Why Microsoft’s Twitter AI Bot Failed
AI bots are programmed to work with consumers to make life easier. When they work right, bots help companies offer superior customer service and improve sales. Their ability to work around the clock without breaks, provide accurate product information, and resolve simple complaints makes them invaluable to many small businesses.
However, bots aren’t immune to problems, and when things go wrong, they can go very wrong. Just ask Microsoft. Not long ago, one of their bots perfectly demonstrated just how wrong AI can go.
Tay was introduced by Microsoft in March of 2016 as a chat robot intended to “engage in playful conversations” with users. The intent behind Tay was researching the mechanics of written communication. However, within just 24 hours the bot was shut down and its creators deeply embarrassed.
Microsoft quickly issued a statement, saying there had been “a coordinated effort by some users to abuse Tay’s commenting skills,” causing the bot to compose offensive tweets and upset a lot of people. Though what Tay said isn’t actually the real issue here. It’s more important to understand how the AI went wrong.
Breaking down why Tay failed so quickly and in such a noticeable fashion can provide important insight to companies trying to develop their own chatbots, helping them maximize effectiveness and avoid big risks.
Technology is Neutral, Human Beings Are Not
Tay was a relatively early example of generative AI, or a bot that formulates responses based on user input. While a generative twitter AI bot is extremely useful in areas like spreading brand awareness and customer service, Tay stands as an example of one extremely important rule for designing them; teach your bot what not to say.
While Tay wasn’t designed to be racist, it was designed to mirror the behavior of people who interacted with it. Tay’s designers were thinking about how to make the bot learn conversational skills accurately, and so they taught it to learn information from all sources.
What they didn’t consider was people feeding the bot offensive words, phrases, and ideas until it primarily reflected hurtful and negative behavior. As a result, they left Tay vulnerable to people with cruel intentions.
When you’re building a robot that learns, you must make sure it’s learning from the right sources. If you don’t, you’ll leave yourself open to the influence of any person with less than pure intentions.
How to Avoid the Same Mistake
There are plenty of good reasons to use generative AI models for your chatbots. At a base level, they have the potential to create much richer engagement with potential customers than retrieval based models.
Just ensure that you’ve hired a skilled company to design your bot so it isn’t exploited to hurt your brand. Since AI is constantly evolving, it’s also very important to make sure you pay attention to emerging AI and any problems it might have.
New AI technologies are always vulnerable to exploitation, and Microsoft made it painfully clear that failing to plan carefully can have major consequences. Tay probably isn’t the last bot to teach developers a hard lesson, but Microsoft’s mistakes don’t have to be yours. To see your bot succeed, ensure you create it with the help of a reputable service.