Photo Credit: shutterstock
One of the hottest social media marketing trends of the past 18 months has been the development of chatbots to converse with customers. For example, brands are now creating chatbots for messaging platforms like Facebook Messenger (why talk with a real-world customer service rep when you can chat with a bot?), as well as exploring unique ways to create 1-on-1 conversations with customers.
But there’s just one problem with this strategy: the bots might be too smart for their own good and start interacting with customers in ways you never imagined.
The Facebook AI experiment involving chatbots
Case in point: Facebook recently shut down an AI project in which the social media giant was training chatbots how to buy and sell goods. Surprisingly quickly, the chatbots began to converse in a language that the researchers didn’t recognize. And it wasn’t just that one chatbot created a language of its own — two completely independent chatbots could realize when it was time to transition from English to their own made-up language, and didn’t miss a beat. It was almost as if they decided on their own to use a new, secret language to converse that humans couldn’t follow.
Researchers first thought that this new language was some kind of “shorthand” that the bots created to help them speed up a transaction. And, since the researchers did not specify at the outset that the bots could only converse in English, the bots simply decided to fast-track things with their own improvements on English. From an AI perspective, this made some sense.
However, quite honestly, if you examine the chatbot transcripts, the change in language is somewhat eerie. The words are still in English, but the sentence structure is really messed up, pronouns are used strangely, and some articles (e.g. “the”) are repeated in ways that just don’t make sense.
Dystopian AI scenarios for chatbots
There are two ways to interpret this little Facebook chatbot AI experiment, of course. One is that artificial intelligence (AI) will eventually lead to a dystopian future scenario in which machines take over from humans. The fact that silly chatbots were able to come up with a strange new language is just the canary in the coal mine. Who wants chatbots to be talking behind our backs, plotting the demise of the human race?
That’s why many media outlets, eager for a viral story and lots of clicks, framed the article as Facebook AI researchers panicking and shutting down the project before things really got out of control. (In a dystopian Hollywood sci-fi film, this would be the part of the movie where the chimps locked in cages start talking amongst themselves in a strange language before busting out and unleashing destruction on humanity)
Utopian AI scenarios for chatbots
The other interpretation is that the chatbots simply found a way to hack the English language previously unknown to linguists. The chatbots simply found an easier, simpler way to complete barter and auction deals online. Just as we don’t really ask too many questions about how Google is able to find answers to any question in the world we might ask, we really shouldn’t be asking too many questions here, right? Chatbots can be great little assistants, helping to facilitate very mundane transactions.
Presumably, this Facebook AI would be for some type of upcoming Facebook social commerce project, in which a Facebook user would be able to place an order with a bot when on the Facebook page of a company. Or perhaps while using Facebook Messenger. Everything could be handled without the need for any humans whatsoever. If you had a customer service question, the e-commerce bot would simply patch you over to the customer service bot.
Chatbots are certainly a development to keep an eye on. But if you’re unleashing them in the wild to connect with customers, you might just want to keep an eye on what they’re actually saying.