Turing-award winner and AI pioneer Geoff Hinton confidently predicted in 2016 that “within five years, deep learning is going to do better than radiologists,” and advised medical schools to “stop training radiologists now.” To which radiologist Keith Dreyer responded: “The only radiologists who will lose their jobs to AI will be those who refuse to work with AI.”
Dryer is quoted in a new book by Thomas Davenport and Steven Miller and his more informed prediction could have served as the book’s motto. Working with AI: Real Stories of Human-Machine Collaboration is based on 29 case studies of human-machine collaboration in enterprises that successfully use AI to augment rather than replace their employees. It is an engaging, well-documented investigation into the nature of work today and tomorrow.
Davenport and Miller remind us that Hinton was in a good, albeit wide of the mark, company. Predictions about the impact of AI on employment ranged from 50 percent of all jobs lost to 5 percent jobs lost (and a net gain of tens of millions of jobs). “The two of us are firm believers in augmentation as the primary impact of AI,” say Davenport and Miller. With Working with AI, they prove that their belief is the reality of many enterprises today and provide a guide to the future of work.
The companies serving as case studies of the power of augmentation range from financial services (e.g., Morgan Stanley, Mass Mutual, DBS Bank, Radius Financial Group), to software (e.g., Salesforce, Intuit), to online/digital (e.g., FarmWise, Mandiant, Shopee, Stitch Fix), to healthcare (e.g., Stanford Health Care, Good Doctor Technology) to organizations in other corners of the global economy. For each case study, the authors describe the work context of the system of humans plus smart machines, and include interviews with workers, managers, customers, and technology suppliers. They outline the lessons they learned from each case study and offer their insights and conclusions along themes that cut across all cases.
First and foremost, teamwork is behind the successful introduction of AI in the enterprise. “One of the remarkable aspects of all the case study examples in this book is that they involve highly complex collaborations across a number of different groups and actors, both inside and outside a particular organization,” write Davenport and Miller. It takes a complex and highly-coordinated team and it takes time.
“Everybody’s a techy” is another important observation. The case studies highlight the transformation of the traditional, isolated and insulated, “IT department” into technology and tools that are embedded everywhere in the enterprise. This new technological reality drives the rise of hybrid roles, merging business and technology skills, expertise and experiences. Today, write Davenport and Miller, “The issue is not whether one needs to assume a business-IT hybrid role. Rather, it is a choice of what type of hybrid and to what degree.”
The quality, richness, accessibility, and usability of data are important factors in the success of AI programs and the support systems for acquiring, integrating and managing the data are important success factors in the deployment of AI. These support “platforms” and the people that develop and maintain them, argue Davenport and Miller, must get the attention and support they deserve.
Another important support mechanism Davenport and Miller found in their research is what they call “intelligent case management systems. These are a type of end-user application environment that help in workflow management, prioritization, recommendations, and data integration. An important aspect of these systems is that humans can override AI: “One of the great advantages of humans and smart machines working alongside each other is that humans can confirm that an automated decision is ‘sensible.’”
As to the current and potential impact of AI on jobs, and especially its impact on entry-level workers, Davenport and Miller describe both negative and positive situations they found in their research. They conclude that “a reduction in entry-level employment opportunities for knowledge workers is still a prominent and imminent threat. At the same time, it is still an inconclusive and evolving situation, with multiple counterbalancing factors and influencing economic forces.”
In the course of their investigation, Davenport and Miller have found that remote and independent work, assisted—and facilitated—by smart machines, has become more common. They see both upside and downside to this type of work, but stress that workers need to interact, talk to one another, and socialize, so they can continue to learn and come up with new ideas.
Finally, “what should be done about AI limitations and human strengths?” Davenport and Miller ask. And answer: Let humans override AI decisions; AI experts should understand well the work context in which their Ai solutions are deployed and explain to all involved what AI can and can’t do; and everybody should understand that AI implementation is a change management activity.
Working with AI was published towards the end of a year in which AI “Foundation Models” grabbed headlines for their “creativity,” “sentience,” and human-level “understanding” and “reasoning.” So it was encouraging for this human (and skeptical) reader to find out that Davenport and Miller “suspect that even with the relentless trajectory of improvement in AI capabilities, widespread augmentation is here to stay.”