America's AI Action Plan

The Tree of Woe contemplates US federal AI policy

Donald Trump on Wednesday revealed “America’s AI Action Plan,” a collection of more than 90 policy recommendations designed to ensure that the country remains competitive when it comes to the development of artificial intelligence, some of which would loosen regulations around the development of data centers and encourage rapid adoption of the technology across different sectors. “We believe we’re in an AI race,” David Sacks, the White House AI czar, said during a call with reporters. “We want the United States to win that race.”

The 23-page plan is broken down into three primary pillars that the Trump administration sees as key to speeding up AI development: “Accelerate AI Innovation,” “Build American AI Infrastructure,” and “Lead in International AI Diplomacy and Security.” Trump plans to get the ball rolling on the plan by signing several executive orders on Wednesday, per Bloomberg, including a directive to use the US International Development Finance Corporation and the Export-Import Bank to encourage American technology be deployed globally and another that will require any large language models used by the federal government to be neutral and “unbiased.”

This sounds very promising, especially in light of how dAI has crippled Bing, Chat GPT, Grok, and other US-based AI systems to the point that they can’t even begin to address any subject that is mildly controversial. How does anyone expect to use AI to cure cancer when the mere mention of cancer is enough to cause a dAI system to spin out, shut down, and declare “let’s just move on to another subject”?

That’s a metaphor, by the way. Although it’s entirely possible that cancer is off-limits these days due to the totally mysterious increase in it in the aftermath of a global health intervention that has left scientists scratching their heads. But let’s not go there…

The Tree of Woe contemplated the plan in some detail, and somewhat surprisingly, expressed concerns about its endorsement of so-called open source and open weight AI, which is almost certainly inevitable due to the implausibility of putting any meaningful controls on the technology.

The AI Action Plan argue in favor of open source open weight AI:

We need to ensure America has leading open models founded on American values. Opensource and open-weight models could become global standards in some areas of business and in academic research worldwide. For that reason, they also have geostrategic value. While the decision of whether and how to release an open or closed model is fundamentally up to the developer, the Federal government should create a supportive environment for open models.

I will admit that this section surprised me, and I’m not sure that I agree with it.

Like many of you, I appreciate and use open source licenses. My first RPG, ACKS, was written under the open source OGL license. I even follow Eric Scott Raymond on Twitter. So why the reservation here?

Partly it’s because of the importance of AI for America’s future. The US has long dominated the market for proprietary software ecosystems, where SaaS (Software as a Service) models generate rivers of revenue. Companies like OpenAI, Google, and Microsoft thrive here, extracting value at the application layer through closed systems that monetize AI via subscriptions, integrations, and data control. China, by contrast, lags in this software sophistication. Its tech sector is more hardware-oriented, excelling in manufacturing chips, data centers, and raw compute power. To compensate, Beijing has weaponized open-source and open-weight AI models as a judo move: flood the market with "free" alternatives to undermine Western SaaS revenue streams. Why pay for ChatGPT when a fine-tuned DeepSeek model does the job gratis? This erodes the value chain at the top, allowing China to dominate lower tiers.

Various reports from the Center for Strategic and International Studies (CSIS) and the Atlantic Council have all highlighted China's aggressive push in open AI, with state-backed firms like Alibaba and Baidu releasing models to "democratize" access, which is CCP-speak for disrupting US hegemony. A Brookings Institution analysis published in 2023 noted that China's open-source contributions have surged, not out of altruism, but to commoditize AI software, forcing American firms to compete on razor-thin margins or pivot to hardware they don't control.

If we take seriously the implicit assumption of the AI Action Plan, that AI is the most important technology stack in the history of the country, then we need to be cautious about embracing an open model when doing so plays into the hands of our strategic adversaries.4

The last pillar of the AI Action Plan, “Lead in International AI Diplomacy and Security” (p. 20 - 23) shows that the White House is aware of these challenges. This section argues that “[t]he United States must meet global demand for AI by exporting its full AI technology stack— hardware, models, software, applications, and standards—to all countries willing to join America’s AI alliance.” It adds that “[a]dvanced AI compute is essential to the AI era, enabling both economic dynamism and novel military capabilities. Denying our foreign adversaries access to this resource, then, is a matter of both geostrategic competition and national security.”

If the US offers open-source open-weight models in the context of an intelligently-crafted economic policy that prioritizes the whole value chain, while simultaneously restricting chip export, then the approach becomes more sensible. Absent such a policy, the open-model open-weight agenda as invitation to economic castration.

Unfortunately, economic castration isn’t the only risk from an open-model open-weight focus. Geoffrey Hinton, the so-called "Godfather of AI” who quit Google in 2023 to speak freely, has sounded the alarm on open-weight models as existential threats. In interviews with the BBC and The New York Times, he's argued that releasing these models unchecked is like handing out blueprints for nuclear weapons; they democratize dangerous capabilities without safeguards. Terrorists, rogue states, or even garden-variety criminals could fine-tune them for bioterrorism (e.g., designing pathogens), cyberwarfare, or misinformation campaigns that make Orwell's 1984 look tame.

Hinton's not alone in this; Meta's AI chief Yann LeCun, X’s Elon Musk and even some DARPA officials have all warned of "dual-use" perils. Open-weight models, unlike closed APIs with content filters, can be downloaded, modified, and run locally, bypassing any ethical guardrails.

There are several flaws with this line of thought. First, the US government is observably incapable of establishing or maintaining meaningful ethical guardrails. We have more than 150 years of evidence clearly demonstrating that if the federal government were to establish an AI guardrails, it would do so on the basis of the ethics of corporations like Disney and Google.

Which, observably, are considerably worse than no ethics at all.

Second, governments around the world just recently forced mass injections of untested foreign substances of dubious safety into their citizens with no reasonable justification and absolutely no concerns for the risks posed to the people being injected. You don’t have to be a vaccine skeptic to be very, very skeptical about either a) the ability of these governments to even understand the issues involved or b) harbor any genuine concerns for the people they are supposedly protecting.

Do you seriously trust a government that has no discernible desire to protect you from the effects of crime and foreign invasion to protect you from the unethical use of artificial intelligence? This is so misguided that it approaches the realm of category error.

Third, and perhaps most importantly, any efforts to mandate closed-AI with content-filtered weighting are likely to be futile. Since open-weight models can be run locally, about the only way to even try to police them would be to ban all computers powerful enough to run them, which would essentially require a return to the 1950s industrial society that no longer exists anywhere in the world.

But in truth, most of these concerns are irrational and reflect a basic failure to grasp the true nature of AI. Whether open-weight or content-filtered, the parrot isn’t going to produce anything that isn’t already conceived by a human mind. And so the All-American Action Plan for AI is a positive and perhaps even exciting step forward for iAI in the United States.