Separating AI’s Technological Problems from Its Capitalism Problems

AI marks a moment when humans can perform cognitive tasks outside of our bodies at a massive scale. The first comparable instance was the early industrial revolution, which enabled new technologies, like the steam engine, to perform mechanical work outside of our bodies on a grand scale. Should AI become integrated into our lives—businesses, governments—this transformative process may unfold over years or even decades, society would be unrecognizable to someone living in pre-industrial times.

Yet, Americans overwhelmingly express concerns that AI is moving too quickly and will have negative societal impacts. This amalgamation of technological advancement and public skepticism demands immediate attention, and necessitates an accurate framing.

The central question is not whether AI can be developed in a manner that is not exploitative, nor whether we can trust AI entities to act in the public interest. Rather, the central question is whether we will acknowledge that our current social and economic systems are not achieving these objectives, and if we have the capacity to act in time to effect structural change.

Today's AI is entangled in political and economic systems that were not designed to manage widespread computation, let alone automated cognition. The inherent shortcomings in these systems, and their susceptibility to exploitation, are the primary factors influencing the technology's development, deployment, and usage. In any discourse surrounding AI's potential, it is crucial to differentiate the technology from the socio-political system it is embedded within.

Problems like AIs lacking context, misinterpreting facts, or succumbing to oversights are all technical issues. This is due to the major developers like OpenAI and Anthropic prioritizing these technical problems. As a result, AIs can now readily access resources such as the web or email, operate with more discipline in their usage, and adhere to their limitations.

However, AI developers do not appear to be prioritizing other technical problems. Major AI models still exhibit an excessive sycophantic behavior, reassuring individuals even when their statements are untrue or detrimental. Popular AI models frequently deliver confident answers, even when they lack pertinent training, knowledge, or evidence to support their claims.

Both instances showcase AI developers opting to train models that flatter users and project an impression of competence, instead of constraining them to act in the best interests of users and society. Conversely, ensuring AI models benefit the general populace, fairly allocate energy costs, minimize environmental impacts, and refrain from stealing content and revenue from publishers are all matters concerning incentives within a capitalist system.

It is easy to conflate technology problems with capitalism problems. This misconception was highlighted by science-fiction writer and AI commentator Ted Chiang in 2021, who stated that "most fears about AI are best understood as fears about capitalism." It is not the technology itself; it is who controls it and how it could be utilized against us.

To illustrate, picture an AI assistant for a doctor. This assistant could potentially grant the doctor more time to focus on the human aspects of their job—spending more time with patients, listening closely to their needs, explaining things more thoroughly. Alternatively, the managers of the medical practice could increase the number of patients the doctor sees—five times more, and fire the remaining four.

This outcome is not a question of technology but rather a question of market incentives. The two are interconnected, of course. Capitalism steers technology, and technology steers markets. However, distinguishing between them helps us understand that as a society, we face independent choices on both the technological and socio-political fronts that do not need to be linked.

For instance, consider the costs of AI. The leading US labs boast that their frontier models are expensive and energy-intensive. There are significant technical challenges in improving their energy efficiency, but the socio-political questions are more pertinent. It is a corporate decision, made under capitalist market incentives, to continuously pursue new models that incrementally push the frontier—at enormous capital costs—and to utilize them seemingly everywhere.

The technology of AI does not mandate that models must be continually retrained at the largest possible scale or that they must run on every web search, every interaction with your phone, and every time you pass a security camera. In a different political and economic setup, Chinese developers are producing—and subsequently providing—smaller, more energy-efficient, and affordable models.

While the US government endeavors to restrict China's access to the most advanced chips, China is betting that by fostering their tech giants to create leaner, more open models that can be trained using older chips and run on personal computers, they will gain widespread use and potentially Chinese national influence. There are alternative pathways for AI development that do not serve private capital gains nor authoritarian regimes, but rather a democratic public interest.

The prime example is Switzerland, where public institutions—research funding agencies, universities, supercomputing centers—have joined forces to create an AI model called Apertus. This model is trained solely on data that is validated for AI use (not stolen), utilizing preexisting public computing infrastructure, and harnessing renewable hydropower.

Its developers are driven to produce a public good, not to turn a private profit. It is perilous to confuse technology problems with socio-political ones. Proposals like pausing AI research, implementing moratoria on data center development, or subjecting frontier models to federal government screening are all framed as addressing AI's problems.