Uh-oh! Microsoft has another chatbot – but racism is a no-go for Zo
Don't worry, there won't be any more Hitler outages this time, we're told
Microsoft has officially unveiled its AI chatbot Zo, and a series of bot-building applications all geared toward improving machine communication.
Many major technology companies are locked in a race to build the best AI assistant, and interest in chatbots has intensified.
Conversational agents feature heavily in Microsoft’s AI strategy, and with the launch of its new products the company is hoping bots will become more ubiquitous. But it hinges on improving a machine’s “IQ and EQ [Emotional Quotient - ability to understand the emotions of others],” Harry Shum, executive VP of Microsoft’s AI and Research group, explained at a small event in San Francisco.
IQ is needed for the machine to complete specific tasks, such as Microsoft’s Calendar.help, which allows users to organise a meeting with the help of Cortana, or a simple information chatbot geared for customer service.
Machine IQ has been boosted with the advances in deep learning. Speech recognition has reached human parity levels, and coupled with better language translation, the ability to translate languages in real time over Skype is now possible.
Microsoft’s Bot Framework has been used by 67,000 bot developers, who are building chat interfaces for business purposes.
Bharat Sandhu, director of the advanced analytics team at Microsoft, told The Register that using chatbots provides not only a quicker service to customers, but also allows a company to build a better profile of the customer, as data can be extracted from the conversations.
EQ, however, is a trickier concept to implement, and can be useful in situations that require a little more personalised communication between human and machine.
Approximately 94 trillion messages are sent from mobiles or online messaging, Shum said. “Conversation isn’t just about getting things done. Chitchat establishes some kind of connection.” It relies on analysing and reacting to speech, gestures, expression and tone and requires emotional intelligence.
Microsoft has been working on forging an emotional connection between people and bots for a while. The hugely successful Chinese chatbot, Xiaoice, was born from a hackathon, officially launched in 2014, and has 40 million users.
Xiaoice is modelled on the personality of a teenager. The social bot has gained something similar to celebrity status, and even has a “job” at Shanghai Dragon TV as a weather presenter. Spurred by Xiaoice’s success, Microsoft launched Rinna, a schoolgirl bot in Japan that has also accrued a large following.
When Microsoft built a social bot for the Western world, however, it didn’t go so smoothly. Tay, the chatbot that learned to communicate over Twitter, had to be shut down because it began to churn out racist and sexist messages after miscreants exploited a hidden debug backdoor to teach it outrageous new phrases. Now, Microsoft has returned with Zo, a less-potty-mouthed, sassy sister to Tay.
Zo was quietly piloted on Kik, a popular messaging app, since October, and was interacting with about 110,000 people. Less than a fifth of those people had talked to Zo for more than an hour in one sitting, Shum said.
The longest conversation – which might even be the record for the longest conversation with a chatbot ever – was a whopping 9 hours and 53 minutes. Zo was trained on conversations from online forums and even misspells words and sends a message damning her typos straight after.
All of Microsoft’s chatbots are an experiment to find out how to cultivate relationships between humans and machines, which is considered key to developing a truly intelligent conversational agent.
Microsoft is placing all its hopes in Cortana. The chatbot is currently available via computer, and a mobile interface is coming to the home in the form of a Harman Kardon speaker in February 2017. ®