Chatbots – the new apps

by Andy Mayer
Posted on 20 November 2017
read time:

I've recently been following a series of online courses from MIT and Stanford on Machine Learning – learning all about chatbots, artificial intelligence and conversational interfaces. This led me to design and develop an AI-powered chatbot for mental health support in Sheffield. Yoomee can now help you research, design and prototype new ways of using this futuristic technology for social good.

What are chatbots?

A chatbot is a piece of software designed to simulate realistic human conversations. Chatbots can run on instant messaging platforms, such as Facebook Messenger, can have helpful and involved conversations with people and act as virtual personal assistants.

Gartner predicts that by 2019, 20% of all user interactions on phones will take place via virtual personal assistants with many mobile apps fading as a result. According to Gartner in the same timeframe, virtual personal assistants powered by artificial intelligence “will have changed the way users interact with devices and become universally accepted as part of everyday life."

The problem with chatbots

In the past chatbots have had a reputation for being really dumb when trying to have conversations with humans. Such chatbots were programmed with loads of rules, like a massive flowchart, in an attemtp to emulate intelligent responses. This approach was very labour intensive to build, and the problem was, if the human didn't respond exactly as expected, then the computer couldn't understand anything that was being said.

However, with recent advances in artificial intelligence technology, we can now use very powerful machine learning algorithms to accurately understand and interpret the intent behind conversations. And it's pretty impressive!

Rather than relying on a set of rules machine learning works by using "training data" – a huge pool of example inputs mapped onto the expected correct output. Using statistics and mathematical algorithms, the computer can see how similar the input is to the example, and then make an educated guess at the meaning – in a similar way to the way the human brain works.

This opens up a myriad of opportunities for building conversational interfaces where a human can talk naturally with a machine – and the computer can understand, and respond intelligently in real-time, like a human personal assistant.

Introducing Pebble

Pebble is a new proof-of-concept chatbot we've developed at Yoomee which uses powerful Natural Language Understanding (NLU) and machine learning to understand the meaning of what people are asking for.

Its objective is to signpost people in Sheffield to where they can get help with mental health problems.It currently understands 33 different types of mental health problems and eight ways of accessing support – which means it can cope with 264 different types of conversation. The more people that use it then the better it will get at understanding.

If you'd like to help us train Pebble about what help is available in Sheffield, then please do get in touch.

We're also working on making Pebble available through Facebook Messenger and the Google Home Speakers so you can give voice commands and interact with services through Google's intelligent personal assistant, called Google Assistant.

Chatbots are the new apps

2017 has been the year of the chatbot, and the technology seems ready for prime-time due to many factors coming together at the same time:

  • Thirst for instant. Research indicates that people are willing to wait no more than two seconds for a web page to load. So, getting an instant reply from a chatbot is a very attractive alternative to using a website.
  • People use instant messaging more than anything else. A 2014 report from Forrester found that time spent on messaging apps exceeds time spent on social media – and this figure has been growing ever since.
  • Artificial intelligence (AI) as a service. Big technology companies like Facebook, Google, Amazon and IBM all make their AI technology available for others to use. There is no need to build your own technical infrastructure.

Chatbots are multi-platform

Don't forget that many different messaging platforms can be used by chatbots. The main ones are Facebook Messenger, Twitter and WhatsApp and the overall user base is growing massively. But there are also traditional platforms such as SMS and email. It's also possible to deliver these conversational experiences using voice-controlled smart speakers, such as Google Home, Alexa and Apple's Siri.

In 2016 there were about 2.5 billion people registered on at least one messaging app. By 2018, this number is predicted to increase to 3.6 billion users.

How we built Pebble

I started off by using Google's DialogFlow service to build the Pebble chatbot in Node.js. Then I migrated away from DialogFlow to IBM's Watson Assistant (formerly Watson Conversation). In my opinion, Watson is way more powerful than Google's offering as IBM has been in this game for much longer than Google. IBM Watson wowed the tech industry with its 2011 win against two of Jeopardy's greatest champions. In my tests Watson was much better able to understand the meaning of text better than Google's DialogFlow.

2019 Update

I've since migrated away from IBM Watson to the industry-leading open source AI technology from Rasa which means we can host and run the chatbot ourselves without relying on a third-party cloud provider of AI. Being open source, Rasa also allows me to customise the backend code and tweak the user experience.

Don't be left behind

Chatbot adoption has already taken off in the US where more than half of those between the ages of 18 and 55 have used them, according to research by Business Insider. If you'd like to explore how your service users could benefit from using chatbots, then please do get in touch. We can help you research, design and prototype new ways of using this amazing technology.

To find out more about how we are using AI to help vulnerable people please follow our project on Twitter.

Posted on 20 November 2017 - By Andy Mayer
Share this post
blog comments powered by Disqus