A License to Compute Would Concentrate Computation and Speech, Stifling Both

Decentralized innovation and institutions can defend against runaway artificial intelligence: Part 4.

Artificial intelligence is a potential saving grace of the United States, which has worked hard to spend, print, and otherwise bumble its way into a dysfunctional malaise. A.I. could substantially boost productivity growth across most industries, generate thousands of new start-ups, and thus reignite the economy, which has been stuck in a low-growth rut for two decades. Robust growth made possible by explosive technology is one way to avoid continued economic and social decline. 

Yet the dominant response of public officials to OpenAI’s ChatGPT, Google’s Bard, and Facebook’s LLaMA, among thousands of new experiments in the A.I. space, is one of horror. Politicians and many industry experts are calling for pauses and the creation of both national and international regulatory bodies. A.I. researcher Eliezer Yudkowsky famously called for tracking the sale of every graphics chip (GPU) and airstrikes to destroy computing farms before they reach “super intelligence.”  

Policy by panic has dominated public action for the last two decades. Some of the challenges were real (terrorism, Covid-19), some imagined (Russian collusion). But in nearly every case, our public institutions reacted with haste and hysteria, disproportion and profligacy, and of course, vastly expanded bureaucratic power. The resulting distractions deflected attention from crucial priorities. 

A.I. may be the latest panic. 

Yes, A.I. will disrupt labor markets and cultural interactions. As a kind of social media on steroids, it is likely to cause similar psychological derangements. The provenance of even larger floods of content will grow more uncertain. A.I. will be politically weaponized. 

Halting A.I. development, licensing A.I. models, or micromanaging A.I. firms, however, are unlikely to mitigate these threats. Yes, these constraints could slow down innovation along some axes, perhaps delaying the corresponding threats for a short while. More likely, however, top-down regulation will tend to empower a few large A.I. giants, including governments themselves. Perhaps especially foreign governments who are unlikely to comply with our A.I. shackles. Entrepreneurial upstarts will struggle to keep up with legal requirements and will depend even more heavily on the colossal data centers of the A.I. oligarchs, thus creating powerful gatekeepers. 

An A.I. oligopoly will be easily captured by political forces, and vice versa. A.I. will be politically weaponized, just like Big Tech in the social media era, or in the way the big three index funds adopted environmental, social, and governance (ESG) investing without ever asking millions of shareholders.

Thousands of productive new ideas will never emerge. Among these suppressed new ideas would be products and protocols aimed at ameliorating A.I.’s very real threats to economic and social wellbeing. In some cases, only powerful A.I. defense will be able to stop runaway A.I. offense, from cybercrime to ornery robots. We’ll enjoy far fewer creative solutions if power is concentrated and capped at the top of the heap. 

Regulatory Metastasis

Beyond direct regulation of narrowly defined A.I. firms, an A.I. regulator could expand its reach to cover vast swaths of the economy and society. 

Consider the recent example of the Cybersecurity and Infrastructure Security Agency (CISA). Presumably, CISA arose to combat threats to our information networks from foreign enemies and malicious hackers. This sounds like a reasonable objective. Quickly, however, CISA defined “infrastructure” in wildly expansive ways, stretching its meaning to the “cognitive infrastructure” of the country. It labeled “misinformation, disinformation, and malinformation” as threats it must stamp out.  

Thus did CISA appoint itself secret censor across the spectrum of our domestic democratic functions, from elections to public health. 

Soon, CISA, along with the FBI and a dozen other federal agencies, was coercing and collaborating with a sprawling web of social media firms, non-profits, and university centers to erase and throttle true speech about Covid-19 and voting. These and other topics demand robust public discussion, even when, or perhaps especially when, the speakers are wrong. The new war on misinformation, however, overturns this basic American ideal. 

Now, consider A.I., which will turbocharge social media and content creation in general. A.I. is already embedded in every smartphone, much of the Web, and many automobiles. 

Even more advanced and generative A.I. will become ubiquitous, the way computing is today. An A.I. regulator might thus quickly metastasize to cover every firm, industry, network, device, and social function. 

The networked economy, in which governments and hyper-scale firms control information and impose gatekeeping rules like never before, may require new legal protections for citizens. Some even suggest we need new civil rights for the digital age. I am sympathetic.

New rights, however, are tricky. They could backfire and paralyze real innovators. The best protection against tyrannical A.I. also happens to be the best chance to maximize A.I.’s economic potential – decentralized entrepreneurship in both technology and the institutions necessary to constructively shape its radical disruptions.