From The Wall Street Journal:
In August, first prize in the digital-art category of the Colorado State Fair’s fine-art competition went to a man who used artificial intelligence (AI) to generate his submission, “Théâtre d’Opéra Spatial.” He supplied the AI, a program called Midjourney, with only a “prompt”—a textual description of what he wanted. Systems like Midjourney and the similar DALL-E 2 have led to a new role in our AI age: “prompt engineer.” Such people can even sell their textual wares in an online market called PromptBase.
Midjourney and DALL-E 2 emerged too late to be included in “Working With AI: Real Stories of Human-Machine Collaboration,” by Thomas Davenport and Steven Miller, information-systems professors at Babson College and Singapore Management University, respectively. But the authors note other novel titles: chief automation officer; senior manager of content systems; architect, ethical AI practice. As AI’s influence expands, its borders with the work world gain complexity. Next up: deputy regional manager of AI-prompt quality and security assurance.
The bulk of “Working With AI” comprises 29 case studies in which corporate teams integrate automation into a workflow. Each chapter ends on three or four “lessons we learned.” For each study, one or both authors typically interview not only a worker interacting directly with the AI but also the worker’s supervisor, the manager who decided to adopt the technology, the software’s developer and the company’s customers. Though they include some statistics on, say, time saved, the reports are largely qualitative.
The book is aimed at managers, consultants and students planning their careers. As none of the above, I still appreciated the accessible narratives as a diverse survey of how current technologies can expand the range of human capabilities. Some of the applications came to resemble each other, but the mild level of bland business-speak, like “stakeholder” and “buy-in,” was positively tolerable.
Early cases lean toward desk-ridden workers. One system helps financial advisers at Morgan Stanley personalize investment ideas for their clients. Another helps fundraisers at Arkansas State University target potential donors and drafts emails for them. Others suggest life-insurance premiums to underwriters at MassMutual, or help forecast sales for Kroger. In all cases, humans have the final say. And in many cases, the systems provide explanations for their outputs, listing, for example, the variables that most heavily influenced a decision.
Later cases breach the office walls. One system predicts which field activities will be most dangerous to Southern California Edison workers, and recommends precautions. Another highlights neighborhoods where crime is likely to occur and recommends that police officers patrol the area. (The latter, a form of predictive policing, has raised concerns about biased algorithms. The vendor says they’ve implemented countermeasures, but the book doesn’t elaborate.)
The benefit in most cases is increased efficiency. AI relieves employees of boring and time-consuming work, freeing them to address other tasks, such as strategic thinking or client interactions. The authors spend less time discussing ways in which machines might perform with more accuracy than humans, though they do point to Stitch Fix, where algorithms assist stylists in clothing recommendations. The company’s director of data science notes that it’s usually best not to override the AI, whose choices tend to be superior. While algorithms can be biased, so too can humans. Stitch Fix’s styling supervisor said the software nudges stylists away from their own preferences and toward those of the clients.
Many readers’ first question might be: Will AI take my job? Or: Can I replace my expensive employees with AI? The short answer from the authors is: In the near future, no. Wealthy countries are actually experiencing a long-term labor shortage. And there are still many things AI (often) can’t do, such as understand context, deal with dynamic settings, create a coherent story, coordinate people, frame a problem and know when to use AI.
The authors include an oft-quoted comment from the radiologist Keith Dreyer: “The only radiologists who will lose their jobs to AI will be those who refuse to work with AI.” The authors elaborate: “If you’re a human reading this book—and we suspect you are—that means you need to shift your focus from worrying about being replaced by a machine to worrying about whether you can add value to a job that you like where a smart machine is your collaborator. Adding value can mean checking on the machine’s work to make sure it was done well, making improvements to the machine’s logic or decisions, interpreting the machine’s results for other humans, or performing those tasks that the machine can’t or shouldn’t do for some reason.”
Link to the rest at The Wall Street Journal