Reinforcement learning (RL) is a sub-area of artificial intelligence (AI), which offers distinctive advantages over other AI approaches. While most AI models are only “trained” once and then rely on their existing training when processing new examples, RL models continue to “learn” based on their performance after they have been deployed.
An example of reinforcement learning in action is when software plays Monopoly or chess, and continually improves, allowing it to beat a human. The best-known example of reinforcement learning in action was when AlphaGo beat a human in a game of Go. Go is known as the most challenging classical game for artificial intelligence because of its complexity, and was impossible for programs built on supervised learning to win.
Our client was investigating hardware and software capabilities related to reinforcement learning.
Before investing in this emerging technology, our client required a five-year roadmap covering future customer demands, potential use cases, key enabling technologies and techniques that could unlock new white space opportunities for their organisation. Specifically, they asked for our help in determining which new products or services they should provide to capture new business opportunities and satisfy customer demands. The client also sought specific use cases for reinforcement learning and technologies and techniques that would be most suitable for these use cases over the next half-decade. The client's third request was that our expert team find and evaluate innovative commercial and academic players who could become potential partners (or competitors) over the next five years.
To research this emerging technology area, we assembled a project team that included two world-class academic and industrial experts in reinforcement learning. Our academic expert detailed some of the leading open-source and proprietary frameworks for the development of solutions using reinforcement learning. Our industry expert evaluated specific use cases, identifying those where the new technology offered distinct advantages over other machine learning and AI approaches.
Our research identified reinforcement learning-enabled products/services that our client should consider adopting over the next half-decade as part of its innovation pipeline. We advised on key technologies, techniques, and breakthroughs required to scale RL-based solutions to an enterprise level, and determined seven mature application areas for potential new product/service development and six emerging application areas where potential products and services could be developed. We also provided a final shortlist of 10 technology-enabled product/services that the client should consider adopting into their pipeline within 12, 36, and 60 months of commercialisation.