There are numerous recent blog posts and newspaper articles discussing this and similar issues. For example, an article in The Nation entitled “The Supreme Court Didn’t Fix Racist Jury Selection” outlines and addresses the limitations of Batson v. Kentucky, a 1986 case that ruled on the impermissibility of using race as a reason for peremptory strikes against jurors. If a defendant or defending attorney believes that strikes were made on the basis of race, he or she can make a “Batson” motion, at which point the prosecutor may defend the strikes, putting forward “race neutral” reasons. The judge then determines whether such strikes were permissible or in violation of the ruling. Though the Batson case is a victory against explicit bias, the author argues that “At best, this scheme tolerates implicit bias in jury selection, because it is so easy to develop race-neutral reasons; at worst, the Batson scheme encourages prosecutors to develop race-neutral reasons as a pretext to shroud discriminatory intent.” Indeed, the article describes North Carolina “training materials” that provide non-race based reasons that prosecutors may use to circumvent the ruling, which reminds me of the idea of plausible deniability in that racial decisions will just be veiled by other explanations.  The author of this article does not suggest that jurors should be screened for implicit bias, but offers very vague suggestions for prosecutors, such as being “screened for bias before hiring” and receiving “diversity training”.
While it is unclear what this recommendation for “bias screening” means, one interpretation is that prosecutors, judges, or jurors should take some test of implicit bias, such as the IAT. Some recent meta-analytic evidence is pessimistic about the utility of the IAT in predicting behavior. However, other evidence suggests that, while explicit attitudes have a stronger relationship with behavior, implicit attitudes also have a moderate correlation; Greenwald, Poehlman, Uhlmann, and Banaji found an average IAT-criterion correlation of .274 over 184 independent samples. They also found that the predictive validity of explicit attitudes was sharply reduced in predicting “socially sensitive” topics, which are likely becoming more frequent, and that IAT measures were better predictors of interracial behavior, as is often the case in juries. If we can reasonably conclude that implicit attitudes affect jurors’ behaviors, it seems a reasonable target to reduce unfair sentencing. Though it is also unclear what the article author above means by “diversity training”, he may be referring to the common approach of educating individuals about the existence of bias. In a study of multiple interventions, Lai and colleagues found that implicit attitudes were malleable in the short term, but observed changes were not lasting, at times disappearing after several hours. Explicit attitudes and motivation to be unprejudiced were also unaffected. 
Given these results, I am unconvinced that implicit bias may be a target for jury selection or training. I certainly agree that those with outright bias and prejudice should be removed, but I am not sure what might be done to reduce the effects of implicit bias. While there seems to be reasonable evidence that implicit bias can predict behavior, I would imagine that the courts would be very hesitant about using IAT results. More discouragingly, there seems to be little promise of intervening to reduce implicit bias. However, in consideration of the research above and the historical under-representation of minorities on juries, one potentially feasible target would be to encourage formation of juries that are actually a “fair cross-section” of the community and not overly white. Since I don’t know anything about law, I don’t know if this is actually a possibility; I imagine quotas wouldn’t work, but I wonder if some sort of adverse impact approach could be used to determine if there were courts or prosecutors that were systematically biased in jury selection. If juries were more diverse, I would hope it would discourage or check outright bias in decision-making, if not reduce implicit bias. 
