In 2016, the first international beauty contest judged by artificial intelligence was held. Contestants from about 100 countries submitted photos for the AI robots to judge. Sadly, all but six of the 44 winners across all the age groups were white (Benjamin, 34). These unfortunate results indicated that racism was potential during AI processing, challenging the assumption that technology is color-blind and nondiscriminatory.
As a society, we can learn from this event that just because robots are making judgments does not mean they are fair. The bias of the programmer can rear its ugly head when it comes to determining arbitrary traits like beauty, which appeared to be the case in the beauty contest. Because a disproportionate number of AI programmers are white, the potential for AI discrimination against people of color, or any other demographic, is a valid concern for civil rights activists in the coming decades.
Biased programming is just one of many ways a robot can engage in racist behavior. The data sets from which AI learn or update their algorithms can also be racist. These include social credit systems that factor race and ethnicity into their scores (Benjamin, 41:2019). Even when race is not a factor in these scores, I believe it would still lead to racism because members of certain races are likely to receive lower scores on average if the algorithm includes unfair environmental factors, like zip code and lifestyle. Since minorities are more likely to live in certain regions and engage in different activities, a social credit system could discriminate against them simply because of their geography. This has a net effect of decreasing liberty for some races. Preventing someone from getting a loan or an education based on a social credit score that hinges on race, or the environmental factors that stratify racial systems, is a subtle form of discrimination that should not be taken lightly.
Another way AI can update its algorithm is by copying the algorithm of another AI. If the host AI had racist programming written into its algorithm, it could potentially spread that algorithm to other AI at an alarming rate (Whitaker, 2018). Findings from Whittaker’s (2018) study involve “individuals (robots) updating their prejudice levels by preferentially copying those that gain a higher short-term payoff…”. To me this suggests that if a racially charged algorithm is profitable in some way, it can be copied indefinitely to other AI that will reinforce those profits. Notice the parallel this robotics has to humans when the Atlantic slave trade first appeared in international relations. Because it was immensely profitable for capitalists, the slave trade grew it at an alarming rate, and laws were slow to protect civil rights for the victims of slavery. I fear something similar would happen with this AI, though on a less barbaric scale.
In popular culture, we tend to view AI and big data as the capstone of technological development, one that will lead to a utopia of non-discrimination and “free” labor. Both accounts are false. For the reasons stated above, AI can become partial to demographics like race, gender, and class, among others. The idea that AI is nondiscriminatory because it can’t make real decisions or have intentions is a terrible oversight on civil rights. Racism is not restricted to humans; it can also apply to any systemic or social structure that contributes to discrimination. We see systemic racism in many phases of society, whether it’s exposure to environmental harm, lack of educational opportunities, or susceptibility to criminal justice. “Legal codes, financial practices, and medical care often produce deeply racist outcomes” (Benjamin, 40). AI is no different from these other forms of lifeless racism; in fact, it is much closer to human racism because humans are the ones programming it. Technology companies that violate our current protections must be held accountable by law, and no country should be immune from that.
My intuition tells me this will be an emerging problem in the U.S. and many places where people put profit over protection. Keeping our liberties safe from inequality is a cornerstone of our Constitution, so I predict movements to protect minorities from AI racial profiling will be relatively successful. However, in countries like China, where civil rights are not held in as high regard, I would expect social credit systems and programming bias to influence the population more heavily. This engineered inequity is already happening in parts of the country (Benjamin, 45-46) where Muslims are vulnerable to discrimination. If left unchecked, AI profiling for citizen scores could exacerbate an already abusive government that is becoming more capitalist over time. Many countries do not have the protections that western democracies have for their citizens, providing this technology with a scary number of opportunities to grow.
Perhaps we need to consider outlawing this technology on a global scale, through international law agreements. We must consider the ways in which technology like this does not indicate progress or increase liberty, but a reversion to pre-industrial globalization tactics, when rights were not guaranteed to protect minorities from opportunistic capitalists and bona fide racists. Technology like this would make it easier for a genocide to occur, for if the government is able to track minorities through social credit systems, it would likely be able to monitor the activities of all persons in a social group. To prevent these social reversions, I believe it would be wise of us to cut the head of the snake before it is fully grown.
References:
Benjamin, Ruha. 2019. Race After Technology: Abolitionist Tools for the New Jim Code: Chapter 1, e-book, pp. 33-52. Polity Press. ProQuest eBook Central.
Whitaker, Roger. September 7, 2018. Could AI Robots Develop Prejudice on Their Own? Cardiff University. Retrieved from https://www.cardiff.ac.uk/news/view/1273236-could-ai-robots-develop-prejudice-on-their-own
No comments:
Post a Comment