ALCMY recently hosted Birmingham Artificial Intelligence group (BrumAI) for a series of talks on how organisations are leveraging Artificial Intelligence (AI) to automate time-consuming tasks, giving them more freedom to focus on tasks that require human creativity.
The speakers were truly inspiring, each sharing their experiences on problems they were facing and how they managed to solve them with AI.
The talk that particularly caught our attention was delivered by Justina Petraitytė, who discussed how she developed an AI assistant called SHIBA (Slack Hosted Interface for Business Analytics) to automate data analytics at her workplace using an Open Source Software called Rasa NLU.
Rasa NLU is an open source NLP tool for intent classification and entity extraction. You can think of it as a set of high-level APIs for building your own language parser using existing Natural Language Processing (NLP) and Machine Learning(ML) libraries.
SHIBA is an AI data analyst chatbot that fetches, aggregates and visualises data on Slack. Prior to building the chatbot, Justina had observed the convoluted and inefficient ways of how her colleagues at different levels of the organisation shared data and information.
She embarked on this project to make data more accessible to the decision makers in her company. This meant the CEO of the company could log into Slack and have a conversation with SHIBA by asking simple questions and receive the data insights they requested in a visual format.
As you can imagine this was no easy task, especially when it comes to building conversational interfaces. The chatbot had to understand the different questions and nuances of how humans ask the same questions.
She used an Open Source Software called Rasa NLU (Natural Language Understanding) to leverage its intent classification and entity extraction features. In layman’s terms, intent classification is the relationship between what a user says and action to be taken, while entity extraction is the representation of useful information such as places, dates, people, durations, money and ordinals. In order to tie both features together, she used Rasa Core to predict what action SHIBA should take at specific points of the users’ conversation.
As with many Artificial Intelligent systems, the challenge is teaching the machine to be contextually aware of the domain specific language and information in order to clearly facilitate conversations. She talked about the different approaches she took to generate and enrich the various training datasets in order to make the chatbot both flexible and intelligent.
Key lessons about building conversational interfaces
Understand your problem space
With the rise of chatbots and voice assistants such as Alexa and Google home, it’s easy to get seduced into jumping into the chatbot craze without clearly understanding the problem you’re trying to solve. You also need to ask yourself critical technology-based questions such as, “Is a chatbot or an Alexa app actually the right solution to your problem?”
SHIBA is an excellent innovative solution to a problem Justina’s organisation was experiencing which leverages cutting-edge technology to eliminate inefficiencies and increase productivity.
Understand your user profiles
Conversational interfaces are fairly new in the technology scene and everyone is trying to grapple with different applications of the technology, especially when it comes to user experience.
It’s important to understand that the intelligent chatbots you will be building will be interfacing with real people with real problems, therefore keeping the intended users at the centre of the conversation and not the technology will be key to delivering business value.
SHIBA was purposely built to make data more accessible to the decision makers in the company. The success or failure of the chatbot lies in the understanding of context, different personality traits, motivations, goals and frustrations of the decision makers during interactions.
Focus on the User Experience
Elvia Vasconcelos defines Conversational User Interfaces succinctly as any User Interface that mimics chatting with a real human.
It is evident that this presents a significant challenge both now and in the future as societies and cultures evolve.
Elivia also points out that one key challenge that applies to both bot designers and engineers is getting conversations started between the user and the bot.
The bot interaction paradigm is quite different from human interaction as it relies on users asking questions. The current human-computer interaction model to date has always been pressing buttons to carry out commands rather than explaining to computers what they want them to do.
Justina pointed out that her biggest challenge while developing SHIBA was building a relatable chat interface, where a lot of business and domain-specific languages and acronyms are used, without putting constraints on how the users requested data.
Her solution to this involved both manual and automated ways of training the chatbot on the nuances of what was being asked and retroactively applying new rules on the bot to make it more intelligent and aware.
Augment your bot with UI Elements
Humans are visual beings with an attention span of a goldfish, that constantly need to be stimulated otherwise they’ll switch off and find something else interesting.
Conversational Interfaces provide a unique opportunity to keep users engaged as information presented to the user is under the user’s command and control. We can augment the responses returned with user interface elements such as buttons, pictures, funny gifs, animated emojis, graphs e.t.c., thereby enriching the user experience and making it more memorable.
SHIBA managed to achieve this by fetching, aggregating and presenting data in the form of visual charts on Slack.
As technology evolves so will our approach to work. Our workplaces and organisations will be completely transformed and redefined by Artificial Intelligent systems.
The way we think about work will be augmented by computational systems that will assist you in thinking and robotic systems that will help you create and make products, leaving workers to focus on their creativity and human intuition to solve more complex problems.
So the main question you should be asking yourself as a decision maker in your organisation is “How is my organisation leveraging Artificial Intelligence in my day to day activities to solve mundane and time-consuming tasks to thinking creatively on innovative solutions?”.