals that data shows people often converse with Jenn for extended periods, especially late at night. They probe her with questions about her likes and dislikes. Some users even seem to be trying to flirt with her, despite her obvious virtual nature.

Morals also spoke with Nova Spivack, who sits on the board of Next IT and who was involved in the research work that eventually led to the creation of Siri. Spivack seems to express the opinion that today’s virtual agents are realistic and believable enough that people might be lured into developing an emotional attachment to them. But Morals is skeptical. Her own interactions with virtual agents haven’t been as impressive. She references Sgt. Star, used by the U.S. Army to answer questions from potential recruits. While the Sergeant can answer most of her questions, she suggests that he clearly doesn’t display a full understanding of the intent of a person’s questions, especially when you try to drill down deeper into a conversation or ask questions about the Sergeant’s own emotional response to certain aspects of Army life. (As a side note, I wrote about the technology behind Sgt. Star in an earlier post).

According to Morals, it turns out that Spike Jonze was inspired to create the film “Her” based on conversations he had with the AIML-based ALICE bot years ago. ALICE’s ability to understand context and adjust her conversation to the desires and habits of dialog partners is extremely limited. But Moral’s discussions with Nuance’s Gary Clayton lead her to be more optimistic about the future capabilities of virtual agents. Clayton believes that as virtual agents have more access to your personal data, they will be better equipped to proactively assist you. He gives the example of a virtual assistant that knows you’re driving on the highway and recognizes that you’re suddenly exceeding the speed limit and asks you if you’re ok. (It may also know that you didn’t sleep well the night before or that you’ve just left from a stressful meeting).

As our virtual agents become more capable and indispensable, it’s not hard to imagine a day when we’ll develop emotional attachments to them. Morals touches on the question: will our virtual agents be able to love us back? But that’s fodder for an even more esoteric discussion. Read Morals’ article for yourself. Just be warned that if you’re planning on going to see “Her,” the article gives away the ending.

Aldebaran’s NAO Robot Wants to Have a Talk With Us

NAOMashable recently published a short article on Paris-based Aldebaran’s humanoid robot. The robot, called NAO (apparently pronounced “Now”) already has the ability to mimic human movements. It comes with a programmable architecture and a CPU running a Linux kernel.

Now Aldebaran has ambitions to transform their android into a fully functional conversational robot. The company recently inked a deal with Nuance to integrate voice recognition and natural language processing technology into NAO’s architecture.

According to the Mashable article, NAO will have access to data in the cloud that will enable it to build its vocabulary and eventually carry on conversations. When combined with its face and object recognition, NAO might be useful in many different environments. Aldebaran considers the android to have a place in education and research. If it develops its dialog skills successfully, it might even be a suitable companion and assistant for people of advanced age. Aldebaran is currently working on software modules that enable NAO to interact effectively with children suffering from autism related disorders.

It certainly remains to be seen how successful NAO will be at carrying on a meaningful conversation. Building a vocabulary doesn’t translate into having the ability to hold a true conversation, where parties listen to one another and ask questions that layer upon earlier elements of the dialog. Having a real conversation is different from just answering questions. Will we be able to talk with NAO, or will NAO just talk to us? It’ll be interesting to watch the robot’s progress and find out.