As internal audit functions race to keep up with their organizations' artificial intelligence (AI) initiatives, two studies reveal current trends and where the technology is going.
AI research and development is picking up speed, notes the
AI Index 2018 Annual Report (PDF), based on trend data from a variety of studies. Categories of greatest growth include machine learning and probabilistic reasoning, neural networks, and computer vision, according to an analysis of AI research papers. Most papers published in 2017 covered machine learning and probabilistic reasoning.
In the learning space, one of the biggest trends is language processing, according to a Stanford News Service
press release about the AI Index. Most information on the internet is text, but AI struggles to learn the intricacies of human languages. Computer scientists are trying to improve AI's comprehension of written languages to "understand that treasure trove of information," says AI Index leader Yoav Shoham, professor of computer science, emeritus, at Stanford University's Human-Centered AI Initiative.
Shoham explains that AI has learned to solve "narrow" problems such as translating languages and keyword searches. The next step is teaching AI to put different pieces of information together to answer more complex questions.
In many ways, AI already has mastered some tasks such as identifying images — often better than people can do it, the index states. And AI is learning things much faster. For example, in about one year, the amount of time needed to train an AI network to classify pictures from the ImageNet database dropped from one hour to about 4 minutes.
Companies are ramping up efforts to exploit AI, as well, the index notes. In the U.S., the number of AI startups has more than doubled since 2015, according to Sand Hill Econometrics.
McKinsey & Co. notes widespread adoption of AI across industry sectors and business functions worldwide. Telecommunications, travel and logistics, and financial services are leading users of AI for service functions. High-tech and telecommunications exploit it for product development, while retail and telecommunications are the leading AI users for marketing and sales.
Whether any of these AI trends will benefit people is the focus of a new Pew Research Center study,
Artificial Intelligence and the Future of Humans. The nearly 1,000 technology pioneers, innovators, business and policy leaders, researchers, and other respondents say networked AI may make people more effective. For example, they say computers could exceed human capabilities for complex decision-making, sophisticated analytics, and speech recognition and language translation. Moreover, smart systems could save time, money, and lives, they say.
Despite such potential benefits, these experts are concerned about the long-term effects that AI could have "on the essential elements of being human." Concerns include:
- Loss of personal control over people's lives as decision-making in digital life is increasingly performed by AI, with little input or knowledge of how AI works.
- Data abuse and surveillance by systems designed for profit or to exercise power.
- Job loss from AI taking over jobs, which could widen economic divides.
- Dependence on AI that results in people losing cognitive, social, and survival skills.
- Mayhem from AI-based weapons, cybercrime, and information.
"Questions about privacy, speech, the right of assembly, and technological construction of personhood will re-emerge in this new AI context," says Sonia Katyal, co-director of the Berkeley Center for Law and Technology, in the Pew report. These factors may throw beliefs such as equality and opportunity for all into question, she notes.
Despite such concerns, 63 percent of respondents are hopeful that most people will be better off in 2030. Stanford's Shoham notes that AI is more likely to supplement people with smart technologies and automated processes than to replace their jobs. "Historically, technology has been a net job creator," he says. "It just changes the nature of the jobs."