The prevailing computing environment of any particular era shapes what people think is possible and, in turn, what technologies we build.
When the personal computer became mainstream in the 1980s, we saw the invention of spreadsheets and word processors and, with these inventions, the transformation of the workplace. When the web became widely accessible in the late 1990s and early 2000s, the browsers and, eventually, cloud-based apps became ubiquitous, leading to huge shifts in how we entertain ourselves, collaborate with colleagues, and do business.
Today’s computing environment is increasingly shaped by AI and the data that powers it. From recommendation engines to smart email filters to predictive text, AI-powered systems have become ubiquitous in our modern society. The norms around how AI is developed transform what kinds of tools, platforms, and experiences we end up building. As so much of our lives become digital, we will continue to see these norms shape our everyday lives.
For example, one current norm is that companies develop products and services that collect as much data about people as possible, and then use sophisticated models to analyze that data and provide personalized experiences. The results of this norm can sometimes be delightful: Spotify suggests songs we like, and Gmail’s autocomplete feature finishes our sentences. But the results can also be harmful: There is evidence that video recommendation engines like YouTube, which optimize for user engagement, profit by introducing people to increasingly extreme viewpoints. In addition, targeted advertisements on Facebook have been shown to manipulate people and exclude vulnerable communities.
Another computing norm is that companies with access to the most data have a competitive advantage in the AI landscape, incentivizing further data collection. Big tech companies have an outsized advantage over both smaller competitors and the people who use tech. Smaller companies find it almost impossible to access enough data to compete on the personalization or recommendation front, and people are often locked into one platform.
As these two examples illustrate, our current paradigms for building technology limit what we think is possible. What if we radically adjusted these norms in AI development? What would it look like if people had greater control over the data collected about them? What kinds of processes and tools in the AI development pipeline will lead to greater accountability?
If our current computing environment is not working, then we must invent a new one. If people feel like they’re not in control of their own data, we can incentivize companies to build technologies that give people more agency. By changing the rules around how data is collected and stored, we can invite smaller players to participate. By imagining new processes for how technology is developed, we can shape the platforms, tools, and products that strengthen collective well-being.
Changes like these are necessary if we want AI that strengthens – rather than harms – society and communities.