The rise of cognitive agents Will humans prefer to talk with machines?

One of the great concerns about artificial intelligence (AI) is that it will replace people altogether. Ironically, however, it is increasingly apparent that interacting with people is one of the key tasks of AI. “Cognitive assistants”—systems that employ cognitive technology to interact with people and make our lives easier—are among the fastest-growing areas of this genre. They are likely to transform many aspects of business in the near future.

Thus far, of course, the primary examples of this technology have served consumers rather than businesses. Cognitive assistants already set your sleep alarm, turn down your thermostat at night, and tell you what movies are playing at the mall. These applications will likely become even more common as a new generation of personal, social robots is introduced to consumers in the next few years.

But the opportunities for cognitive assistants in business have been unexploited thus far. There are many different business tasks and processes that could benefit from them, however, and this category of application is beginning to emerge. They can provide access to complex information, perform digital tasks like opening a bank account or checking a patient into a hospital, make recommendations about products or services, and bring awareness of context to an interaction. Cognitive assistants can even understand the emotional content of a conversation and react accordingly.

The technologies behind cognitive assistants build on decades of research in natural language understanding and generation, semantic decomposition, and even machine and deep learning. Because the underlying capabilities of these technologies are increasingly modular (provided as application program interfaces, or APIs), companies can assemble them as needed to address particular interaction needs with customers, suppliers, or employees.

Some AI applications are extremely difficult to implement, such as how best to treat cancer. But cognitive assistants are easier. They usually don’t have to be perfect. And because they involve iterative interactions with humans, they can take advantage of humans’ willingness to repeat or rephrase comments or requests. In some cases they may even ask humans to listen to a set of choices and choose among them, as did previous automated interaction systems. But the idea behind cognitive assistants is to require much less patience of the humans interacting with them, and to respond naturally and accurately to human speech and text.

In what areas of business are we most likely to encounter cognitive assistants in the near future? One area that is often mentioned is customer service, and it’s certainly true that many aspects of this function could be augmented with cognitive helpers. This might encompass billing and account interactions, tech support, furnishing lost passwords, and so forth. Since existing technologies do not often provide high-satisfaction customer support now, cognitive assistants have a good chance of improving things dramatically. It still may be the case, however, that human agents will need to intervene at times, and companies should provide an opportunity for this.

But not only customers can benefit from cognitive assistants. Suppliers, for example, can check on whether ordered products have arrived and when they’re going to be paid. Employees can check their vacation balances or learn whether they’ve exhausted their prepaid medical balances. At one company we’re familiar with, warehouse employees using a cognitive assistant can get detailed information about the products they’re seeking—and perhaps not finding—in the warehouse.

Of course, cognitive assistants can’t take over human interaction processes out of the box. Even if an existing system is acquired from a vendor, it will need to be trained for the specific set of tasks it is supposed to perform. This can be quite complex. One company we worked with found that there were 3,000 different types of problems that customers might want addressed during a call. The “triage” process was so challenging for cognitive technologies that the company decided to use human agents for the task, and then route customers to cognitive assistants that could handle specific problems. Even then, however, training on all those issues was a very large problem.

Cognitive assistants are likely, of course, to have some implications for a company’s employees. Those who staff call centers, for example, are certainly at some risk of losing their jobs. As we’ve noted, it’s unlikely that all human agents will be made redundant, but some will. Human agents who understand how cognitive assistants work, their strengths and weaknesses, and how to monitor and improve their performance over time will certainly be the most likely to preserve their own employment. It’s also likely that the most experienced agents and supervisors will be retained, making entry-level call center jobs harder to get.

Cognitive assistants are also likely to change the outsourcing landscape dramatically. Call centers have historically been among the functions most likely to be turned over to BPO (business process outsourcing) firms. It may be that those BPO providers will themselves adopt cognitive assistants to offer to their clients, and we know of some that are planning to do so. But many companies hired outsourcers because of the challenges of managing large call center staffs, and they may feel that managing much smaller staffs and a coterie of cognitive assistants is much less burdensome. In other words, cognitive assistants may drive firms to take back call center management. To try to avoid this, BPO firms will have to strive to add value to cognitive assistant work, perhaps by offering to train them rapidly or endowing assistants with detailed knowledge of industries and processes.

The world of cognitive assistants for businesses is a brave new one, and few organizations have detailed experience with it. There are sure to be some surprises and disappointments. But we are confident that the technology holds considerable promise for improving human interactions with businesses. Before long, we may even see that humans actually prefer to talk with machines because of their vast knowledge, ability to take detailed contextual factors into account, and inability to become upset or befuddled. And to never again be told, “All of our customer service agents are busy—please wait until the next one is available to take your call” would be a small boon to humankind.