Viewpoints

About Artificial Intelligence

Virtual assistants that listen and speak, software that outperforms human masters of strategy games, and other newsworthy developments have led many people to wonder and worry about the future of artificially intelligent systems, especially systems that take advantage of recent innovations in machine learning for deep artificial neural networks. This Technology Map mainly defines AI as systems that perform activities that otherwise require an intelligent human. Although the methods that AI software uses to perform these activities are quite different from the way that human intelligence works, people have reason to monitor whether software and intelligent robots will introduce economic efficiencies, displace many workers, cause social discord, or have other combinations of disruptive effects.

All current AI-development work that shows measurable progress is for specialized systems that perform narrow tasks: Machine-learning algorithms seem to be driving AI toward uncanny capabilities that do not emulate human intelligence, as in the case of game-playing computers, whose styles of play differ greatly from those of human players. But a system that plays games does not recognize speech, and a system that identifies cancer in chest X-rays does not perform other tasks that occupy the majority of a radiologist's professional time. Commercial success stories in the field are all examples of narrow AI (application-specific solutions), including big-data analysis platforms that automatically generate insights about business operations, speech recognizers for virtual assistants, and face recognizers for automated tagging in social-media posts and police files. Some uses of AI have been counterproductive, confirming people's varying and mixed opinions about AI's promises and perils.

Although a faction of researchers' labors toward artificial general intelligence, current progress in AI mainly depends on efforts to solve particular problems. Active domains of development include transportation, robotics, office productivity and collaboration, security, transactions, logistics, entertainment and advertising, and social interaction. Some of the work promises or threatens to replace human workers, such as efforts to automate customer service or develop driverless trucks. Many other developments emphasize human-machine collaboration, such as machine-translation systems and intelligent cybersecurity tools for trusted personnel. Improved collaboration and increased autonomy alike could arise from efforts to produce explainable AI systems that provide a rational justification for their behaviors. AI researchers also aim for breakthroughs in efforts to incorporate the latest findings from the field of neuroscience, which has its own road map for discovering how natural brains work. Other hoped-for breakthroughs that involve quantum computers and other novel hardware might accelerate existing machine-learning processes by orders of magnitude. Whether AI technologies become generalized or remain specialized, whether they are mostly autonomous or mostly collaborative with humans, and whether or not their actions are sufficiently explainable to people, for good or ill the current technology road map seems to point to potential changes to people's roles in workplaces and communities. Monitoring AI developments will be important for evaluating expectations of business conditions and the factors that affect constructive AI-development and public-policy decisions.