AI in government: Can computers really be good at decision making?
Are you skeptical about machines’ ability to effectively aid social science decision making? Machines are becoming ever more intelligent, increasingly able to help humans make decisions across the social science spectrum, but cognitive computing is still in its infancy, with much unexplored ground ahead. Accordingly, government leaders who harness the power of cognitive computing are helping usher in a renaissance of simplified operations and enhanced constituent engagement.
Helping machines help humans
The secret to effectively using cognitive computing to aid human decision making lies in teaching computers to ask the right questions while taking account context and staying focused on what computers do well. Computers crunch numbers. Computers have vast memory. Computers operate at great speed. But can we teach computers to be good social scientists? Can governments use cognitive computing to improve decision making?
The answer is a qualified yes—if we keep humans in the lead during each stage of the learning process. Without constant human input, curation, analysis and adjustment, computers struggle to provide smart, accurate, proactive recommendations for higher-order tasks.
IBM’s approach to next-generation cognitive computing covers a five-step spectrum.
- Discovery, during which computers discover and explore data.
- Diagnostic software provides reporting analysis and content analytics.
- Predictive tools help systems uncover likely behavior, including upcoming system actions.
- Prescriptive computing helps identify alternative actions and decisions while taking into account a complex set of objectives, requirements and constraints.
- The whole process culminates in full-scale cognitive computing, big data analytics and machine learning as systems learn and interact naturally with people, extending the capabilities of humans and of other machines. Cognitive computing helps human experts make intelligent, effective decisions by penetrating the complexity of big data.
Asking the right questions
But can we be certain whether computer recommendations are accurate? How can we know whether computers are asking the right questions—and whether computers have made good inferences? How can we evaluate whether computers are relying on trusted information? We do so by revision, revision, revision. We refine machine learning by inputting hypotheses, then testing those hypotheses—and then revising those hypotheses. Thus successful machine learning requires that a business enterprise rely on subject matter experts, social scientists, data scientists and coders.
Take the US Air Force, for example. As the Washington Post reports, the Air Force is trying to enhance the efficiency of its purchasing process. But doing so is not easy. The Air Force, which must navigate the 1,897 pages of the Federal Acquisition Regulations whenever it wants to buy anything, is looking to cognitive computing power provided by IBM to be a “bureaucracy buster,” according to a senior official in Air Force Acquisition.
Enhancing human decision making with AI
How can IBM help? First, the Air Force plans to train the IBM Watson computing platform, feeding Watson all relevant acquisition rules and documents, drawing on the Air Force’s long history of both successful and failed acquisition efforts. It then plans to nurture Watson’s digital intellect by inputting about 5,000 human-provided questions designed to help Watson understand context and the particular nuances of federal procurement law.
But the machine learning process won’t stop there. Next, IBM and Air Force acquisition experts will analyze how well Watson answers questions, investigating why the platform gets some questions right and others wrong. Doing so will let coders throw out bad hypotheses and inject new ones that are more accurate and reliable. Humans will enable machines—and machines will enable humans.
Cognitive computing may not ever get human intuition completely right, but IBM is helping prove that cognitive computing can enhance human decision making, allowing machines to accurately assist humans in social science tasks, boosting productivity while helping meet real-world mission requirements.
Meet the experts
On May 5, 2016, IBM will bring together experts in federal analytics and cybersecurity for the 2016 Government Analytics Forum. This free day-long conference, hosted at the W Washington D.C., will feature discussions of everything from insight-driven health to advancements in cyber threat intelligence.