The government has no business in AI
People worry about the ways artificial intelligence will change jobs, education, creativity, and daily lives, according to polling. But they harbor skepticism about the ability of government to regulate it — and rightfully so.
Therefore, it’s particularly concerning when prominent thought leaders and lawmakers crusade for larger roles for the state in AI. For example, Ezra Klein, a futurist thinker and leading voice on the Left, has just proposed a government-developed “public option” AI model. Something like a government-provided or government-controlled AI entity that competes with private providers — AI.gov instead of OpenAI, Anthropic, Google, and so on.
Whatever gripes people may have with current and future AI offerings, the notion of government-run AI poses several major problems.
Government cannot keep pace with innovative, cutting-edge tech companies. Government is pretty good at collecting and distributing data and administering programs under well-defined rules, such as sending Social Security checks and averaging hospital costs, but it’s just not set up to produce commercial services efficiently or quickly. Consider that the IRS recently needed billions of dollars to update its systems, including migrating from a software coding language that is practically extinct in the private market. Healthcare.gov was a colossal failure at launch, despite having years to prepare. How is government supposed to build or improve an extremely sophisticated product, such as AI, or compete with what the rest of the world is producing? A government AI would be obsolete before it was finished.
Also, a public AI would be enormously expensive and inefficient. Building its own AI would be a massive waste of resources.
Data centers are being constructed around the world, and AI models are being coded and trained even more broadly. The scope of the buildout has caused high-end processors to sell at a premium. AI equipment is scarce, and its manufacturers’ stock prices have reached all-time highs because they’re all trying to keep up with demand.
Government would need to recruit from the same pool of highly sought-after software engineers that multitrillion-dollar tech companies are competing over, which means paying prices bid up by the most well-financed companies in human history. Suffice it to say, these engineers are as scarce as the hardware and harder to substitute. And government would have to build its own data centers, deciding on locations based on what is politically acceptable to its voter base, donors, and powerful colleagues.
Some legislators may fight to have the data center in their district (or even fight against it), which means the decision will depend significantly on the legislator’s political power. These same problems exist if government builds the AI or owns the companies that are doing the work, as Sen. Bernie Sanders (I-VT) has recently proposed.
As was seen with the CHIPS Act, government likes to pile on unnecessary and counterproductive requirements based on political pressures. Under future administrations, data centers could be required to be run only by union employees or to be built only with equipment manufactured in the United States. These considerations will likely mean the government data center is going to be more expensive than the private one down the road — and take far longer to build.
Read the full article at Washington Examiner.