On AI and Expectations: Does AI have more reasonable expectations? And is discrimination reasonably unexpected? - Zac

 In Chapter 7 Crime Shelby asks Is crime unreasonable? AI and Expectations provide two avenues for exploring this question and Shelby's discussion of it.

Some rhetoric around AI use in criminal justice advocates for solely algorithmic decisionmaking removing the judge or human elements from criminal decision-making. A recent speaker of the Ath supported this view, and identified 12 factors that ought be used in criminal guilty verdicts and recidivism estimations including whether the suspect is Married, Age, etc. In this role AI seems to magnify proxy discrimination on a massive scale by enabling discrimination across minute 3rd factors. Perhaps more concerning for Shelby is AI from the utilitarian computer scientist's lens completely side steps the quesiton about whether crime is unreasonable. For them crime causes suffering, so they ought to reduce crime, and thus design and AI algorithm that has the highest prediction accuracy for reducing it with no relevance to justice or the reasonability of crimes committed. 

I fear this utilitarian computer scientist is all too happy to agree with Shelby's diagnosis, and propose these algorithms as the solution to ensuring fairness in decisionmaking by removing the human element:

“As we human beings are imperfect, no social arrangement can ensure that citizens will face no injustices. Yet no citizen can be ex[1]pected to tolerate serious, burdensome, and repeated injustices over the course of their lives.”

I fear the rise of AI black box decisionmaking in criminal decisionmaking which makes uninterpretable decisions undermine any possibility of reasonable expectations on what constitutes noncriminal activity or characteristics. Since conviction can be decided by a near infinite number of factors which could not be reasonably expected by humans. It seems to substitute reasonable human expectations for reasonable AI expectations (or unreasonable accurate ones). This is obviously independently problematic, but also means an AI Shelby account begins to sound dystopian. i.e. If an AI develops a life plan based on its expectations for a human, yet its expectations are frustrated not because of the humans lack of effort or ability, but because X other factor, the AI is not being unreasonable when it chooses unlawfal means to attain expectations i.e. coerces the human into a higher paying career path in the way economic conditions/prevailing opportunity structures do and Shelby critiques (see the quote below in 3rd to last paragraph I'm psuedo paraphrasing and modifying). I feel limited by my own creativity in describing the many ways AI computer scientists can use Shelby for their accounts in ways Shelby likely wouldn't intend, the ways it can be used to justify denial of human decisionmaking in favor of AI for more reasonable expectations (or perhaps through influencing and setting human expectations), or used to justify limitless AI criminal activity due to humans violating expectations.

Furthermore, unpacking Shelby's analysis of expectations more generally has some interesting implications. Before deep delving here I will say what can be "reasonably expected" has a variety of meanings, and I'm not using it as a synonym for the reasonable person or ethical actor here. Many people "reasonably expect" things and practices which are unethical and they do not accept as just, as something more along the lines as 'likely to occur', particularly practices highlighted by Shelby.

Specifically, while contemporary society may agree toleration of injustice is difficult, I would argue people do understand toleration of injustice as reasonably expected or even valorized rather then unexpected. This ranges from the obvious bootstraps stories of rags to riches and overcoming wealth disparties via "The American Dream", to the success of minority figures which is celebrated in part due to their persistence through and ability to tolerate repeated injustices. It seems not only do marginalized groups expect to face discrimination in pursuit of goals, but that dominant groups (and perhaps marginalized groups in a different relation) expect some valorous few to emerge as successful in spite of discrimination.

This is problematic for Shelby because if discrimination is considered reasonably accepted (even though it ought not be) it seems poor citizens no longer are granted freedom to violate the law due to a violated expectation/promise.

Furthermore a model of people understanding discrimination as reasonably expected appears to have worse implications then Shelby's reasonably unexpected model for two reasons. First it becomes a way to separate the "able disabled" or the successful class of discriminated against members from their communities as a tool for discrimination against the broader group. Serving as proof if they worked harder or did something different they could be successful i.e. Black Billionaires being weaponized against ghetto poor Black people. Second they appear required to adopt, maintain, and further the false expectation of success as I will further explore with the following quote:

“My point is rather that affl uent cap i tal ist socie ties encourage the expecta[1]tion that, with a reasonable degree of effort, any able- bodied person has a fair chance to live a life of material comfort. So if one develops a life plan based on this expectation yet the expectation is frustrated, not because of one’s lack of effort or ability, but because of inequities in the prevailing opportunity structure, one is not necessarily being unreasonable when one chooses unlawful means to attain the expected standard of living.”

To further explain the distinction of my argument this from Shelby's account, for me the issue isn't that the expectation does not hold up or is frustrated for a specific person. Rather it is that the bad expectation of success exists. Therefore even if the expectation of success does hold up for a marginalized person, as it does in some cases, those cases seem problematic because they are held up, and build and spread the expectation, not okay because they were held. In other cases a reasonable expectation that one will face discrimination clearly does not make a person anymore vindicated to take illegal action for it not being fulfilled then if it were fulfilled (arguably the reverse is the case).

The conclusion being A. I fear Shelby's consideration of discrimination as unexpected prevents him from granting license for reasonable crime or violation of civic obligations to ghetto poor and B. if not fulfilling expectations grants license for reasonable crime or violation of civic obligations, it appears that license can and should be extended to those whose expectations are fulfilled for themselves but not others.