An Obituary & Autopsy Report for Facebook’s Assistant M: 3 Take-Aways for Chatbot Creators
Earlier this month, Wired Magazine reported the imminent death of Facebook’s Assistant M (“M”), a text-based virtual assistant chatbot with a very Fox & Friends/Morning Joe-type headline touting the death of all chatbots. Scott Thurm, business editor of Wired, tweeted wistfully, “It wasn’t long ago that everyone had high hopes for chatbots. Facebook’s decision to kill M, it’s entry in the field, suggest those hopes were misplaced.”
And there were other tweets from users like @LauraKlein, asking “Who could have possibly predicted such as thing?” (implying that possibly she had predicted such a thing) or @dendisuhubdy who tweeted, “YESSSS!!!” (Rude, Dendi. Glad I’m not your aged family member. #schadenfreude)
With Assistant M’s passing, I think it only fair to look past headlines, and take time to properly analyze and remember M’s place in the world. Hence, I have created an obituary, (I hope it wins a Grimmy Award), an autopsy report, and 3 key takeaways from Assistant M for all chatbots creators.
OBITUARY: REMEMBERING OUR BELOVED ASSISTANT M
FACEBOOK’s ASSISTANT M, known affectionately as "M" in the tech community, passed away Friday, January 19, 2018, of totally-determined causes. She was born August 8th, 2015, in an office with much white space in San Francisco, California. M grew up with her developers, trainers, product managers, and a litter of bot cousins—though it was noted that she was the darling of the family, doted on by tech press and David A. Marcus, Vice President of Messaging Products. M was most known for being thoughtful; never one to forget a birthday or recall an article you liked—she’d save it for you, she was just the kind of gal!
From the time she was young, M was resourceful, so much so that if you wanted a hamburger, she would find a way to get it for you. Plane tickets? No problem. Oftentimes she would enlist her trainers to do this. M became so good at her hobby, that she decided to make it her career: making 10,000 friends/clients in the process. With so many friends, it became difficult to do what she knew best, even with the most excellent trainers. This lead to unfortunate case of exhaustion (of resources). We don’t remember M for what she couldn’t do. We remember M for what she tried to do. Some people have small dreams, and their lives are simple and easy, and some people have big dreams—which can never quite be realized in their lifetime. Though one should ask, what kind of life is best.
She is survived by her developers and trainers, who will miss her dearly. She will be survived by her grandchildren, who undoubtedly, will carry her vision to fruitfulness. A service will be at 12 p.m. Friday, January 26th, at Instabot offices, New York, New York, for coffee, mimosas and the singing of My Way by Frank Sinatra, because although she “bit off more than she could chew”, we are sure she had few regrets (RSVP and Join Us). Although M would have appreciated the beauty of flowers, she asks that those that are so inclined instead make a donation to your favorite STEM or tech charity (Mouse.org, Girlswhocode.com, to name a few) so that her work may be carried forward.
LEARNINGS + TAKEAWAYS:
Assistant M is not an example of why chatbots don’t work; it is an example of how using chatbots incorrectly, within the context of currently-available technology, leads to failure. Conversely, this also leads us to create a set of best practices for all chatbots creators to take forward:
Key Takeaway 1: Chatbot solutions should be laser-focused on a limited set of core-needs.
Chatbots are most successful when they develop a core set of competencies and build onto those skill-sets over time, as with Amazon Alexa. Due to this technology being so new, most chatbots creators cannot anticipate how chatbots will be interacted with by users. Therefore, starting with a large breadth of skills will lead to expectations which cannot be met.
Key Takeaway 2: The first thing chatbots should do is quickly introduce their core competencies.
Not only should chatbots start within a specific scope, they should always firstly tell you how they can help you and what they can do best. By managing expectations for users, you can prevent boundary testing by users. For text-based chatbots, this means conversations should start with a prompt, “What can I help you with? Option A, Option B, etc.” Voice-activated chatbots should provide verbal clues.
Key Takeaway 3: Even the best natural language processing (NLP) system has limitations.
Facebook, with all its resources failed to make an NLP system that could truly comprehend the needs of Assistant M’s users. NLP is still in its infancy, and the amount of training required is still very resource-heavy. Hence, it’s best to use NLP within scope and augment its capabilities with structured conversations. Also, creating appropriate hand-offs between chatbots and humans is critical in best leveraging chatbot technology.
I see Facebook’s Assistant M as an important part of the story of chatbots. Facebook was brave enough to push the limitations of chatbot technology, showing us what limitations exist, the traps to avoid, and where to go next. Assistant M is a parallel to Ham the Astrochimp, the first chimpanzee launched into space on January 31, 1961. Sure, Ham never made it to the moon, but he it was his flight that directly resulted in Alan Shepard’s later accomplishments. We salute you, Assistant M. Much like Ham, it is by seeing your path that ours has been made clearer, and for this, we are grateful.