The Social Structures That Shape AI

Nearly three out of four businesses already use artificial intelligence, and the AI market is projected to surge to $1.4 trillion by 2030. Amidst this new tech boom, Assistant Professor of Sociology Benjamin Shestakofsky thinks it’s more important than ever to pause and explore the social systems that shape AI growth.

“Even the definition of AI itself is ambiguous, contested, and ever-changing,” says Shestakofsky, who studies how digital technologies affect employment, organizations, and the economy. “Regardless of what AI can actually do, people’s ideas about what AI is can influence their behavior.”

With questions swirling around how AI will shape society, Shestakofsky and Devika Narayan of the University of Bristol set out to analyze the social systems that affect artificial intelligence. In an article published in The Journal of Applied Behavioral Science, they outline four big sociological themes that help make sense of AI.

1. Follow the Money

Finance is the not-so-invisible hand that shapes technology; in 2022, the U.S. tech industry attracted more than $130 billion in venture capital funds. With the potential to turn relatively small investments into millions or billions, venture capitalists encourage technology start-ups—including those that create AI—to take big risks and grow quickly in the name of profit, Shestakofsky says. “These funding structures are setting the agenda for technology development and the goals it’s aimed at achieving,” he notes, adding that investors often care more about the skyrocketing short-term valuations of AI startups than whether these companies will be sustainable in the long run.

Shestakofsky says he believes this environment can make technology development snowball, something apparent in the past few years with the rapid release and improvement of generative AI chatbots like ChatGPT.

2. Are We Playing Monopoly?

It’s no secret that the biggest tech giants in the country have consolidated their power to near-monopoly status, and much new research and development of artificial intelligence filters through these companies, Shestakofsky says. Whether it’s Microsoft partnering with OpenAI, the creator of ChatGPT, or smaller AI start-ups using Amazon’s servers, much of AI flows through these goliath companies.

But big tech companies aren’t the end all, be all of AI, either, he adds. Shestakofsky says innovation in AI is often driven by new, smaller companies, many of which build apps designed to run on tech giants’ popular platforms, like iOS and Android. In this way, he believes it’s important to understand how AI development is a push and pull between industry titans and scrappy start-ups, both of which rely on each other to prosper.

3. Founders Versus Followers

Whenever new technology is developed, entire industries adapt around it. Take the example of Netflix that Shestakofsky and Narayan include in the paper; when Netflix revolutionized media by introducing streaming, its rivals had to follow suit, and now nearly every media company from HBO to Disney has its own streaming platform.

While it’s still early days for AI proliferation, Shestakofsky says it’s likely AI development will push many industries to adopt new practices. He points to one already happening in India: Companies that previously specialized in maintaining clients’ in-house IT systems must adapt as cloud computing and digital platforms become the new norm, leading to an “unbundling” of technology infrastructure.

4. It’s Not All Glamorous

Thinking about the AI industry often conjures luxurious Silicon Valley offices, job perks like free meals and exercise classes, and sky-high salaries. But that image excludes the low-paid workers who make AI possible.

These workers—often located in places like India and the Philippines, the researchers note—are responsible for tasks including labeling data, testing and evaluating models, and screening out harmful content. “AI development doesn’t happen without these armies of low-status workers who are out of view and out of mind for the consumers using these systems,” Shestakofsky says.

Even the definition of AI itself is ambiguous, contested, and ever-changing. Regardless of what AI can actually do, people’s ideas about what AI is can influence their behavior.

Many of these workers are now responsible for a task called red teaming, in which they prompt generative AI systems to create harmful or offensive content to identify and patch holes in their moderation systems. This means these workers may get exposed to disturbing, violent, or pornographic imagery and content. “This is another emerging frontier where we don’t yet fully understand the consequences for workers’ health and well-being,” he says.

There’s still a lot to learn about AI, but the researchers say they hope these lenses can help demystify the new technology and situate it in the broader context of existing social, cultural, and economic systems.

“As a sociologist, the name of the game is always to understand what you’re studying in relation to broader social structures,” Shestakofsky says. “So, there may be a lot of incentives to ‘move fast and break things,’ but I think organizations can benefit from paying close attention to some of the second-order effects associated with introducing new technologies.”