What’s on the horizon for cognitive computing is not only a matter of technological capability; organizations must embrace cognitive computing in practical terms by understanding the risks and limitations for implementing those new technologies to solve longstanding knowledge work challenges. Lessons from recent analogous technology shifts can help predict how cognitive computing will play out and how the knowledge management (KM) field should respond.
Last year, member-based nonprofit APQC assembled its 9th KM Advanced Working Group to explore the practical potential of cognitive computing in KM. Over five months, KM leaders from APQC, Deloitte, EY, NASA, Pfizer and the U.S. Army Training and Doctrine Command evaluated existing and anticipated capabilities.
The first article in this series explained the group’s analysis of basic cognitive computing concepts and how it believes the technology will influence the future of knowledge work. The article examined KM areas in which cognitive systems show the most promise, including content curation, search and discovery, expertise location, lessons learned analytics, data visualization and intelligent personal assistants.
This final article in the series looks at what may impede the adoption of cognitive computing in the context of KM. It also offers implementation advice related to strategy, change management and adoption, and measurement and sustainability, as well as predictions for the next three years of rapid maturation.
Limitations and challenges
Obviously, the vast potential of cognitive computing technology is balanced by significant risks and caveats that might impede adoption. Some of the challenges and limitations that cognitive computing faces are like those of any new enterprise technology, whereas others are specific to this field.
From a technical perspective, cognitive computing and machine learning were originally designed to make sense of massive amounts of data. The systems require a lot of raw material from which to extract insights, learn and predict. Large global organizations may have a sufficient baseline of content and transactional and digital information to derive value from machine learning algorithms, but small and midsize ones may struggle to reach critical mass for the more advanced systems. And if sufficient data must be compiled before the cognitive system can start generating results, implementation may not represent the “quick win” required by leadership. See chart on page 25 of KMWorld, March 2017, Vol. 26, Issue 3 or download pdf of chart.
Another significant challenge for implementation involves “training” cognitive systems. In many cases, enterprises need not only a sufficient data set, but also skilled resources who can invest time in tuning the cognitive engine before valuable outputs can be gained. That may create an initial barrier to entry for some organizations.
The next challenge is the cost of implementation. Cloud and app-based systems are being developed that at a minimum facilitate affordable trials of the technology. However, initial full-scale implementations that harness cognitive computing’s true potential are likely to focus on areas where the most ROI can be gained. Those may be domains such as sales, marketing, and risk and fraud detection—not necessarily the realm of knowledge management.
Furthermore, there are significant legal and privacy implications to searching all that data, especially when it comes to information that people perceive as personal, such as their email exchanges, search queries and downloads. Laws vary from country to country, and business rules for access and use are still being written. Organizations should tread carefully to avoid alienating the very employees they hope to aid and empower.
But even if KM can make a compelling business case and get over the legal hurdles, are employees ready for the new tools? Currently, cognitive systems position smart machines as “assistants” providing recommendations to human actors, but letting an algorithm answer your questions still requires a certain level of trust. The next generation of cognitive applications will put computers even more firmly in the drivers’ seat and require increasing confidence in their capabilities.
When Dutch company MotivAction conducted research on that issue, it found that 75 percent of people wanted to remain in control of all decisions, 20 percent were comfortable allowing computers to decide occasionally and only 5 percent would leave all decisions to computers. Moving to a cognitive computing-powered world may require more user hand-holding than programmers and early adopters realize.
A final challenge for cognitive computing is the scarcity of data scientists and developers relative to the demand created by the current era of big data. Organizations may not have the skills required to develop those applications (where data scientists are required) or to interpret the results.
Lessons from previous technology shifts
As with any new technology, cognitive computing enthusiasts (and vendors) promise that the new applications are so appealing and intuitive that adoption will be seamless and automatic. But anyone who has been through a major IT project knows that new tools, no matter how wonderful, require design time and adjustments in employee mindsets and behavior. And organizations that ignore change management do so at their peril.
Because the application of cognitive to KM is so new and untested, organizations must gather relevant lessons from analogous technology shifts, most recently with the rapid adoption of social networking and mobile devices. A few lessons follow from the Advanced Working Group’s experiences implementing new technologies.