Artificial fairness? Trust in algorithmic police decision-making

Abstract

Objectives Test whether (1) people view a policing decision made by an algorithm as more or less trustworthy than when an officer makes the same decision; (2) people who are presented with a specific instance of algorithmic policing have greater or lesser support for the general use of algorithmic policing in general; and (3) people use trust as a heuristic through which to make sense of an unfamiliar technology like algorithmic policing. Methods An online experiment tested whether different decision-making methods, outcomes and scenario types affect judgements about the appropriateness and fairness of decision-making and the general acceptability of police use of this particular technology. Results People see a decision as less fair and less appropriate when an algorithm decides, compared to when an officer decides. Yet, perceptions of fairness and appropriateness were strong predictors of support for police use of algorithms, and being exposed to a successful use of an algorithm was linked, via trust in the decision made, to greater support for police use of algorithms. Conclusions Making decisions solely based on algorithms might damage trust, and the more police rely solely on algorithmic decision-making, the less trusting people may be in decisions. However, mere exposure to the successful use of algorithms seems to enhance the general acceptability of this technology.

‘Channel shift’: Technologically mediated policing and procedural justice

“…Contextual inquiry methods would seem particularly appropriate for understanding experiences and reactions to technologically mediated contact encounters, alongside approaches that explore differences within and among different demographics and user types that make up the 'public' end user. Online survey experiments (Hobson et al, 2021), combined with follow-up interviews, could provide another way explore the challenges and opportunities of public experience of digital systems. Fundamentally, however, it is also crucial to engage with the service designers, strategists and leaders within policing who are currently driving delivery, and within the technology companies and designers who provide the infrastructures underpinning technologically mediated contacts.…”

“…Conversely, studies have also found that people may see algorithmic decisions as less fair and appropriate than police officer decisions (Hobson et al, 2021). Dietvorst et al (2015) use the phrase ‘algorithmic aversion’ to describe a complex set of reactions to AI, which Burton et al (2018) argue include: false expectations that affect responses to algorithmic decision-making (for example, the idea that error is systematic, ‘baked in’ and irreparable); concerns about decision control and a general sense that the decision-maker cannot be considered trustworthy; and an emphasis on the need for human decision-making in contexts marked by uncertainty.…”

AI for decision support: What are possible futures, social impacts, regulatory options, ethical conundrums and agency constellations?

“…sions made by human experts (Araujo et al 2020). Others, however, suggest a preference for decisions made by human actors in a policing context (Hobson et al 2023). Human decisions to accept or reject AI suggestions are furthermore not only contingent on the confidence vested in the system but also on personal self-confidence (Chong et al 2022).…”

Algorithms in the court: does it matter which part of the judicial decision-making is automated?

“…Studies on the perception of procedural fairness when AI is involved (not necessarily in the context of courts; see, e.g., Woodruff et al 2018 ; Lee 2018 ; Lee et al 2019 ; Saxena et al 2019 ) 19 yield some mixed results. Some studies find that algorithms are seen as less fair than humans (Newman et al 2020 ; Hobson et al 2021 ), e.g., because they lack intuition and subjective judgment capabilities. Other studies, however, find that that the difference in perceived procedural fairness of human decision-makers versus algorithmic decision-makers is task-dependent (e.g., Lee 2018 ).…”