This article was originally published by AI Frontier.
Facebook virtual assistant M is dead. That’s the state of chatbots
Translator | Liu Zhiyong
Editor | Vincent Chen
AI Front Line introduction: In August 2015, David Marcus, Facebook’s director of Messenger, announced the launch of M, a personal virtual assistant within Messenger that can perform simple tasks and find information on behalf of users. Be trained and supervised by people. Unlike other AI-based virtual assistants, it can represent the user to complete tasks such as shopping, gifting, and restaurant reservations. At the beginning, Facebook launched VIRTUAL assistant M with great ambition, aiming to build M into an O2O project that can complete the closed-loop of online to offline services and directly turn Facebook into an entry point of physical goods and e-commerce. However, Facebook has announced that virtual assistant M will be shut down on January 19, 2018, and members of the team will move to other departments. Let’s take a look at this article and see what’s going on with chatbots.”
It’s hard to think back now, but in early 2016, many in the tech industry believed that “chatbots” (text-based virtual assistants) would be the next big thing. Chat application company Kik has bet its future on the machine and “ChatVertising [1]”. Start-up Betaworks has launched an accelerator program called Botcamp[2]. And at its F8 conference in 2016, Facebook unveiled “smart bots” to developers as the best way to connect with 900 million Messenger users [3].
AI Front: ChatVertising is a feature that US chat app company Kik launched in July 2014 to let people talk directly to brands via bots, a format they thought could be the future of advertising, or at least a sign of a new era of brand interaction. The Wall Street Journal wrote that it was the model for the future of native advertising. Botcamp is Betaworks’ first startup accelerator program to promote chatbots.
Few could have predicted the steady growth of voice assistants like Amazon’s Alexa and Google’s Intelligent Assistant, and the decline of text-based chatbots into a joke. Betaworks says its accelerator was designed once and has moved on to other themes. Kik switched to blockchain technology [4]. Now Facebook says it will shut down M, its full-service virtual assistant, on January 19.
In some ways, it’s impressive that Facebook has kept M running for so long. Despite the hype, M in Facebook Messenger is being presented as an experiment. In the San Francisco area, the free service is available only to 10,000 employees, who use it to do things like reserve restaurants, change flights, give gifts and wait for customer service. For those who have access to this service, M is an amazing benefit; But for Facebook, it’s a cost center.
AI Front: Facebook virtual Assistant M was launched in August 2015 and ended on January 19, 2018, running for two and a half years.
That’s because most of the tasks that M does need to be done by people. Facebook aims to develop artificial intelligence technology that will automate almost all of M’s tasks. But despite Facebook’s engineering resources, M’s performance has been less than impressive: one source familiar with the project estimates that M has never been more than 30 percent automated. Last spring, M executives admitted [5] that the problems they were trying to solve were more difficult than initially realized.
For M’s executives, it was easy to get internal support and resources for the project in 2015, when people were feeling new about chatbots and their future was full of possibilities. But as it became increasingly clear that M would always require a lot of expensive human resources, the idea of expanding the service to a wider audience became less feasible.
The core problem with M is that There is no limit to what Facebook can ask M to do. Alexa has proven adept at dealing with a smaller range of questions, many of which are factual or core to Amazon’s strengths in shopping.
Another challenge: When M can do the job, the user pushes M to do more difficult tasks. A fully automated M would need to do far more than existing machine learning technologies can do. Today’s best algorithms are a long way from truly understanding all the nuances of natural language [6].
Facebook’s M did succeed in automating some of the contractor’s work. If you ask a robot to deliver flowers, it can automatically get suggestions from an online florist, asking only one person to choose the price offered to the user.
Facebook didn’t come away empty-handed. The omniscient assistants who use the service and role-play have generated valuable data that can be used by the company’s AI researchers. Using machine learning to make software better understand natural language and conversation is one of the team’s main interests [7].
In a statement, Facebook said: “We started this project to learn from users’ needs and expectations for virtual assistants, and we’ve learned a lot from it. We’re applying these useful insights to other AI projects on Facebook, and we’re very pleased with M’s performance in Messenger. That’s what this experiment has taught us.”
References:
[1] Advertising’s New Frontier: Talk to the Bot
www.wsj.com/articles/ad…
[2] Botcamp
betaworks.com/botcamp/
[3] Introducing Bots on Messenger
Developers.facebook.com/videos/f8-2…
[4] IS YOUR STARTUP STALLED? PIVOT TO BLOCKCHAIN
www.wired.com/story/is-yo…
[5] Facebook’s Perfect, Impossible Chatbot
www.technologyreview.com/s/604117/fa…
[6] AS ARTIFICIAL INTELLIGENCE ADVANCES, HERE ARE FIVE TOUGH PROJECTS FOR 2018
www.wired.com/story/as-ar…
[7] Facebook’s Head of AI Wants to Teach Chatbots Common Sense
www.wired.com/2016/06/fac…
For more content, you can follow AI Front, ID: AI-front, reply “AI”, “TF”, “big Data” to get AI Front series PDF mini-book and skill Map.