The inevitability of AI dominance

     The issue of ethics in the development of computer science and machine learning is not new and is perhaps one of the most researched areas in the field of ethics in recent years. This is due to the sheer level of advancement that AI has seen in the past two decades. Johnson's paper provides a scathing account of the rather naive notion that our society, inherently rife with bias and bigotry, could create computer programs that are inherently free of bias. Yet it begs the question, what is the alternative? 

    The development of AI has advanced to such a degree, that it has become nearly impossible to artificially correct biases in systems. During the 90s and early 2000s, machine algorithms and data sets were primitive enough that we can artificially play with datasets (such as the weighting of certain data, etc.) to correct for the outcome. This is no longer possible in the status quo due to the prevalence of the internet. When algorithms are constantly absorbing data from the internet, spitting out values, and learning from them, it becomes near impossible to correct any biases artificially.  

    However, not only is this inevitable, one could even make the argument that this is better than the alternative. What an algorithm does is that it seeks to produce the most correct result with the data it is presented. This is not unlike what a human (such as a judge) does - we try to make the best decision with the data that are presented. The difference between us and AI is that while we can only process a finite amount of stimulus, the AI's capacity for information processing is only bound by our technology, which improves essentially every year. While it may be true that we and AI both suffer from societal bias, an algorithm's bias is actually capable of being corrected and improved, whereas our biases are static. For example, let's take Johnson's account of the criminal recidivism algorithm, and compare it to the only viable alternative, which is that the decision to grant bail/parole is at the judge's discretion. When we see that the algorithm unfairly punishes African Americans, we can easily attribute that mistake to a misevaluation of the data that fed its systems. This means that we could be proactive, and correct, or at least try to correct such biases. While this is obviously easier said than done, the fact that we could speak volumes. Especially when we compare this to a racist judge. A judge is said to be impartial, but we all know that everybody has biases. This is especially the case with racist judges, who suffer from confirmation bias, stereotype bias, among other objectively unfair metrics that they can use to make life-changing decisions. Whereas we can fix an AI, we cannot fix an entrenched racist mind. Furthermore, to replace a judge is nearly impossible, and as such these individuals would remain in the criminal justice system unless we have extremely solid proof that such judges are acting on racist impulses, which is extremely difficult to prove. 

    As such, I think the takeaway is as follows. Society is biased, which leads to the things we produce, whether that be algorithms or software, to be biased as well. I agree with Johnson that such things must be recognized. What also must be recognized, however, is that the alternative is for a human to decide and act in lieu of the algorithms, and I would argue that the human mind is perhaps the most biased software of them all.