Despite the widespread fears about the automation of vast swathes of the workforce, such disruption has not really emerged in the years since those predictions grabbed headlines around the world. Instead, the common belief today is that the best applications of AI and automation will be very much a joint affair, with humans doing what humans continue to do best, and technology doing what technology does best.
This general heuristic is, of course, extremely broad brush in its nature and it’s far from straightforward understanding just what it is that man and machine thrive at. Indeed, a central theme of Lumina Foundation CEO James Merisotis’ latest book, Human Work, is that we often struggle to truly understand what skills we have, and indeed what skills we need to have in order to thrive in the modern labor market.
New research from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) suggests that AI itself might be able to accurately adjudicate whether a task is best suited to a human or to technology.
Task management
The researchers have developed a machine learning system to assess a task and identify whether it’s one that would be best performed by a human expert or technology. Interestingly, the researchers claim that the system is capable of learning and adapting as it goes, such that it can identify, for instance, when the expert isn’t available or whether they have a certain level of experience, before choosing whether to defer to them.
The system was trained on a range of tasks in the medical field, including chest X-ray analysis, where the system was tasked with identifying conditions such as cardiomegaly and atelectasis.
“In medical environments where doctors don’t have many extra cycles, it’s not the best use of their time to have them look at every single data point from a given patient’s file,” the researchers say. “In that sort of scenario, it’s important for the system to be especially sensitive to their time and only ask for their help when absolutely necessary.”
Kasparov’s law
Chess supremo Garry Kasparov famously noted after his battles with IBM’s Deep Blue, that an intermediate human player teamed with a computer was better than either an expert human or a super-machine operating independently. This appeared to be borne out by the MIT team, whose human-AI hybrid model was able to perform 8% better on both tasks than either human or technology could on their own.
It’s noticeable that the rapid adoption of health technologies during the Covid-19 pandemic has largely seen (relatively) new technologies bolted onto existing processes. Rather than a face-to-face consultation, for instance, they are done instead over telehealth platforms.
To integrate something such as the MIT triaging tool will require more akin to what Vijay Govindarajan refers to as task shifting, where AI will take on some of the tasks it is well suited to, freeing up the highly skilled doctors to do the work they are best suited to.
Reinventing healthcare
It’s a point that Ravin Jesuthasan and John Boudreau make in their book Reinventing Jobs, in which they outline a number of steps we should undertake to make the best use of AI.
- Deconstruct the job. This is the vital first step to make. As mentioned in a recent post, jobs are made up of dozens of individual tasks, some of which will be suitable, some of which won’t be. It’s only when you break down each job in this way however that you can make this kind of assessment.
- Assess the relationship between job performance and strategic value. The next step is to then analyze each job to understand how important it is to the strategic outcomes of the business. There are likely to be certain tasks that are hugely important, and certain tasks that are much less so.
- What automation is possible? From here, you can begin to explore the way technology can help improve processes, either fully automating them or augmenting the human being. They break potential automation down into three categories: robotic process automation, cognitive automation and collaborative robotics, with each technology best suited to perform certain tasks. They also highlight how tempting it is to jump to this point straight away, bypassing the first two steps that are so important.
- Optimize the work. The final step then brings all of this together and implements automation based upon your findings about where technology can make a real benefit to the work your employees do.
It’s a process that goes back to Michael Hammer’s famous missive in the Harvard Business Review in 1990, in which he outlines the inevitable failure of any new technology so long as it is transplanted onto processes that were designed for a previous generation of tools.
This was reflected in a recent joint strategy paper from McKinsey and EIT Health, which explored the ways in which AI was being introduced into healthcare today.
“One of the most telling messages from healthcare provider leaders was that the implementation of new AI technology was not about the solution per se, the implementation of new AI technology was not about the solution per se, but about how it was implemented and the success of the change process but about how it was implemented and the success of the change process around its introduction. around its introduction.”
Or as Mount Sinai Hospital’s Robert Freeman said, “these projects are about 5 percent technology, and 95 percent change management”. Developments such as that highlighted by the MIT team are a fascinating indication of the progress being made, but it’s clear that there is a long way to go before such technologies are a mainstream part of healthcare as we know it.
