Can AI help new workers get up to speed?

Originally posted on The Horizons Tracker.

It’s often said that new hires require up to a year to be completely productive, with this timeline not just due to learning the particulars of that specific role but also the organizational culture and the networks within the company that underpin how work is done.

These challenges are particularly strong when you’re working in a remote role that doesn’t give you the kind of face-to-face exposure that can make tacit learning easier. Research¹ from UC Berkeley proposes an AI-based system to help this learning process.

“Think about gig workers or physicians in rural areas who don’t have a chance to learn from their peers every day,” the researchers explain. “Of course, performance suffers during this learning period.”

Learning the ropes

The researchers hope that their algorithm can help new hires learn things that may be counterintuitive and therefore difficult to pick up without someone physically there to support them.

“What was especially interesting is that people in our study didn’t blindly follow the tips but combined them with their own experience to learn parts of the optimal strategy that weren’t even mentioned,” the researchers continue.

They believe that this approach is valuable because it doesn’t automate human judgment out of the equation, but instead, provides targeted support and guidance to help us learn more effectively. We’re essentially being helped to come up with the best solutions ourselves.

Complex decisions

The study found that one-off decisions are fairly easy, especially if they have immediate consequences. These types of decisions are easy to optimize for, but those that entail many more steps are far harder, as each choice affects everything that comes after in a cascade of complexity.

The researchers believe that the data that many organizations are collecting on their workers may help, as this data can be used to train AI to better understand what best practice looks like, and then to distil that into relatively straightforward rules or heuristics to help people learn.

This concept was put to the test via a virtual kitchen management game. The game required players to complete various challenges, all of which involved a number of subtasks. Participants would have to assign a range of tasks to virtual workers, each of whom had varying skill levels. The ultimate goal was to ensure customers received their food as quickly as possible. In some of the sessions, problems were introduced, such as the removal of a key team member.

The optimal approach

To begin with, the researchers developed an algorithm using reinforcement learning to help them devise the best approach and then turn that into a straightforward rule for people to follow. This was then tested on around 2,300 volunteers who were split into one of four groups, one of which received no advice, the second of which got advice from their peers, the third got advice from a simple computer program, and the last got advice from the algorithm.

The results were pretty conclusive, with the group that received advice from the AI performing the tasks much faster than those that did not. The performance gaps were especially stark when the tasks were made more complicated by the disruption, with 19 times as many people getting the optimal performance when they got help from the AI.

Interestingly, the group that received support from the standard computer performed particularly poorly, as they failed to adjust their strategy. Similarly, the advice given by other humans was often either completely wrong or too general to be of any use.

“The algorithm captured the discrepancy between the existing human action and the optimal policy, which helped identify the best performance-enhancing tip,” the researchers explain. “This opens up exciting possibilities for using the wealth of workplace data that companies already collect to automatically identify and share best practices.”

Not so straightforward

Getting the participants to actually follow the optimal advice was far from straightforward, however. None of them were told whether the advice they got was from a human or a machine, but they were nonetheless more likely to follow the advice given by a peer than they were to follow that from the AI.

The researchers suggest that this is because the human advice was quite intuitive, whereas the AI advice was often anything but. They only learned to follow it after completing a few rounds, so their experience allowed them to appreciate the importance of the advice and combine it with their previous experience.

“This suggests that AI can guide human learning in ways that go beyond simple instruction-following,” the researchers conclude. “In most settings, there are nuances where it’s just not feasible to automate everything. But if you can find that one simple and effective piece of advice, people can start to figure out best practices on their own.”

Article source: Can AI Help New Workers Get Up To Speed?

Header image source: Andrea Piacquadio on Pexels.

Reference: