Seeing past the hype around cognitive computing
Given the hype around artificial intelligence, you might be worried that you’re missing the boat if you haven’t yet invested in cognitive computing applications in your business. Don’t panic! Consumer products, vehicles, and equipment with embedded intelligence are generating lots of excitement. However, business applications of AI are still in the early stages.
Research at MIT Sloan’s Center for Information Systems Research (CISR) suggests that small experiments in cognitive computing may help you tap the significant opportunities AI offers. But it’s easy to invest huge amounts of cash and time in failed experiments so you will want to carefully target your investments.
The biggest impact from cognitive computing applications is expected to come from automation of many existing jobs. We expect computers to do—faster and cheaper—many tasks now performed by humans. Progress thus far, however, suggests that we have significant obstacles to overcome in our efforts to replace human intelligence with computer intelligence. Despite some notable exceptions, we expect the displacement of human labor to proceed incrementally.
The business challenge is to determine which applications your company is ready to cash in on while resisting the lure of tackling processes that you can’t cost-effectively teach machines to do well. We have studied the opportunities and risks of business applications of cognitive computing and identified several lessons. These lessons offer suggestions for positioning your firm to capitalize on the potential benefits of cognitive computing and avoid the pitfalls.
Business processes need to meet four conditions to effectively apply cognitive computing. These include:
1) prescribed outcomes
2) lots of repetition
3) massive amounts of relevant, interpretable electronic data
4) complex interactions among the parameters that influence optimal outcomes
AI has been successfully applied to games like chess, Jeopardy, and Go because these games completely meet these conditions.
An example of an effective business use of cognitive computing is Kabbage, which uses AI to determine who should or should not receive a loan, given a company’s predetermined goals (like maximizing interest revenue or minimizing bad loans).
Kabbage has access to vast amounts of personal data because it gets permission from loan applicants to collect their financial information from their financial services providers. Kabbage also scours social media. The loan granting process is highly repetitive and Kabbage can learn from every loan repaid (or not). As the database grows from thousands to millions, the application can repeatedly review its algorithm and assess the impact of new kinds and instances of data.
One size does not fit all
Even when all four conditions are met, a business need might not be well suited to cognitive computing. For example, some banks might be attracted to the capability of artificial intelligence to grant loans automatically with optimal results.
However, if the bank’s customer experience expectations require that employees work with customers who are turned down for loans, they may not want to apply an algorithm that adapts with new data (i.e., a cognitive application). Because it’s constantly adapting, the application cannot explain what it would take for a rejected applicant to become a successful applicant.
Machine learning requires a lot of teaching
Machines must be trained to be smart, which takes a lot of time and effort. First, you need to identify the problem and propose the parameters that could influence an algorithm. Then you must repeatedly provide massive amounts of data and report outcomes so that the machine can learn from new circumstances.
These investments can be substantial. Watson won Jeopardy only after IBM invested six years in training and development. Google’s energy management application, which reduced energy usage for cooling its data centers by 40%, required lots of sensors and the determination of data scientists who spent two years developing the system. While Google consumes .01% of the world’s electricity supply, not every company could justify that type of investment.
Consider using a vendor
Given the expense, companies may be better off waiting for specialized vendors to develop their cognitive computing applications for business processes. Vendors can recoup their investments by selling systems to multiple companies. Kabbage is an obvious example, as is ROSS Intelligence for law firms who are looking for better, faster ways to identify legal precedents.
If you decide to invest in cognitive computing, it’s best to proceed incrementally as you increase your understanding of business rules. Start with the automation of well understood, repetitive tasks. This allows people to test business rules in different contexts, analyze relationships between inputs and outputs, and revise the rules. If you go slowly, you have a better likelihood of finding relevant parameters and more useful data.
Approached this way, cognitive computing won’t be terribly disruptive, but it may be extremely valuable and possibly even a game-changing force in how work gets done.