I’ve looked previously at the rise of so called automated leadership, with the scheduling and appraisal of employees largely done via algorithm, with researchers exploring just how people feel working under this kind of leadership.
That is but one part of the infusion of automation into leadership however, with things like forecasting and other forms of data analysis handed over to computers for a while now.
Automated decision making
A team from the University of York and software company MooD International are teaming up to use a mixture of AI and gaming technology to help management decision making.
The work revolves around the so called Monte Carlo Tree Search, which is a commonly used algorithm for decision making in video games. The aim is to make a similar algorithm for use in the workplace.
This will involve a visualization the kind of impact a decision will have, thus hopefully making it easier for managers to ensure they make the right choices.
“Despite huge developments in AI research and gaming intelligence in recent years, there is still a big gulf between such techniques and how they can be applied to real-life situations,” the researchers say.
Data mining
A recent paper builds on this approach and deploys a specialized algorithm to help find the best solution to tricky, multi-objective problems. The approached, dubbed MOEA/D (multi-objective evolutionary algorithm based on decomposition), works by breaking multi-objective problems into single sub-problems that can then be tackled individually. The algorithm aims to create an optimal set of possible solutions for each sub-problem.
Researchers have further improved this approach by utilizing a chain-reaction method that deliberately overlooks duplicate solutions to a problem, thus ensuring higher diversity levels in the solution population.
It also aims to order existing solutions so that neighboring solutions aren’t deleted, even if new solutions are found to be better. Instead, it defers to the end user to decide which solution is best for their particular problem.
The algorithm does provide for version control however, and will replace existing solutions and back track to update any alternative search directions accordingly.
The system has been trialed in environments with 2-8 objectives, and improvements were shown over existing methods, but it goes without saying that a lot of work is needed to transform this into a solution fit for the marketplace. It does however, underline the progress being made, and the potential for algorithmic support around the corner.
