Difficulty-skill balance does not affect engagement and enjoyment: a pre-registered study using artificial intelligence-controlled difficulty. Issue 2 (1st February 2023)

• Record Type: Journal Article
• Authors: Cutting, Joe; Deterding, Sebastian; Demediuk, Simon; Sephton, Nick
• Is Part Of: Royal Society open science. Volume 10:Issue 2(2023)
• Publication Date: 2023-02-01
• Subjects: difficulty -- skill -- flow -- engagement -- artificial intelligence; Science -- Periodicals; 500
• DOI: 10.1098/rsos.220274
• Languages: English
• ISSNs: 2054-5703

Abstract

Abstract : How does the difficulty of a task affect people's enjoyment and engagement? Intrinsic motivation and flow theories posit a 'goldilocks' optimum where task difficulty matches performer skill, yet current work is confounded by questionable measurement practices and lacks scalable methods to manipulate objective difficulty-skill ratios. We developed a two-player tactical game test suite with an artificial intelligence (AI)-controlled opponent that uses a variant of the Monte Carlo Tree Search algorithm to precisely manipulate difficulty-skill ratios. A pre-registered study ( n = 311) showed that our AI produced targeted difficulty-skill ratios without participants noticing the manipulation, yet different ratios had no significant impact on enjoyment or engagement. This indicates that difficulty-skill balance does not always affect engagement and enjoyment, but that games with AI-controlled difficulty provide a useful paradigm for rigorous future work on this issue.