We recently sat down with George Sazandrishvili, CEO of AlgorX, Inc. to get his thoughts on what he’s seeing in the field of conversational tech and for a sneak preview of his talk “Rise of Deep-domain Conversational AI” during the Conversational Computing Conference Oct 3-4, 2017. Learn more about this conference at: http://www.convoconf.com/.
George, could you please introduce yourself and tell us a little about your background and your role at AlgorX, Inc.
I wrote my first "toy" program at the age of ten on a PDP-11 compatible computer in BASIC. Just to give an idea how old it is, PDP-11s were minicomputers at the size of a modern refrigerator. I started serious programming on the first generation of 32-bit processors from Intel and even though I come from a business administration background, programming has been my lifetime passion. I discovered Artificial Intelligence in early 1990s when I had a chance to interact with ELIZA, an early conversational software agent created by Joseph Weizenbaum at the MIT AI Laboratory in 1960s. I am not sure what amazed me more, the ingenuity built into ELIZA or the fact that conversational technology had not moved ahead much in more than 30 years.
AI takes a lot of resources: time, money and computational power. AI research may be prohibitive for most people unless they happen to be in academic circles or in large corporations with a focus on AI. However, I have been managing to dedicate resources to AI endeavors through my other businesses for more than ten years. As we approach the third decade of the 21st century, AI seems more viable from a commercial perspective. Hence, we see a number of successful AI start-ups. In 2012 I started working on the software that would later become the core of our AI technology.
As a founder and CEO of AlgorX, Inc I have to wear many shoes. Thanks to modern technologies many tasks can be delegated and this allows achieving goals even with lean budgets. The world has become increasingly freelance-oriented (as of now 55+ million freelancers make up more than 35% of the US workforce) and this also makes it easier to outsource non-core functions. As AlgorX grows I plan to dedicate even more time to AI research.
Your session at the CCC17 event is titled "Rise of Deep-domain Conversational AI". Without giving away the details, tell us a bit about what we can expect from your presentation.
Understanding and processing human language lies at the heart of conversational technology. Current generation of conversational agents like Google Home, Amazon's Alexa or virtual assistants like Siri, Cortana and Google Assistant are very simple. While they are able to answer discrete questions and do certain commands, they are not yet able to maintain meaningful and deep conversations with us. We are much closer to building successful self-driving cars than building truly intelligent conversational agents. I will shed light on this during the session.
It is anticipated that once self-driving cars mature, they will make hundreds of thousands, probably even millions of drivers jobless. The same can be said about conversational technology. Once it reaches levels that are enough to replace humans, we will see serious effects on the jobs market. While a self driving car can replace only a driver (but not an office manager for example) conversational technology will have even bigger impact on the job market. Simply, there are more people employed for being able to talk than to drive. Even in self-driving cars conversational technologies play important roles. Sounds scary but we have to give a thought to this.
I will briefly review the major “building blocks” of conversational technology, key trends that are happening now (supervised and unsupervised deep-learning) and those that will shape the field in the coming years. The culmination of the session will be the question about how close we are to building the real "strong" conversational AI.
Where are you seeing Conversational technology being applied now, and where are you seeing it applied that is unexpected?
The rise of conversational technology is tightly coupled with proliferation of mobile devices. In essence, conversational technology is a form of human-machine interaction. While we may feel perfectly comfortable with a keyboard when interacting with a PC, the same cannot be said when we use a mobile device. The so-called software keyboard on a mobile device is getting better and better but still for the vast majority of people typing on a tiny keyboard is an awkward experience. What about interacting with your car or domestic electronic devices? In these cases conversational technology is the only reasonable way to go. An ideal situation would be to have a specialized microchip that could be integrated into any kind of electronic device to “conversation-enable” it. When this happens we will see conversational technology used in such unexpected places as microwaves and refrigerators. However, another alternative is to have one centralized device (think Google Home or Amazon's Alexa) that controls all of our home devices. Right now it's a bit difficult to predict which path the manufacturers of domestic electronics will take. Nevertheless, one thing is for sure: the future of human-machine interaction is conversational.
What do you see as the most exciting innovation or development in Conversational tech?
Conversational technology is exciting but challenging at the same time. It's exciting because the idea of giving computers the ability to process human language is as old as the idea of computers themselves.
It's challenging because the problem is AI-complete (implying that the difficulty of this task is equivalent to making computers as intelligent as people). Though conversational technology is not about humancomputer interaction only. Conversational technology helps with human-human interaction as well.
Think about a machine that can do high-quality translation of human languages in real-time. This technology would finally help us break language barriers and enable people all over the world communicate freely with one another. There are several attempts (for example Skype's chat translation) but the state of the art is still far from being practically usable. Innovation is happening in practically every area of conversational technology. The latest technologies in speech synthesis use deep neural networks exclusively and achieve remarkable results. The same can be said about speech recognition. With the help of deep neural nets the latest generation of speech recognition agents recognize our speech practically flawlessly. But something is definitely missing. Take your smartphone, say OK Google and what happens then? Unless you give a specific command, in the
great majority of cases the conversations end in Google search simply because the software is not able to maintain a deep and meaningful conversation with you. The next generation of software is going to change this. How soon this happens though depends on how quickly we will be able to create the real, strong-AI.