Is artificial intelligence really accessible to everyone?
Artificial intelligence is making inequality worse nationally and globally, but diverse high-quality data, better access, inclusive practices, and strong regulation can help make algorithms better.
Those Nerdy Girls is your trusted source for all things health. Whether there’s too much information, not enough information, or lots of misinformation, we have you covered. If you’re already a paid subscriber, we greatly appreciate your support. If not, please help us keep the lights on by becoming a paid subscriber today.
Is artificial intelligence really accessible to everyone?
Artificial intelligence is making inequality worse nationally and globally, but diverse high-quality data, better access, inclusive practices, and strong regulation can help make algorithms better.
It seems like artificial intelligence (AI) is everywhere these days and that everyone you know is using it one way or another. Yet if we look closer, not everyone has the same kind of access or benefits from AI use.
Current trends in AI growth actually risk widening income gaps between countries, concentrating wealth among those who have control over AI technology. Put another way, richer countries that have more resources and existing infrastructure, like the United States, are better prepared to benefit from AI. In contrast, poorer countries are at higher risk from downstream disruptions from AI, such as AI tools replacing lower-cost labor in developing nations with weaker social safety nets.
The United Nations has called this phenomenon “the Next Great Divergence”. The numbers are already telling—2 in 3 people use AI tools in some high-income countries, but only 5% do so in low-income ones.
AI also risks worsening inequality within countries as well. Many predictive policing algorithms in the US, for instance, target some neighborhoods more heavily than others, often along racial lines.
There are 4 key factors that contribute to this gap in AI benefits:
📊 Bias encoded in data: AI algorithms are trained on massive amounts of data, much of which is usually publicly generated. That means AI programs often “learn” the many biases present in the world (think: racism, sexism, ableism, ageism). Even if AI engineers do try to curate the data AI algorithms are trained with, they could still miss any implicit bias patterns that AI would be sensitive enough to pick up on. One Cedars-Sinai study found that AI algorithms would provide very different recommendations for African American versus white patient cases with the same psychiatric illness and characteristics, even if race was only implied.
This bias arises because many minority and/or marginalized populations are underrepresented in AI training data, either because they aren’t intentionally included or because there’s a lack of data overall. Large language models (LLMs) like ChatGPT, for example, underperform for many non-English-speaking users because they are mostly trained using English data. (You can read more about LLM chatbots in my previous article here!)
AI algorithms may also use assumptions based on a lack of data that can result in worse outcomes for minority populations. To illustrate, one popular health risk-management algorithm used healthcare spending as a proxy or “stand-in” variable for their actual healthcare needs—the less someone spent on healthcare, the healthier they were. But this didn’t account for people who didn’t spend more money because of issues affording, accessing, or distrusting mainstream healthcare. As a result, the algorithm wrongly categorized Black adults as healthier than they actually were compared to white patients. While AI algorithm designers can adjust their models to compensate for a lack of minority data in the short term, it still risks worsening health inequalities or unintentionally promoting race-based science.
💻 The digital/AI divide: According to Merriam-Webster, the digital divide is “the economic, educational, and social inequalities between those who have computers and online access and those who do not”—basically, the gaps in educational, financial, and environmental resources keeping people from using the Internet, computers, smartphones, and AI.
People who lack digital/AI literacy, or the know-how to use and navigate AI tools, may not be able to reap the same kind of benefits from AI, fully understand its impact on their lives, or become at increased risk of being harmed by AI. For instance, vulnerable populations like older adults or people with learning disabilities are at increased risk of being deceived and scammed by AI tools, partly because they may not fully understand how to recognize AI-generated content or how AI works.
🙅Exclusionary design: The AI workforce and academic research is mostly dominated by young white men. Even if they are trying to design for minority groups like older adults, intended users are often not included in the design process. As the World Health Organization put it in one of their AI policy briefs, “the tendency is to design on behalf of…people instead of with [them].”
⚖️ Lack of regulation: Although AI legislation varies widely throughout the globe, the US largely relies on corporations to self-regulate, with the federal government providing use guidelines but very limited oversight. Much of the legal responsibility for AI use currently falls on users, who either may not fully understand what they consent to or who aren’t fully protected from any AI-related harms.
There’s also a lack of consensus on what AI algorithms should be regulated and how. For example, should a LLM chatbot providing medical advice or claiming to be a therapist be regulated as a medical device? Is personal health information uploaded to an AI algorithm subject to the same privacy laws? These are some questions that many governments unfortunately don’t have the solutions to yet.
So what can we do to fix this?
⭐ We should work to advocate for more diverse AI training data and a more diverse AI workforce.
⭐ We can work to increase our own knowledge about AI and its impact, as well as encourage others to learn alongside us as AI continues to evolve.
⭐ We can advocate for more regulation and cross-collaboration between government, AI companies, and advocacy/academic groups.
⭐ Finally, we can work to help our communities become more digitally/AI literate.
Left as they are, AI tools risk worsening inequalities on local, national, and global scales. We all should make an effort to design AI algorithms well, carefully monitor them, and train them on more inclusive data, for ourselves and for others riding the AI wave.
Stay nerdy, stay well.
Those Nerdy Girls.
Additional Resources:
Ada Lovelace Institute: Vulnerability in the Age of AI
Center for Global Development: Three Reasons Why AI May Widen Global Inequality
TedX: AI is dangerous, but not for the reasons you think (Sasha Luccioni)
ICYMI - Digest of Recent Posts:
Mental Health: Mental Health Awareness Week is May 11-17.
Reminder that we now have an online store filled with amazing nerdy merchandise.
Your purchases will help financially support the science communication mission of Those Nerdy Girls.
P.S. We’d love to see pics of you and your friends and family sporting their new TNG swag. You might even be featured in our shop! Email us your pics at info@dearpandemic.org
Like what you read? Please share it with others!
If you have a question, let us know!
We read every question and use them to inform our upcoming content, though we are unable to respond to each specific question.