Are Algorithms Value-Free?

Abstract

As inductive decision-making procedures, the inferences made by machine learning programs are subject to underdetermination by evidence and bear inductive risk. One strategy for overcoming these challenges is guided by a presumption in philosophy of science that inductive inferences can and should be value-free. Applied to machine learning programs, the strategy assumes that the influence of values is restricted to data and decision outcomes, thereby omitting internal value-laden design choice points. In this paper, I apply arguments from feminist philosophy of science to machine learning programs to make the case that the resources required to respond to these inductive challenges render critical aspects of their design constitutively value-laden. I demonstrate these points specifically in the case of recidivism algorithms, arguing that contemporary debates concerning fairness in criminal justice risk-assessment programs are best understood as iterations of traditional arguments from inductive risk and demarcation, and thereby establish the value-laden nature of automated decision-making programs. Finally, in light of these points, I address opportunities for relocating the value-free ideal in machine learning and the limitations that accompany them.

Varieties of Bias

“…In the functional framework of bias, as presented and supported by Johnson (2020a, 2020b, 2023a, 2023b), social bias functions like these other psychological and computational biases. Specifically, social bias is a functionally‐defined mental entity that takes structured mental states as inputs and yields structured mental states as outputs in ways that mimic inductions made on the basis of social‐group membership 12.…”

“…Indeed, we get things wrong exactly in cases where the environment is abnormal, like an image cleverly disguised to exploit our tacit assumptions, leading to visual illusions. 10 In the functional framework of bias, as presented and supported by Johnson (2020aJohnson ( , 2020bJohnson ( , 2023aJohnson ( , 2023b, social bias functions like these other psychological and computational biases. Specifically, social bias is a functionally-defined mental entity that takes structured mental states as inputs and yields structured mental states as outputs in ways that mimic inductions made on the basis of social-group membership.…”

“…Additionally, the functional view lays a path for recognizing how biases emerge out of group‐based inquiry. Johnson (2023a) extends comparisons of biases in algorithms to the adoption of values in scientific inference 38. Drawing on the functional account of bias, these comparisons can be extended further, recognizing that the sources for values in machine learning programs and scientific practice more generally are the same sources that give rise to the need for bias in everyday thought and inquiry.…”

“… Following Johnson (2023a, p. 4), we can consider an inference to be inductive if it is ampliative, i.e., non-deductive. This includes enumerative induction and abduction, i.e., inference to the best explanation.7 For a thorough discussion of the relationship between underdetermination, induction, and bias, seeJohnson (2023a).Johnson uses the terms 'canons', 'values', 'virtues', or 'biases' interchangeably to denote the same natural kind.8 Right image borrowed fromTodd and Oomes (2002), p. 847. For more on the problem of underdetermination in vision, seePalmer (1999), p. 5 ff and Burge (2010), p. 89.…”

“…I discuss the relationship between bias and psychological essentialism further in Section 3.4.31 While empirical evidence directly linking being a man and being dangerous in perception is currently unavailable, existing literature does suggest that this example is empirically plausible(Correll et al, 2002;Cloutier et al, 2014;Eberhardt et al, 2004;Payne, 2001;Trawalter et al, 2008; Wilson et al, 2017, among others). SeeJohnson (2023a), footnote 32 for discussion.32 Johnson (2023a) additionally argues that these differences might likewise be relevant for differential epistemic evaluation. To summarize, cognitive social biases that encode theoretical assumptions about category membership are justified only when those assumptions are true.…”