Is AI doomed to bigotry?

The question of morality and ethics has always been intertwined with technological development.

A good case in point is science fiction works, as they are usually filled with highly intelligent robotic beings that perform dubiously moral actions. These books and movies usually depict AI as sentient life forms that have learned the ways of evil. In our current world, we do not possess such advanced artificial intelligence, but the question of AI bias is still relevant due to its developers. Can humans truly code something completely neutral, or is it bound to be marked by its creator’s ideals and views?

What is meant by biased AI?

Complex algorithms do not grow on trees. They are carefully produced systems that “make decisions based on training data, which can include biased human decisions or reflect historical or social inequities, even if sensitive variables such as gender, race, or sexual orientation are removed” (Manyika et al., 2022). Biases can come in many different shapes and forms, but they can be particularly harmful to discriminated groups. In 2016, Amazon had to shut down its recruiting AI tool; it was actively rejecting women for software engineering and other IT-related jobs because of their gender. This was due to it having been trained to choose top applicants based on people already hired by the company over a 10-year period, most of whom were men (Dastin, 2018). Other similar instances include AIs performing racial profiling (Larson et al., 2016), underestimating black patients’ needs in healthcare (Obermeyer et al., 2019), and spewing racist ideologies on Twitter (Victor, 2016).

So is AI always biased?

As previously established, the reason for biases in AI is not because they’ve grown sentient and now exhibit bad behavior, but rather because their behavior is a direct reflection of the data being fed into them by their human creators. Despite being a pattern of mathematical calculations, machine learning is still dictated by what humans “consider suitable, (…) which information is relevant, and (…) the outcomes they consider best—ethically, legally, and, of course, financially” (Fitter & Hunt). Everything from the original designers of the model to the collectors of the data pool will have an effect on how the final program learns and adapts. It is, therefore, not the machine itself that is to blame (for it has no opinions of its own), but the responsibility of creating good models lies in the hands of data scientists. “The key question to ask is, ‘Is my model biased?’, because the answer will always be yes” (McKenna, 2019). 

A better question is, “Can it reduce human bias?”

Misunderstandings as well as expectations of fault-free algorithmic predictions could be harmful, as they “can result in unnecessary AI fear in society, exacerbate the enduring inequities and disparities in access to and sharing the benefits of AI applications, and waste social capital invested in AI research” (Zhai & Krajcik, 2023). There is research that suggests that machine learning can be used to minimize discrimination. Fitter and Hunt (n.d.) create an example of how a business can create a model that analyzes gender-biased language and adjusts it to encourage a more diverse pool of applicants (Fitter & Hunt, n.d.). It can also be argued that since AI itself isn’t biased, it can expose already-existing biases that were previously implicit. To further exemplify how recruitment bias could be solved with AI, there are already existing tools such as Knockri, Ceridian, and Gapjumpers that “remove or ignore characteristics that identify gender, national origin, skin color, and age, so that hiring managers can focus purely on candidates’ qualifications and experience” (Gow, 2022). It would, therefore, not be wise to completely discourage research in this field because computers’ blindness to human differences can be utilized with the correct datasets and models.

In conclusion, the moral dilemma of technology is as muddy as one might imagine.

What really constitutes AI neutrality is hard to know. While an algorithm does not perceive the world, it is still shaped by it. It is consequently very important to consider the type of data that was used to create an AI tool, as an answer may appear to be factual without actually being factual. Just because a computer doesn’t inherently have opinions, it will never be completely neutral because its creators are humans who can never be completely impartial. Think about that next time you ask ChatGPT or any other AI about your homework. Who are you really asking?