To Avoid a 'Tokenocracy,' Ensure Popular Access to AI Systems

Uneven access to AI is already expanding the chasm that exists between the "haves and have-nots," according to the Washington Post. Rich people in more affluent areas are making more use of AI, while “disadvantaged regions have lagged behind.” If this trend persists, the United States may be in danger of creating a ‘tokenocracy’ in which justice, opportunity, and self-governance flow toward those with the largest AI budgets.

The reporting on the growing gap in AI access across the country corresponds to my own observations during my travels as the director of the AI Innovation and Law Program at the University of Texas School of Law. From talking to students in Morgantown, West Virginia to attending salon dinners in San Francisco, I’ve seen first hand that AI is bifurcating the country into those that use AI and those who bear its costs. Absent immediate intervention by Congress, this divide will worsen and undermine the country’s ability to realize the full potential of AI to increase human flourishing.

The AI abyss—the gulf in the rate of AI use and the intensity of that use—is playing out in real time in key domains. The justice system is one in which Davids frequently lose to Goliaths. Landlords hire attorneys; tenants don’t. Landlords win. Access to representation in removal proceedings reduces the likelihood of being removed by 35 percentage points. Disparate access to and use of AI will soon tip the scales of justice in favor of the AI haves. Big law firms are developing proprietary AI models while pro se litigants are still unsure of whether and how they can use AI to assert and defend their legal interests. AI could theoretically help level the playing field—providing ‘Davids’ facing removal with far more than a slingshot. Stanford Law Professor Julian Nyarko recently found that in a contest between AI-generated responses to complex legal questions and those offered by law professors the AI answers were favored 75 percent of the time by fellow law professors. If Davids or their attorneys had access to such AI, then they might actually stand a chance. But that’s a big if under the status quo.

AI haves will likely also dominate the economy. Fortune 500 companies are first in line for additional tokens—the “the snippets of text by which the output of a large language model is counted.” Despite AI tool providers increasing the costs of such tokens, the biggest firms are ready and willing to shell out millions of dollars so their employees can leverage AI on a consistent basis for more complex tasks. Microsoft estimates that 80 percent of the Fortune 500 use AI agents—tools that require significant token budgets but are able to assist with a wide range of tasks with minimal to no oversight. Meanwhile, the mom-and-pop shop on Main Street will have to wait—perhaps for a long time—for costs to come down and their AI usage to increase.

Similar disparities will pop up in other domains. Americans with doctors leveraging the latest AI will experience better health outcomes. Students at schools with robust AI tools and AI-savvy teachers will see test scores jump. Individuals able to afford the latest AI-equipped cars will avoid the worst crashes and dodge the worst congestion (if they’re driving at all).

It's a public imperative that such a disparity does not become entrenched. Access to tokens is increasingly access to knowledge, to productivity gains, and to an increased ability to self-govern. If that access is not widespread, then we will experience a significant societal flourishing overhang—the gap between the increased quality of life the average American would experience if advanced AI tools were ubiquitous and the average quality of life under the current distribution of advanced AI.

We’ve seen this disparity play out in prior technological waves. Rural Americans waited decades for electricity; they literally lived in the dark because of poor policy choices. Later, rural Americans and Americans in less affluent communities then watched their neighbors gain early access to the future by way of the Internet; lawmakers failed to adequately respond. In fact, millions of Americans are still on the wrong side of the Digital Divide. When COVID-19 hit, those communities lacking affordable access to high-speed Internet experienced significant hardships. (I started a nonprofit, No One Left Offline, to try to help rapidly close such gaps, but bandaid solutions do not close fractures.)

Congress must not repeat those mistakes. The remedy is a federal floor of token access for every American: publicly funded, model-agnostic, and delivered through a competitive market of certified providers.

Consider a hypothetical: Maria, the owner of a taqueria in Laredo. Her landlord's attorney used AI to draft an eviction notice. Her insurance company used AI to deny her claim. Her largest competitor used AI to optimize its supply chain and undercut her prices. Maria has a smartphone and a free-tier chatbot that times out after five exchanges. The existing AI abyss is the difference between staying open and closing down.

The Token Floor changes that calculus. Congress can and should establish an AI Access Fund and direct the Treasury to negotiate bulk purchase agreements with certified AI providers. Maria gets a monthly token allocation she can spend on any certified model—legal help, business planning, translation, whatever she needs. Funding mechanisms include a revenue-based assessment on frontier model providers above a defined threshold or a reallocation of existing Universal Service Fund obligations that already cover broadband access.

Interoperability as a core feature of this policy prevents it from becoming a corporate subsidy. Federal tokens must be redeemable across any certified provider. Maria chooses her model the way she chooses her bank.

OpenAI CEO Sam Altman has proposed something adjacent—"Universal Basic Compute," a monthly slice of GPT-7 for every citizen. The vision deserves exploration, too. However, it misses the mark. A single private company leveraging such control risks causing a tokenocracy of a different sort. Likewise, Meta founder and CEO Mark Zuckerberg, in his new manifesto on “superintelligence,” proposes an auction system to “ensure the benefits” of advanced AI systems are “distributed widely.” But a “dynamic auction system” is not the same thing as a guaranteed resource.

Rural Americans waited decades for electricity. Millions are still on the wrong side of the broadband divide. Maria cannot wait. Congress should not make her.