31st March 2021. AI | Robots
AI will create jobs; robots being put through their paces
Welcome to Just Two Things, which I try to write daily, five days a week. Some links may also appear on my blog from time to time. Links to the main articles are in cross-heads as well as the story.
#1: AI will create jobs
So let’s just imagine for a moment that everything we’ve been told about the impact of AI on jobs is wrong. That rather than causing mass unemployment, it will increase the demand for labour. At least we’d have to file that under “Interesting if true.”
The London School of Economics academic Leslie Willcocks makes this case in an open access article published in the Journal of Information Technology last year.
The abstract lays out the many assumptions in the “AI will eat our jobs” hypothesis succinctly:
[that] automation creates few jobs short or long term; that whole jobs can be automated; that the technology is perfectible; that organizations can seamlessly and quickly deploy AI; that humans are machines that can be replicated; and that it is politically, socially and economically feasible to apply these technologies... Adding in ageing populations, productivity gaps and skills shortages predicted across many G20 countries, the danger might be too little, rather than too much labour.
One of the strengths of the article is that it has taken an overview of the two stories that dominate the discourse—the ‘hype’ story and the ‘fear’ story. Clearly the discourse has been dominated by the ‘fear story’ ever since Frey and Osborne announced in 2013 that 47% of American jobs were at risk from AI. More on that in a moment. But the data is typically problematic:
We should exercise caution about the numbers used to support the above views. We have to become sceptical about the more macro studies on the technology and future job numbers. The problem with all too many is that they are projections going forward, with not necessarily good data sets, often carrying questionable, even tacit assumptions and few make their methodology transparent.
That 2013 Frey and Osborne article still dominates the discourse, of course. The AI debate on jobs has moved a long way since that piece of work, but it’s never completely escaped the shadpdow of that 47% figure, certainly at the level of media coverage. Willcocks observes, of the Frey and Osborne research, “the researchers do not try to specify the speed of technology development, nor a time period for the loss of jobs”. There’s more:
First, the study, like many others in this area, does no analysis of jobs likely to be created by changes in work and technology. Second, it focuses on job and occupations, not on activities and tasks, nor the amount of work that needs to be done, which seems to be increasing exponentially… Third, the study largely factors out the key bottleneck of how commercially feasible, viable and organizationally adoptable the emerging technologies are.
It’s a long article, but the detail is valuable. In particular, it describe a number of qualifiers that tend to get lost in research around jobs and AI. There’s eight of these, so I’m not going to list them all, but they include:
“Are whole jobs lost as a result of automation? Of key interest is the percentage of the job or activity that is automatable…Studies that look at work activities as a better unit of analysis than whole jobs suggest that job restructuring will be the more normal pattern.
The impact of job creation: “until 2018 very few studies focused on job creation from new technology, though job creation has invariably happened in the past. New jobs, historically, also come from new services and business models and innovations that are made possible by changed technologies.”
“How fast will automation technologies be deployed pervasively in work organizations?.. This is more complex than studies tend to assume. “Does it give relative advantage? Is it compatible with existing ways of operating? What is the risk level? Is it too complex or not administratively feasible? Is it easily trialable with tangible outcomes?” Implementation is tough as well.
We may be short of workers. “Ageing populations in the G20 countries may well lead to significant global shortfalls in labour and skills over the next 30 years… Declining birth rates and ageing populations may well [leave] a workforce too small to maintain current economic growth, let alone meet aspirational targets.” Automation may be one way of coping with such shortfalls, and we have seen signs of this already in Japan.
There may be more work to be done. There’s evidence of work intensification since the financial crisis —“sweating the assets”—but this is unlikely to be sustainable. Secondly, the increase in data that is part of the AI story brings its own work demands. Third, there’s an increase in audit, management, and bureaucracy—“We have been creating, we would argue, a veritable witches brew of data, technology and bureaucracy.” The available solutions bring their own problems, which creates more work.
He concludes that the Robo-Apocalypse story is unlikely:
the later the study, the lower the job loss estimates. When job creation is added in, several reports even suggest that the net job loss over the next 12 years, at least, is going to be negligible.
This leaves the question of how Robo-Apocalypse has become such a powerful story, and Willcocks suggests that since ancient times technological change has acted as a repository for wider anxieties about social change.
To which I might add that in some ways technologists and economists (at least, non-specialist economists) are the worst people to tell these stories. Technologists tend to self-fulfilling stories about the future in which technologies have magical and immediate properties. Economists tend to assumptive models which treat labour markets as if they behave like other markets. In practice, labour markets are much stickier, much more social, more culturally determined, and much more dependent on knowledge than other markets. But in a world where technologists and economists have the attention of media outlets and the ear of policy makers, these flawed stories are the ones which get heard.
(h/t to Leah Zaidi)
#2: Robots being put through their paces
The science fiction writer Christopher Brown has a weekly newsletter where he explores his experience of nature in the scruffy urban fringes around Austin, Texas, where he lives. It’s always well-written, and nature often becomes a springboard for other topics. Like robots, for example:
Most of the robots in the news look more like robots, in the way the robots of the 20th century science fictional imagination looked like people or animals. Consider Spot, the robot dog developed by Boston Dynamics. Variations on this basic design have been around for a decade, and the videos from the lab never cease to deliver the wonder of juvenile fantasies made real.
Boston Dynamics, Spot’s developer, rarely tires of videos in which Spot goes through its roster of tricks—cleaning, skipping, planting—and even the name could have been taken from a children’s book.
With an eye to his nature newsletter blog, Brown sees an upside to this:
It doesn’t take much imagination to see the tremendous potential for robot labor to aid in habitat restoration, tree planting, and the monitoring of wildlife populations, even as you wonder how the coyotes and deer will react the first time a creature like Spot walks into their woodland.
But he’s a science fiction writer too, and he also reminds us of the downside: that the rise of robots has been accompanied by the emergence of organisations such as the (very serious) Campaign to Stop Killer Robots.
j2t#068
If you are enjoying Just Two Things, please do send it on to a friend or colleague.