Implicit Bias, Reinforcement Learning, and Scaffolded Moral Cognition

Abstract

Recent data from the cognitive and behavioral sciences suggest that irrelevant features of our environment can often play a role in shaping our morally significant decisions. This isn't always a bad thing. But our inability to suppress or moderate our reflexive reactions can lead us to behave in ways that diverge from our reflectively held ideals, and to pursue worse options while knowing there are better ones available (Spinoza 2002, 320). Nowhere is this clearer than it is where racial biases persist in those who have adopted egalitarian ideals. My primary aim in this paper is to sketch a computational framework of implicit biases, which can explain both their emergence and their stability in the face of egalitarian ideals; I then use this framework to explain why some strategies for intervening on implicit biases are likely to be more successful than others. I argue that there are plausible ways to adjust our goals and manipulate our local environments to moderate the expression of implicit bias in the short-run, but I also maintain that the dynamic nature of learning and valuation, as well as the impact of stress on cognitive processing, necessitate more comprehensive interventions designed to reshape the cognitive niche we inhabit. Put bluntly, the freedom to act on what we know to be best can only arise in a world where our reflexive and reflective attitudes are, by their very nature, already aligned. So if we wish to gain control over our implicit biases, we must intervene on the world to which our attitudes are attuned, and do so in a way that instills anti-racist attitudes.

A Plea for Anti-Anti-Individualism: How Oversimple Psychology Misleads Social Policy

“…However, the centerpiece of the structural 14. See, for example, Anderson (2010;2012a;2012b), Dasgupta (2013), Huebner (2016), Jolls and Sunstein (2006), and Kang and Banaji (2006). prioritist defense of affirmative action as a two-bird stone is that it will change the biases of members of other social groups (Kang & Banaji 2006: 1109-1110, but research increasingly suggests that having women professors has no effect whatsoever on undergraduate men's implicit or explicit stereotypes about math ability and gender.…”

“…See, for example, Huebner (2016) for an approving reference to Dasgupta. In some very general sense, it is obvious that social biases depend on social environments.…”

“…The view comes in more and less explicit forms, and is more likely to rear its head in certain contexts than others. Specifically, structural prioritizers including Anderson (2010;2012a), Haslanger (2015), and Huebner (2016) tend to be much better at moving past mirror and appreciating the complex interplay of structural and psychological factors when they are explaining the perpetuation of injustice than when they are envisaging the pivot toward justice. They seem to be much better at characterizing vicious cycles than virtuous cycles.…”

Implicit attitudes and the social capacity for free will

“…In this section, I will focus on those specific attitudes that are stigmatizing-implicit prejudices. Bryce Huebner discusses the relation between these learning processes and implicit prejudice in detail (Huebner, 2016). Different types of learning processes seem to be involved in the constitution of implicit prejudice.…”

“…Such forms of reinforcement learning are also typically unconscious but are slightly more complex than Pavlovian models of learning. In these more complex forms of associative learning, one learns from the consequences of one's own behavior by means of a much larger variety of subtle rewards or punishments (Huebner, 2016).…”