Things went haywire, and you don’t want that to happen to your brand. Here’s our view:
Operationally, this is a notable fail from one of the leaders in the Machine Learning (ML) and AI space. Aside from a technological assessment of how they could have done better, it’s amazing that they turned this system loose into the real world without more closely monitoring its launch.
Although embarrassing, the business consequences for Microsoft will be pretty minor. However, as chatbots and other AI-enabled tools become more common in the customer support world, there are an increasing number of sobering examples of why you should never turn AI loose onto your customers without a human circuit breaker.
There’s been a tremendous amount of hand wringing recently in the popular media around the evolution of AI and it’s potential to replace human beings in a variety of functions, especially in the workplace. While it’s true that certain tasks will, and are already being, automated, this is exactly the wrong way to think about the application of AI in a customer-facing environment.
A lot of AI systems perform great offline, and then fall apart when pushed into production within the very unpredictable environment of the real world. When there are business costs associated with making errors, it’s critical to add human decision making into the AI feedback loop to act as a circuit breaker, and to prevent AI from taking actions that could be damaging. When human beings become part of that system, we see a shift from full automation of tasks to augmentation.
Human beings, especially when interacting with other human beings, will always demonstrate superior best judgement compared to an automated system. The huge opportunity, especially in the customer support space, is not to replace agents with automaton, but to use ML as a powerful tool to make them better.
The best application of AI in the support world is as a tool to enhance your support team. Augmentative tools powered by ML can make agents happier, more productive, and help them deliver the best customer experience possible.
In the human-machine interface of ML-based systems, providing mechanisms to communicate next-best actions that are both reliable and interpretable is crucial. Human trust ultimately decides the fate and efficacy of augmentative tools in a business environment.
Trust, in the face of what is fundamentally an imperfect and fallible system, is grown by fostering and rewarding interactions that incorporate human-powered feedback.