November 2, In 2022, I participated in a Google AI event in New York City. One of the topics is the person in charge AI. When I listen to executives talking about how they align their technology with human values, I realize that the forging of AI models is a double-edged sword. The model can be adjusted to minimize bias, but a specific perspective can also be enforced. Governments can demand manipulation to review unpopular facts and promote advocacy. I think this is something that an authoritarian regime like China might adopt. Of course, in the United States, the constitution will prevent the government from messing with the output of AI models created by private companies.
On Wednesday, the Trump administration issued its AI manifesto, a plan of action for one of the most important issues in the country and even humanity. The program is often focused on beating China in the AI supremacy competition. But some of it seems to be synchronized with the Chinese script. In the name of truth, the U.S. government now wants AI models to abide by Donald Trump’s definition of the term.
In the 28-page plan, you won’t find the intention clear. Instead: “It is crucial that these systems must be constructed with freedom of speech and freedom of expression, and that the U.S. government policies do not interfere with this goal. We must ensure that in the AI era, freedom of speech flourished in the AI era, and that AI objectively sourced by the federal government can reflect the truth, not the social engineering agenda.”
It’s all good until the last sentence raises a question – according to who is the truth? What exactly is the “social engineering agenda”? We get clues about this in the next paragraph, which directs the Department of Commerce to study AI rules in the Biden-er-e era and “eliminate references on misinformation, diversity, equity and inclusion, and climate change.” (As in weird capitals written in published plans.) Acknowledge Climate change Is it a social engineering? As for the facts, the White House said in its case note on the program: “LLMS should be true and prioritize historical accuracy, scientific inquiry, and objectivity.” It sounds good, but it comes from a government that restricts the “uplift” interpretation of American history, denies climate change, and believes that Donald Trump’s claim to be the greatest president of the United States is an objective truth. Meanwhile, just this week, Trump’s Truth Social Account reposted Obama’s AI video in prison.
In a speech touting the program in Washington on Wednesday, Trump explained the logic behind the directive: “The American people don’t want to awaken the madness of Marxism in the AI model,” he said. He then signed an executive order titled “Preventing the Awakening of AI in the Federal Government.” It declared in designation that “the federal government should hesitate to regulate the functionality of AI models in the private market”: “In the context of federal procurement, it has an obligation not to procure models that sacrifice the authenticity and accuracy of the ideological agenda.” As all the large AI companies are proposing, the order appears to be a backdoor effort to ensure that the general LLM shows loyalty to the White House’s explanations for history, gender identity and other hot ass issues. If there are any questions about the government’s violations, the order took several paragraphs of demonized AI to support diversity, declare racial bias or value gender equality. POGO Alert – Trump’s executive order prohibits top-down ideological bias is a blatant exercise in top-down ideological bias.
Marx is crazy
It is up to the company to determine how to deal with these needs. I posted this week about model behavior with an open engineer who told me that the company is already working to be neutral. They say that in a technical sense, meeting government standards such as government standards, such as strike back, should not be a huge obstacle. But this is not a technical dispute: it is a constitutional dispute. If companies like Humans, OpenAI, or Google decide to try to minimize racial bias in LLM, or make conscious choices to ensure that the model’s response reflects the dangers of climate change, the First Amendment may protect these decisions because it exercises the “freedom of speech and expression” touted in the AI action plan. The government authorizes the denial of the government’s contract with the company exercising its rights as the nature of intervention.
You might think that companies that build AI will fight back, citing their constitutional rights on this issue. But so far, no major tech companies have publicly opposed the Trump administration’s plans. Google celebrates White House support for pet issues, such as enhancing infrastructure. Anthropic has published an active blog post about the program, despite complaining about the White House suddenly abandoning strong export controls earlier this month. Openai said it is close to achieving objectivity. There is no freedom of speech about yourself.
action
Silence is understandable because overall, AI action plans are a scam for AI companies. The Biden administration has scrutinized large-scale technologies, but Trump’s plan is a big fat green light for the industry, a partner in the nation’s struggle for China. It allows AI power to essentially blow past environmental objections when building large amounts of data centers. It promises to support AI research that will flow to the private sector. There is even a regulation that restricts some federal funding, and these states try to regulate AI on their own. This is a consolation prize for the failed portion of the recent budget bill, which would ban state regulation for a decade.
But for the rest of us, the “strike back” order is not easy to remove. Artificial intelligence is increasingly becoming the medium for us to obtain news and information. The United States’ establishment principle is the independence of this channel and government intervention. We have seen how the current administration makes the parent companies of media giants like CBS clearly undermine their journalism principles in favor of corporate goals. Expanding this “anti-strike” agenda to AI models, it is not unreasonable to expect similar accommodation. Senator Edward Markey wrote directly to the CEOs of Alphabet, Anthropic, Openai, Microsoft and Meta, urging them to fight the order. “The details and implementation plans of the executive order are not clear, but it will create significant financial incentives for big tech companies … to ensure their AI chatbots don’t give speeches, which will frustrate the Trump administration,” he wrote. In his statement to me, “Republicans want to use the power of the administration to make chatgpt sound like Fox and friends. ”
You may suspect that this view was not shared by the White House team at the work AI program. They believe that their goals are truly neutral and that taxpayers don’t have to pay for AI models that do not reflect unbiased. Indeed, the plan itself points its finger to China to manipulate what happens when truth is manipulated. It directed the government to examine the border model of the People’s Republic of China to determine “consistent with the key points of the conversation and censorship of the Communist Party of China.” Unless the AI corporate overlord has some backbone, it will be evaluated in the future American The border model is likely to reveal key points of conversation with the White House and the locked route of censorship. But you probably won’t find this by querying the AI model. Too awake.
This is a version Steven Levy BackChannel Communication. Read previous reports by Steven Levy here.
