Crossing the boundaries of automation—Function allocation and reliability

• Research suggests two boundaries that determine if automation is beneficial.
• 1. Adverse effects of automation are most likely when decision-making is automated.
• 2. Automation less reliable than 70% cannot be considered as helpful.
• We crossed both boundaries at the same time: function allocation (FA) and reliability.
• Reliability determined attention allocation and performance, FA showed no impact.

Two important automation characteristics are crucial when considering human performance consequences of automation support. One characteristic concerns the function allocation (FA) between human and automation. Adverse effects of automation seem to be most likely when the human operator is taken out-of-the-loop from active decision-making, excessing a boundary from information automation to decision automation. The second characteristic is the reliability of automation. Previous research suggests a critical reliability boundary around 70% below which automation support cannot be considered as helpful. This study explored differential effects of crossing both boundaries at the same time. Within a multi-task simulation consisting of a monitoring task and two concurrent tasks, participants were assigned to one of six groups, two manual control groups and four automation-supported groups. Automation support differed with respect to FA (stage 1 vs stage 4 automation) and reliability (68.75% vs 87.5%), both factors varied across the critical boundaries. Results suggest that reliability determines human operators׳ attention allocation and performance. When reliability was below the boundary, participants showed an increased attentional effort and a worse performance compared to fairly reliable support. Against the stated assumptions, FA did not reveal any impact. In combination with previous research this result might indicate that the FA boundary might rather be some kind of “function allocation valley” concerning decision-making automation (stage 3) in which negative consequences for human operators are most likely. Results are discussed in the context of recent automation research.