MIT Explores What Impact AI Will Have on Work
MIT Explores What Impact AI Will Have on Work
Let's take a brief look at how MIT is exploring the impact that Artificial Intelligence will have on work.
Join the DZone community and get the full member experience.Join For Free
As AI technologies have become more powerful in recent years, there has been growing interest in how the technology will impact our work. The latest of these comes from one of the doyens of the field. MIT Sloan’s Erik Brynjolfsson has recently penned a paper that attempts to provide a more nuanced exploration of the topic than the utopian and doomsday predictions that occupy either end of the spectrum.
The paper argues that whole jobs are very unlikely to be disrupted or replaced by machines, with a much more likely scenario being that specific tasks will be replaced instead, although Brynjolfsson does concede that some jobs are likely to have more of their tasks automated than others.
“Our findings suggest that a shift is needed in the debate about the effects of AI: away from the common focus on full automation of entire jobs and pervasive occupational replacement toward the redesign of jobs and re-engineering of business practices,” the authors say.
The True State of AI
The authors argue that many of the more outlandish predictions for AI are based upon unrealistic expectations of the technology. Indeed, many make the assumption that artificial general intelligence is just around the corner, when the reality is that we are still a long way off.
Instead, the applications of AI today are in very narrow domains. Of course, that’s not to say that these narrow applications aren’t powerful and can’t have a profound impact upon the world. Such technologies are working in areas such as image and speech recognition, natural language processing and predictive analytics.
From this basic reality of the situation today, the researchers then examined the kind of tasks AI can and cannot do today. They developed a 23-question rubric to gauge whether a task is suitable for current AI capabilities.
“Any manager could take this rubric, and if they’re thinking of applying Machine Learning [to a task] this rubric should give them some guidance,” the researchers say. “There are many, many tasks that are suitable for Machine Learning, and most companies have really just scratched the surface.”
Susceptibility to Automation
The team then used data from the Bureau of Labor Statistics to understand the kind of tasks that typically make up a job. It’s a similar methodology to that used by Oxford researchers Carl Benedikt Frey and Michael Osborne. The MIT believe that this has given them an insight into the risk of automation of some 900 different professions.
Indeed, the analysis is very similar to that earlier work by Frey and Osborne, with one noticeable and crucial difference. Whereas the Oxford pair identified occupations where there was a likelihood of certain tasks being automated and conflated that with the entire job being automated, the MIT team haven’t made that leap. This gives their analysis a more realistic outlook, finding as it does that many tasks are likely to be transformed by AI, but whole jobs much less likely. Instead, the transformation will be the reorganization of tasks rather than the replacement of jobs.
For instance, radiologists are a profession that are frequently in the crosshairs, as numerous AI applications have proven to be effective at rapidly and accurately diagnosing various conditions based upon medical scan images. That most of these applications have been both faster and more accurate than human radiologists has led many to predict their demise under the stampede of technology’s advance. This is unlikely, however, as there are 26 distinct tasks associated with being a radiologist, and whilst analyzing medical images is well suited to AI, interpersonal skills are currently not.
“In almost every occupation, there are at least some tasks that could be affected, but there are also many tasks in every occupation that won’t. That said, some occupations do have relatively more tasks that are likely to be affected by Machine Learning,” Brynjolfsson says.
This is likely to result in a much better matching of skills with job requirements, with Machine Learning being used for tasks that it is inherently well suited to, and this then freeing up humans to do the things that we are inherently well suited to.
The ability to have a more realistic understanding of the impact technology is likely to have is crucial if we’re to have a reasoned and successful response to the changes wrought by the technology. This kind of research will undoubtedly help to do just that.
Published at DZone with permission of Adi Gaskell , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.