Rethink AI Objectives

When customers feel like no one is listening, they may be right. Or, they may be unknowingly talking to a machine. Last week, Google announced Duplex, an artificial intelligence (AI) assistant that can handle customer service requests, such as booking an appointment or providing basic information. We’ve had automated phone attendants for years, but the new buzz is that customers can’t tell that this automated attendant isn’t human.

The Duplex announcement quickly brought questions about transparency. Should companies include a notice that “objects on the phone are less human than they appear?”

Of course they should (and Duplex will). Investing massively in technology to be intentionally duplicitous seems like it belongs in the junkyard of bad ideas we shouldn’t even consider, such as earthquake generators or encryption back doors. What would be the benefits of being dishonest with customers? Is the “gee whiz” worth the potential backlash?

Just because we can do something doesn’t mean we should. Consider the benefits of being up-front about interactions with bots:

• Customer satisfaction can increase. For example, if customers know they are interacting with a computer and not using another person’s time, they may spend more time customizing their selection. I’ve certainly spent far more time online investigating travel options than I would ever subject a human travel agent to. With better customization, customer satisfaction can increase.
• Customers can be better informed. When interacting with other humans, people hesitate to ask what they perceive to be dumb or repetitive questions. We can be reluctant to admit that we still don’t understand. But when interacting with machines, we are free to clarify to our heart’s content. As the use of technology removes the concern about long call times, we don’t have to worry about using a machine’s time.
• Customer interaction can be faster. Social norms dictate that we first lubricate the conversation with bromides and banalities about weather and pleasantries. I deeply suspect that most customer service people actually don’t care if “I’m doing OK for a Monday.” Knowing that we are interacting with a machine, we can get straight to the point, saving everyone time.
• Customers may be more honest. We care what other people think about us. As a result, we may not be completely honest when what we think differs from what we believe the other person wants to hear — the social desirability response bias. We are less likely to worry about what a machine thinks about us than what another human thinks.

So why are companies even considering the duplicitous option? Like dogs who don’t know what to do once they’ve caught the car they’ve been barking at and chasing, we’ve chased the Turing test (which tests whether a machine can mimic a human’s intelligent behavior) and are not sure what to do now that we’ve caught it.

For science, passing the Turing test makes sense as a goal. For business, other goals most likely make a lot more sense. Even if computers can pass for humans, human mimicry may be a shortsighted goal. We already have humans. We don’t need to imbue these bots with the “hmms,” “ums,” and “uhs” that are part of normal imperfect speech patterns. Duplicating misses an opportunity for more.

If not “humanlike,” what should the objectives be? In this context, they might be:

• Using AI to learn from prior customer calls and avoid them entirely. Encoding and decoding requests back and forth into speech may show impressive technological prowess but probably means the customer hasn’t found the information they need from other channels. The AI attendant may be superfluous, automating processes that shouldn’t exist at all. Customer satisfaction may increase with a clearer online appointment system that avoids the call in the first place. Instead, ongoing analysis of call logs using AI can be part of overall process improvement.
• Using AI to assist humans during a call. By analyzing call content, businesses can use that information to improve security and improve service. These are aspects better done by machines than humans.
• Using AI to anticipate calls and plan staffing. Pretending to be human is far from the first deception in customer service calls. Instead, we now endlessly hear “your call is very important to us” as we wait while the company prioritizes staff efficiency over customer time. While this may be perfectly rational, AI can better anticipate call volumes and improve both company and customer efficiency.

AI can be a phenomenal tool to improve business and customer service (specifically). But there are much better ways to use this technology than mimicking humans and fooling customers. After all, we’re still supposed to be the intelligent ones here.