by Shireen Zaman, May 17, 2024
Concerns over the alignment of artificial intelligence, or AI, to human goals, preferences, and values are at an all-time high as AI plays increasingly large roles in daily life and is starting to take up space in fields where all of the decision-making once belonged to humans. Researchers discuss how the often erratic actions and values of humans may even contradict, and expecting an AI to understand why and how to replicate those behaviors is often impossible. Researchers have tried workarounds, such as using neural networks to reward and punish behavior, but that often leads to the machines acting in unexpected and almost comical ways because of how they misconstrued the goal. That is why many have turned to focusing AI on interpreting the goals and intentions of humans rather than maximizing output for a certain objective, known as inverse reinforcement learning, or IRL (Mitchell, 2022). However, the question remains to be answered whether AI’s development into the decision-making sphere where humans would have been in control can develop into something ethical with IRL, or that can not be entirely true inherently because it was developed as a tool and is embedded with the values of its developers. That is why ethics within development is incredibly important, because those ethics will be translated into the AI’s neural networks and deep learning models.
AI is a disruptive technology, meaning it has greatly changed the corporate and sociocultural world. While there have been many advances in the medical field and AI algorithms have made huge progress in the world of research, there have also been many concerns over the trajectory of its applications, such as algorithms within social media creating political echo chambers and discrimination in hiring. AI taught to want and be like its human counterparts may also pick up implicit biases, effectively embedding them through training datasets or other instructional methods. Then, once this has made its way into the broader public’s lives, the effects will become untraceable and inevitable, with corporations maximizing profits while claiming AI as “just a tool.”
The ripple effect of AI in the corporate world may be seen in many ways. For example, as mentioned previously, in the hiring process, there are many concerns that an AI algorithm being used to sort through applicants may reflect choices made by companies, but also be shaped by structural norms and practices in the field. These algorithms would be influenced by corporate incentives and regulatory framework, and not necessarily the developer’s choice, so finding the issue’s root would be very difficult. If IRL-oriented AI were to choose applicants with values like diversity and equity in mind, the outcome would be more ethical and fair. Another example would be chatbots used instead of customer service representatives or automatic content generation. If training data contains racist or sexist language, the AI may perpetuate stereotypes if it does not mimic the same kind of language. Narrow datasets, in general, can be very controversial and taxing, where skewed data for facial recognition technology may lead to misidentification or exclusion for populations with certain features, or biased risk assessment algorithms may disproportionately target minorities in sentencing decisions within the criminal justice system.
This all begs the question asked earlier, whether AI can move past this “just a tool” narrative when, in the end, the corporations and governments with the most vested interests in developing AI for marketing (such as social media) or surveillance (such as facial recognition) are largely in control of its deployment and have access to all the information it collects and works with. In the end, it is a tool for these large organizations but has far-reaching effects on the individual. Through this the difference in individual vs structural ethics can be seen because, to these organizations, violations of personal privacy, implicit discrimination, and stereotyping do not outweigh their larger goals/values of maintaining established social order, optimizing profit models, and ensuring the stability and safety of organizations themselves.
Thus, developers are responsible for aligning AI to human interests by ensuring they are trained on large, representative datasets that reflect diversity and are imbued with human complexity. In addition, corporations and policymakers must implement measures that maintain fairness and transparency in AI models by detecting and mitigating biases in AI. Both parties must work together to create AI that is ethical and beneficial to human society and quell fears of a dystopian society where AI becomes destructive.
Works Cited
Mitchell, Melanie. “What Does It Mean to Align AI With Human Values?” Quanta Magazine, 13 Dec. 2022, www.quantamagazine.org/what-does-it-mean-to-align-ai-with-human-values-20221213/.
SBU Brooklogue 