Tell us about your role and the team/technology you handle at RealNetworks.

I am Reza Rassool, CTO of RealNetworks. I help guide our technology strategy to spur growth. Since joining RealNetworks, three years ago, I have encouraged an emphasis in AI in all of our product lines. Three in particular:

• In the RealTimes application we developed computer vision features set to better catalog your mobile camera roll. That development spawned our SAFR business which now claims “the world’s best facial recognition for live video.”
• The Kontxt messaging business now sports an AI layer, solving problems like spam detector, anti-smishing, and message classification to allow carriers to optimize delivery.
• The legacy RealMedia business is also enriched by AI to improve Video Compression, enhance the Consumer Experience, and drive increased Monetization for OTT operators.

How big is your Data Science and AI team?

AI and Data Science are now deeply engrained in the product roadmaps for all our businesses. So, I’d say that all Product Development staff have added Machine Learning to their tool belts. We started with a core team of PhDs but soon disseminated the tools among our global engineering staff through our innovating ThinkDay and Hackathon programs. I ensure that demonstrated AI competence is part of the job description of all Senior Engineering hires.

We hear a lot about the exciting opportunities in Data Science. No one talks about the challenges and risks. What Data Science-related challenges do you deal with at RealNetworks?

Let me be a bit contrary and say that nearly every news article warns of the challenges and risks of AI – most notably the fears of it accelerating a dystopian future. Among the risks and challenges I see:

• Privacy – At Real, we have been very deliberate about addressing the criticisms of facial recognition with respect to its bias and its potential for invasion of privacy. We developed a set of Guiding Principles that steer us towards the ethical use of facial recognition. Our user experiences start with an opt-in process. We provide an SDK allowing businesses to add facial recognition to their loyalty apps. Consumers can opt-in by enrolling their face for automatic recognition. While SAFR for Security enhances and automates existing video surveillance systems we also promote use cases that enhance the consumer experience.
• Bias – Additionally, we strive to reduce the bias in our facial recognition. Bias is defined as the variation in accuracy across, age, gender, and skin tone. Amazon Rekognition has been singled out in the press for bias – exhibiting high accuracy for white males and lower accuracy for darker-skinned females. NIST has consistently scored the RealNetworks facial recognition algorithms as having among the lowest bias. We achieve this through curating a diverse training set sourced from around the world and sound science. We specifically do not classify individuals by race or ethnicity. These are pseudo-scientific classifications of humanity. Many competing FR systems that have chosen to tag people by race inevitably require an army of human labelers to look at faces and make a determination. In doing so, they are enshrining a very human bias into their Machine Learning.
• Perception – The public perception of AI and particularly facial recognition varies around the world and is changing rapidly. We have already accepted devices in our homes that eavesdrop on our conversations in order to allow us to control our devices and interact with services by voice. While facial recognition is viewed with caution in the West, it is embraced elsewhere. Shoppers smile to pay at the checkout in stores in Beijing.
• Regulation – The EU provided clear guidelines for the acceptable use of biometrics through GDPR. We welcome similar clarity in the US. Currently, regulations vary state by state and now even city by city. However, while we wait for legislation, I urge each company building AI-based products and services to develop their own set of guiding principles.
• Explainable AI – Machine Learning creates models that can make predictions such as what movies you’d like, how to price airline tickets, how to route city traffic. These models are opaque ‘black boxes’. We don’t understand the reasons for the decision made by an AI model. The recent DARPA initiative to develop a toolset that’ll deliver models that are more transparent, hopes to make the internal workings of deep neural networks more open to human scrutiny.
• Incremental Training – While current Machine Learning is a rudimentary attempt at mimicking the human brain, the existing state of the art is ‘catastrophically forgetful’. In order to add to the training of a model you need the entire historical training set. The industry in seeking a method of incremental training that allows new training data to be added to an existing model to result in a new smarter model. This is a better imitation of human learning.

How can Marketing and Sales teams better leverage Blockchain and Big Data analytics to grow their businesses?

I’m not a Blockchain expert. We have one SAFR Developer partner building a system that logs and anonymizes smart city computer vision events. It is possible that an indelible ledger may be the appropriate repository for IoT events from sensors including computer vision events from cameras. This ledger would be trustworthy and would also preserve the anonymity of the source.

Which industries and markets have been the fastest and most agile to adopt your technologies?

After an early foray into Retail Analytics, we found significant traction in security. I anticipate that 5G IoT and Smart City applications will deploy computer vision. Additionally, Secure Access and hospitality applications continue to promise to eliminate the ID badge and other physical credentials or tickets and speed users through authentication check points.

What makes RealNetworks different from other high-tech solution providers?

RealNetworks has a history of tackling the hardest technical problems. With our RealMedia business, we brought Streaming Media to the world across the Internet when it was hard – when your connection to the Internet was through a dial-up modem. In our Kontxt business we route and process the fire hose of text messages between mobile carriers and have maintained an uptime of over 12 years. In our SAFR business we’ve taken on the most difficult category of live video facial recognition for in-the-wild subjects. The DNA of innovation persists in RealNetworks.

Salesforce recently announced acquiring Tableau. Google is doing that with Looker. How do you see Data Science and Data Visualization becoming the core of every Marketing and Sales Technology offerings?

Congratulations to my friends at Tableau. Yes, AI is already a ubiquitous tool, wielded by many businesses. I anticipate computer vision permeating many aspects of our lives beyond security from driverless cars to home automation. Facial recognition will progressively replace authentication by physical credentials. I imagine walking into a hotel lobby, recognized by a smart camera, receiving a text message that I was automatically checked-in to my reservation and providing me with the room number and code to open the door. Queuing to check-in will become passé.

Tell us more about your roadmap for Data Science and AI in RealNetworks.

We continue to improve the accuracy of our facial recognition but the gap between the leaders in this space is narrowing. My specific preoccupation is on driving SAFR deeper into devices. The speed and compactness of our model make it more embeddable in cameras. The Kontxt platform will continue to use Machine Learning and Natural Language Processing to derive deeper insights from messages. In the RealMedia tool set, we are also training models to improve the compression of video and enhance its streaming.

Do you see Data Science becoming the core of all MarTech and SalesTech companies? How can Product and Innovation teams work with the Data Science team to make these products better suited to market demands?

All businesses must glean insights from historical data. Data Science and AI provide tools for us to automate the analysis to deliver those insights. Historian George Santayana said that ‘those who do not learn from their history are doomed to repeat it.’ In the business domain, you might not even be able to repeat your past success if you don’t learn and understand the cause of your success.

What is your advice to Data Science Engineers, AIOps Analysts and, MarTech Product Managers?

Beware. AI is not a silver bullet or magic mirror who can answer any questions. Deep neural networks are at best curve fitting to past data – albeit to curves in hundreds of dimensions. Machine Learning, in its current form, can only tell you what your data already knows. Moreover, it cannot provide you insights for eventualities not represented in the ground truth of your training data. Don’t abdicate leadership to the machine!