2016 was hyped up to be the year of Bots. And indeed, despite the lack of public awareness around the topic – most people only read about bots in trendy blog posts – 2016 was, in fact, a very prolific year for this emerging trend. Much was accomplished on the developer-side: the building blocks of the bots ecosystem were laid out and the whole infrastructure to build and deploy them greatly matured. Bots are ready, or soon-to-be, for the mass market.
Below, we’ll explore how bot-building became more accessible to anyone who wants to build one. We’ll present diverse examples, from libraries to actual functional bots, before discussing another major topic: how to make them available to the biggest population of “chatters” there is – the world’s 5.1 billion mobile phone owners.
The building blocks of a conversation
From the outside, bots can seem easy to develop. And indeed, creating sophisticated bots became much easier in recent years thanks to the increased availability of numerous underlying technologies required to create them.
Accessible and easy-to-implement libraries
Ready-to-use bot-building libraries appeared and keep getting better. Here is an overview of three main technologies frequently used to make truly intelligent chatbots:
- First, Natural Language Processing (NLP), the ability to understand and extract data from everyday language made tremendous progress over the last couple years. Many common programming languages have had low or high-level, accessible and easy-to-implement NLP libraries, and new ones keep appearing — Python (NLTK, spaCy, Pattern), JavaScript (Talisman, Compromise, NaturalNode), Java/Scala (Apache OpenNLP, ScalaNLP).
- Secondly, it is now easier to implement. Machine Learning and Deep Learning, which is used to train bots by allowing them to continuously learn from the inputs of their users. Specific libraries are available in the main languages: Python (ChatterBot), Java (DeepLearning4J) or Javascript (Craft.AI). Learning techniques allow for the creation of bots that become more efficient with each user interaction, progressively adapting their behavior to the users’ habits. We’ll see an example of this with George the Dentalist below. Now, there are even high-level libraries with both NLP and Machine Learning capabilities, such as Talisman (JavaScript) or ChatterBot presented above.
- Thirdly, image and video recognitionare sometimes required to make the bots’ interactions natural and easy; Recast.AI explained in a blog post why image recognition was the simplest way to design a seamless conversation flow for the home improvement bot they were building. Once again, the main programming languages now have solid libraries for computer vision, such as OpenCV (Python and now JavaScript), Google’s TensorFlow (Python and C++) or IBM Watson visual recognition (JavaScript, C).
All-in-one bot-building platform
At the same time that libraries full of bot-building technologies became more available, another trend emerged: all-in-one bot building platforms allowing non-developers to start building bots, too.
Recast.AI perfectly demonstrates this trend: the platform facilitates the creation and deployment of bots over the most common channels. Recast.AI takes care of everything from building the conversation flow, to training, bot deployment, and analytics.
Recast.AI’s purpose is to help users easily create messaging interfaces. It’s a collaborative platform, so users can import bots or bot parts from other users’ bots, allowing them to save time and energy by assembling their own creation, piecemeal. Powered by a strong language technology, Recast.AI’s bots offers keyword detection available through an API so bot-creators can focus on crafting meaningful and fluid interactions. Moreover, because Recast.AI is community-based, their services improve with each user interaction made with bots using the platform.
From natural language to data
In terms of actually breaking down language so that bots can process it, they use an “entity” system. This system lets bots extract meaning out of the conversation bots have with users. Using entities allows bots to detect keywords in a natural, human expression. It allows for the automatic detection of basic entities, like datetime, location, and names. These are the 30+ so-called “gold entities”. Bot-builders can go further by creating custom entities using an organization process, the”gazette”, to help them be recognized. Gazette helps specify the perimeter of user-created entities: a gazette is a list of words belonging to an entity associated with a strictness parameter.
For instance, a coffee waiter bot would need a coffee entity. To define this entity, a gazette must be used to help him recognize all the different coffee types:
- Espresso
- Cappuccino
- Americano
- Caffe Latte
- Mochaccino
- Caramel Macchiato
Recast.AI goes even further than helping in the creation and deployment of bots. It can take care of all bot builders’ needs. They offer services to build and train bots described above, as well as analytics and hosting services. Indeed they provide bot-builders with numerous metrics to help them make the most of their users’ exchange. They can see the conversation streams of their users to understand their behavior and adapt their bots quickly
Beyond chatbots in messaging
The conversation doesn’t start and stop with bots. As channels and platforms multiply and diversify, it has become harder and hard to offer a service available on all main channels to reach all customers. Issam Hakimi, developer of Broid.AI, efficiently cornered the issue with his phrase “the Babel Tower of the 21st century”:
In a world where the connectivity grows exponentially, communication tends to become more complex and specialized to specific usages. The Babel Tower of the 21st century is arising.
To destroy this new Babel Tower before it reaches the cloud, he created Broid.AI, an open source conceived as an attempt to unify communications, enrich them, and gather a network of smart systems. Broid.AI’s goal is to make it easier for developers to deploy their services (be it bots, or something else) on multiple channels. Instead of having to work their way around numerous documentations that can be very diverse, they only have to work with one to implement all the channels they need. Deploying a chat bot or a messaging app over dozens of channels has never been easier.
So what can bots already do?
Recast.AI created bots to showcase the potential of their platforms. Among them is the Pokébot, a bot available on Kik and Messenger to answer all questions about Pokémon asked in natural language.
George, the Dentalist is a bot able to process purchase orders received from dentists and sent by SMS, to learn from the orders and ultimately proactively repeat the order the dentist would usually make. It’s powered by Recast.ai NLP capabilities and craft.ai for the learning part. Craft.ai’s team wrote about its genesis. It shows how smart and effective bots are not created per se, but emerge as a solution to solve a real-world situation.
Various platforms opened to bots over the last couple years: WeChat, Facebook Messenger, Amazon Echo, Slack, Google Allo, Snapchat, Telegram… Despite their large user base, everyday user interactions with bots are still far from mainstream. Indeed, most bots are still designed to face hypothetical situations instead of built from the ground-up as an answer an issue occurring in the real-world.
What’s next for chatbots?
Here At CALLR, we are convinced that SMS will help bridge the gap between bots and mainstream communication channels. Indeed, SMS has many advantages over OTT apps (WhatsApp, Messenger), including convenience. But more than anything, the game-changing SMS advantage over OTT is its scope. It reaches more that 5 billion people worldwide. SMS work even on the simplest phones, which gives them the potential to reach those who might actually need the services bots and AI can offer. Ultimately, SMS is the only channel which permits smart services on the most basic phones.
Because of its efficiency and availability, SMS is becoming the preferred communication channel for brands. Moreover, SMS is cross-channel and allow brands to exploit their underused database. For all the above reasons, we are thrilled of both Broid.ai and Recast.AI CALLR’s integration; both are, in their own way, pushing for a wider availability of bots that are smarter than ever.
The post 2017, the year chatbots started texting appeared first on CALLR Blog.
