Bias in Artificial Intelligence
In this month’s episode of On the Vanguard, we discuss how artificial intelligence mirrors human biases and the importance of approaching this technology responsibly and with a commitment to equity.
The Duality of Deploying AI
When we think of artificial intelligence (AI), oftentimes we imagine bizarre scenes from sci-fi films featuring scary tech looming in the distant future. However, the reality is that artificial intelligence is already here and will only continue to evolve. Whether perceived as a villain or a hero, AI is a powerful tool that we must harness to enhance our collective wellbeing. Failure to do so could lead to grave consequences that disproportionately affect communities of color.
So what exactly is AI?
At its core, artificial intelligence is an innovation that seeks to create computer systems that emulate how the human mind works, with cognitive processes such as learning and problem solving. Algorithms are the backbone of AI systems. In short, they are a set of instructions designed to perform a specific task. Individuals who build algorithms attempt to make them reflect reality — thus algorithms are reflective of the perspectives, biases, and values of the people behind them. Its presence in our daily lives is easy to overlook, from chatbots to hyper-personalized music recommendations and ads.
What is bias in AI?
Bias in AI refers to the preferences and prejudice reflected by the patterns computer systems learn from. Namely, AI exhibits the racial and gender bias of those who create the code. Joy Buolamwini, author of Unmasking AI and founder of the Algorithmic Justice League, coined the term “the coded gaze” to describe the algorithmic bias that can lead to exclusionary and discriminatory practices. This can manifest in a various forms such as automated sexist hiring practices and misclassification by facial recognition algorithms used by major tech companies, all of which most adversely affect women of color. Therefore, bias in artificial intelligence is a pressing and multifaceted issue that deserves significant attention.
How does this bias manifest IRL?
Bias can be introduced at multiple levels. Since AI relies on data, any bias present in the training data can be inherited by the AI system itself. In some of its earliest inceptions AI has been used for policing. For instance, predictive policing involves using algorithms to analyze historical crime data to identify “hotspots”. Some may argue this approach is more cost effective and objective than police officers. However, the lack of transparency of such systems could do more harm than good, reinforcing existing racial biases in the criminal justice system. The Markup and Gizmodo completed an investigation on the efficiency of predictive policing software such as Geopolitica and found an alarming pattern. In Plainfield, NJ, the crime prediction algorithm was correct less than 1% of the time. It consistently targeted low-income, Black and Latino neighborhoods and produced a large massive volume of predictions compared to a much smaller number of crimes. In an episode of VanguardSTEM’s webseries and podcast, On the Vanguard, Dr. Anicca Harriot frames the rising concern best,
“Isn’t it just a new age stop and frisk?”
With such low success rate, it becomes evident that our efforts and funds should prioritize identifying the root causes of the crimes rather than solely focusing on its oftentimes inaccurate location.
Bias in AI can also stem from the design of the algorithm itself. Such is the case for racial bias in the algorithms of health care systems. In an attempt to re-distribute medical resources, a predictive healthcare algorithm discriminated against Black patients. As it is, the average Black patient receives less medical funds than a white patient. By basing the algorithm off of previous treatment costs, for which there is already an existing gap, it further exacerbated healthcare disparities. Even though this bias may stem as a byproduct, if unchecked, it has far-reaching effects that demand our attention.
Algorithms do not exist in a vacuum, they affect individuals and entire communities.
AI isn’t the villain
Certainly, AI is a double edged sword. While there are sectors that hold bias in these software systems, other areas trend towards catalyzing equity. Even within the medical field artificial intelligence holds great potential. The expansive capabilities of AI, such as machine learning can offer unique advantages in biomedical research. With large volumes of data, these iterative computations can detect subtle patterns that facilitate early detection and enhance diagnostic accuracy. Now this doesn’t mean that AI will be left to its own devices, but rather validate and cross-reference information from expert clinicians. Additionally, it can automate administrative tasks, freeing up medical staff to provide better care.
With the help of AI, ensuring high-quality care for those historically marginalized can begin to close the gap in present healthcare disparities.
Education is another domain benefitting from AI as seen in the emergence of the EdTech space. In addition to widened accessibility, these smart technologies will help deliver more personalized education so that students don’t fall behind. While some in the field of education have concerns about academic integrity, AI should be viewed as a tool. There are several AI tools that students can take advantage of that mimic one-on-one tutoring instead of simply spitting out an answer. In this way, AI can increase accessibility to educational resources for those that cannot a private tutor. In addition, AI can be equally powerful to help reduce the burden on teachers by helping create lesson plans and complete administrative tasks. Teachers can reallocate their time and energy to provide more personalized education. Ultimately, AI is meant to guide our thinking, not be a replacement for it. When used with this approach, AI can truly power the way we learn.
Toward Equity in AI
Looking ahead, it is necessary to implement strategies that mitigate and address its inherent biases. This can take many forms such as more diverse datasets and designing algorithms with equity in mind. In addition, encouraging diversity in AI development teams would bring a wide range of perspectives and experiences to the table. Angle Bush, founder of Black Women in Artificial Intelligence, takes this a step further by saying “…we also have to embed power, authority and influence because it is not enough to be at the table if we don’t have power, authority, and influence to raise those questions of ethics.” Governments and organizations are beginning to recognize the need for regulatory measures to address AI bias. Such measures are already taking effect such as the European Union’s AI Act, the world’s first regulatory guidelines on AI. It highlights the priority to make AI systems transparent, non-discriminatory and environmentally conscious. Hopefully, other governments will use this as a model to follow.
In conclusion, AI is here to stay, and how we navigate its development and integration into our lives will shape our future. It’s a tool meant to handle routine tasks so that we can spend more time crafting complex ideas and being creative. By using AI responsibly and ethically, we can harness its full potential for the benefit of all.
Check out our latest podcast to hear more about artificial intelligence and all of its varied applications from social media to universe simulations!
This article was originally published on October 31, 2023 as part of our regular content.
If you enjoy our original content, consider donating to our parent not-for-profit, The SeRCH Foundation, Inc., to help support this work.
