Biased societies mean biased AI

By Chyung Eun-ju and Joel Cho

Chyung Eun-ju

Joel Cho

As society makes steady steps into the Fourth Industrial Revolution, artificial intelligence (AI) development has seen a rapid evolution, as more and more companies invest in accelerating AI adoption plans in their businesses.

With the recent announcement from Elon Musk that Tesla is developing a humanoid robot, what would be considered a dystopia just a few years ago is now looking like an imminent reality.

It is undeniable that AI technology is becoming more and more common in our daily lives. From virtual assistants to self-driving cars, AI systems are being incorporated into almost every sector of the market. When adopted successfully, companies have seen various benefits from this technology, assisting them in increasing productivity, efficiency and consequently, profitability.

Although adopting more and more AI technology does have a series of benefits, it is important to ponder the amount of caution that should be applied when incorporating cognitive technology to substitute for human tasks and activities in the context of a diverse and multicultural world.

Looking at AI as a cognitive technology, that is a technology programmed to imitate human capabilities, it is crucial to consider the inherent biases of algorithms programmed with selected data.

According to Olga Russakovsky, an assistant professor in the Department of Computer Science at Princeton University, the root cause of bias in AI is the data. If the data is biased, then the AI will augment such biases, as the corresponding system is programmed to function based on this selective data. Given that AI technology's nature is such, there is a fundamental issue that must be considered as we, as society, incorporate AI more and more into our daily lives: How do we determine that the data accessed by AI will not be discriminatory or unethical?

It is undeniable that modern societies are based on historically rooted values. The data we have today ranges so drastically that AI systems can be programmed, intentionally or not, to propagate outdated biases even further. Examples of AI systems functioning on algorithms that echo prejudices are countless.

A very simple search on Google for the words, “school boy,” will result in images of young boys in regular outfits for school, while conducting the same search for “school girl” will show results of women in a more sexualized context. This example shows exactly how the AI system used by the search engine will collect and process biased data.

In terms of demographics, obtaining more representative data sets can result in better samples, but the process does not guarantee an unbiased AI system. Since AI requires development by human programmers, it is crucial that these are able to recognize their own biases in order to avoid exposing AI systems to subjective data.

Algorithms can amplify the bias in data, so programmers must be very cautious when building such systems, considering not only the technological impacts of their programs, but also the social impacts that the programs will have in our lives.

The book “The Smart Wife,” written by digital sociologists Dr. Yolande Strengers and Dr. Jenny Kennedy, discusses how the male-dominated AI sector developed Siri, Alexa, Google Home and even the Japanese voice assistant Hikari Azuma, to fit into the old-fashioned female stereotype of doing “wifework” for their husbands.

The design of these voice assistants brings back outdated stereotypes leading to advanced technology taking a step backwards in terms of gender equality. Male programmers are most likely to blame for assistants with female persona as women are underrepresented in this sector.

AI makes decisions concerning who gets a kidney transplant, who gets screened for a job and even who gets approved for home loans. The growing reliance on AI calls for new regulations, and so perhaps we will never be fully AI-dependent. These ethical issues prove how human input is still crucial until AI can evolve beyond the human flaws of bias.

Chyung Eun-ju (ejchyung@snu.ac.kr) is studying for a master's degree in marketing at Seoul National University. Her research focuses on digital assets and the metaverse. Joel Cho (joelywcho@gmail.com) is a practicing lawyer specializing in IP and digital law.