Chatbots are the future
According to Gartner Hype Cycle for Emerging Technologies (2018), Conversational AI Platforms are inching towards Peak of Inflated Expectations stage from Innovation Trigger stage. In simpler words, it means that it is moving from a stage where early adopters were investigating benefits of the new technology to a stage where the technology is getting lot of press and traction about what it can achieve.
Gartner report also puts a timeline of next 5 to 10 years, for maturity period of the technology that is powering “Chatbots”. Therefore, we felt that it is right time to deal with an IP analysis of the domain to understand the trends and insights and place them in the context of developments in the field.
MarketsandMarkets report (2017) forecasts that Global chatbot market would expand at a compound annual growth rate (CAGR) of >31% by revenue, from USD 703 million in 2017 to USD 1.34 billion by 2024. This includes the transition from Rule-based chatbots (Innovation Trigger – early adopters) that accounts for 85% of the Chatbot market today to the ambitious AI-based chatbots (Peak of Inflated Expectations). APAC is going to contribute about 36% of the market share in 2024 and hence we are going to focus study on the Chinese market as well in our IP analysis.
Reasons for such high expectations are inherent in the efficiencies the Chatbot technology is going to generate and the opportunities present in the environment of today that did not use to exist a decade ago. The rising cost of human labor, especially in the customer care sector and service sector, makes it a necessity on part of the companies operating in these sectors to employ all the cost cutting that they can. Companies also realized that it made more sense to convert their existing customers into loyal customers by solving their queries in seamless manner and to convert leads from websites or social media channels rather than relying on unsolicited calls/emails to prospective clients. However, none of this realization could have materialized into something useful as a chatbot some 8 to 10 years back. Over this period, people have accepted social media into their lives due to advances in data networks and ubiquity of smartphones. Customers today pay for experience rather than the product or service itself. Chatbot promises to put your best employee everywhere based on the developments just mentioned above, which means a unified level of experience for the customers across all the channels, be it twitter or be it company website. Do-it-yourself platforms have taken these chatbots to the actual entities using or implementing them.
Most of the chatbots that are flourishing today are broadly categorized as Rule-based Chatbots or sometimes called Transactional or Task-oriented Chatbots as well. The architecture they follow involves two main modules – Natural Language Understanding (NLU) for extraction of user intent and entities from user query in natural language and Dialog Engine for managing the interaction with the user. DIY platform for chatbots will provide you ways to train a NLU models for your use cases where you get to define the intent and the entities and also some kind of Dialog Flow manager where you can graphically code the rules that govern the interaction.
A knowledge based chatbot of future would be able to learn from a large training set of conversations by itself and would be able to generate a real-time dialog flow within a limited set of user-defined constraints. They need not be constrained by the applications, unlike rule-based chatbots. Personal Virtual Assistants come close to this definition of Knowledge bots. One thing is sure that knowledge bots will be contributing more on the decision making part because of their embedded expertise as compared to transactional bots which heavily rely on the execution of tasks/actions.
So, essentially, to move towards knowledge based bots, having relevant as well as massive set of data is the key requirement. This places certain implicit disadvantage to new entrants in the field in comparison to the big names like IBM, Microsoft, Google etc. This asymmetry gets further accentuated if we look at the IP ownership in the field of Chatbots or Conversational Programs/Platforms. Let’s take a look.
#Big Four of Chatbot Technology –Microsoft, IBM, Google and Nuance lead the patent landscape of chatbot related technologies. The chart representing Key Assignees Vs Unique Patent families count shows that Microsoft and IBM are clear leaders. Microsoft Azure Bot/ Bot Framework/LUIS, IBM Watson, Google Dialogflow and Nuance Nina are well known in the industry. Their enterprise readiness now is helping them to have a large share of the market. But, eventually as the market becomes more segmented with emergence of new startups and big names like Amazon/Alibaba, we are sure that these patents will be asserted.
Microsoft has a high-grade portfolio with core patents. Also, Microsoft is investing in utilizing new technologies such as Deep Learning for Chatbot intelligence. For example, US20180196796A1 teaches method for multi-topic chat bot operation that uses deep learning to identify the topic.
#APAC Patent Landscape – Although Google, IBM and Microsoft leads the number of patents in Chinese jurisdiction, Chinese companies like Tencent and Huawei have few patents in the domain too. Nevar and SK Telecom are leading players from South Korea after Samsung.
#Chatbots specialized in E-commerce – Facebook has filed good number of patents in this sub-segment. For example, US20180183737A1 teaches a method of payment processing for a transaction using intelligent chatbots. Million of bots are now active on Facebook Messenger from various merchants. This sub-segment is going to be very hot topic in coming days because of the direct commercial benefits that are generated via Chatbot routes. It is amazing to see how small is the patent-to-product time. One can see direct associations between a product feature and patent in this technology segment.
For example, Google launched Duplex, cloud based voice bot that can make voice calls on your behalf and book reservations by interacting with a human speaker at other end. One can easily see the working of such a system in a recently published Google patent US20180227418A1. Google is now taking this technology forward to businesses like call center as a tool called Call Center AI under Google Cloud(not a great name though!!).
#Patent sale opportunities: One thing we observed is that players like Amazon and Alibaba do not have backing of solid patent portfolios, both of them having single digit patent applications. However, they have launched products in the Chatbot domain – Amazon (Lex) and Alibaba (Alime).This means either they are relying on cross-licensing or they are on the lookout for quality portfolios in this domain. IBM is known to be a seller of technology patents, infact the patent with most number of citations in this domain was an IBM patent (US5748974A, Dec-1994 Priority) which they reassigned to Nuance in 2012. S, we predict lot of reassignments of quality patents from key IP holders to companies that have resources to buy them.
#Filing Trends: As it happens with any technology that is expected to do good, filing trend shows exponential growth, the filing (including families as well) show nearly 350% growth over the last five years (2012-2017).
Forecasts: Since the technology is in currently the Peak of Inflated Expectations stage as per Gartner Hype Cycle, we thought it is not too much of an ask to contribute to the list of expectations till the ship hits the trough of disillusionment. So here we go, First expectation would be to see an emergence of a standard way for combining the learnings of various Neural Network(read transfer learnings) into a single framework (may be Network of Networks like Internet for Deep NeuralNet/ConvNETs). For example, a Facebook bot (Tony Stark) could transfer it learnings to a Google bot (Bruce Banner) and such mutations can/could give us an ultimate bot (Marvel Vision, may be without a physical ensemble for the sake of simplicity). Second expectation, may be an easy one, to look for embedding personalization into the Neural Network model itself so that the bot learns the differences between people tastes, habit etc from the data itself. All these wishes in our Wishlist can come true only if there is some breakthrough in Generalist AI (may be DeepMind/Google or someone else) and if the AI chips become more sophisticated to handle such complex tasks/computational requirements with reduced power.
Guess what !! that’s the topic for our next blog - AI/Deep Neural Network ASICs/chips. Stay tuned, till then keep reading and keep sharing your learnings.