On November 2, 2022, I attended the Google AI event in New York City. One of the topics was AI. When I heard about executives how they added their technology to human values, I realized that the defect of AI models was a two -edged sword. Models can be tweeted to minimize, say, say, but also to implement a particular approach. Governments may demand manipulation to censole unwanted facts and promote propaganda. I imagined it as something in which the dictatorial government like China could work. In the United States, of course. , The Constitution will prevent the government from messing with the results of AI models developed by private companies.
On this Wednesday, the Trump administration issued its AI manifesto, which is a very important issue facing the country and even a remote action plan for humanity. The project usually focuses on improving China in the AI’s supremacy race. But a part of it seems to be compatible with China’s playbook. In the name of truth, the US government now wants AI model to follow the word of Donald Trump.
You will not be intended to be clearly stated in the 28 -page project. Instead, they say, “It is important that this system be made from the ground keeping in mind the freedom of speech and expression, and that the policy of the US government does not interfere with this purpose. We have to make sure that the free speech in the AI era is a flour and federal government.
Until the last sentence, this is all right, which raises the question. And what is the “social engineering agenda” exactly? We find a hint about this in the next paragraph, which instructs the Department of Trade to look at the rules of the Biden AI and “eliminate references to false information, diversity, equity, and joining and climate change.” (The weird upper case as written in the published project.) Recognizing Climate change Is there a social engineering? As far as truth is, one of the facts about the project, the White House says, “LLM will be true and prioritize historical accuracy, scientific investigations and objections.” It sounds good, but it has come up by an administration that limits US history to “developing”, denying climate change, and respects Donald Trump’s claims of being the largest president of the United States as a truthful truth. Meanwhile, just this week, Trump’s true social account published Obama’s AI video in prison.
In a speech about the project in Washington on Wednesday, Trump explained the logic behind the directive: “American people do not want to awaken Marxist Luncy in AI models.” Then he signed an executive order titled “Preventing Wokes in the Federal Government”. Although saying that “the federal government should hesitate to manage the functionality of the AI model in the private market, it has announced that” in the context of the federal purchase, it is the responsibility of not buying the models that sacrifice the ideological agendas of truth and accuracy. ” Since all major AI companies are courting government contracts, this order is a backdoor attempt to generally reveal the fault to translate the history of the LLM White House, sexual identity and other hot button. If there is any doubt about the violation of the government, the order spends on making AI a devil in several paragraphs that support diversity, call for racial prejudice, or value gender equality. Pogo Alert-Trump’s Executive Order Top Downward Banning theoretical prejudice is a clear practice in the topdown theoretical bias.
Marks obsession
It depends on the companies how to handle these demands. This week I spoke to an open engineer working on model behavior, who told me that the company was already trying to neutralize. In a technical sense, he said that meeting government standards such as being opposed to Woky should not be a major obstacle. But this is not a technical dispute: it is constitutional. If companies like Anthropic, Openi, or Google decide to try to minimize ethnic bias in their LLMs, or decide to make a conscious choice to reflect the risks of climate change to the reaction of models, the first amendment protects these decisions because the AI is the use of a “freedom of speech”. A government mandate that denies government contracts to companies using this right is the essence of interference.
You think AI -making companies will fight for their constitutional rights on this issue. But so far, no major tech company has publicly objected to the Trump administration’s plan. Google celebrates the White House support for its pet problems such as promoting infrastructure. Anthropic posted a positive blog post about the project, though he complained about abandoning the strong White House’s sudden export controls earlier this month. Open says it is already nearing the pursuit of the object. There is nothing about expressing their freedom of expression.
In the action
Reticence is understandable, because, overall, AI Action Plan is a bona for AI companies. While the Biden administration ordered the Big Tech scrutiny, Trump’s plan is a major fat green light for the industry, which is a partner in the national struggle to defeat China. At the time of the construction of large -scale data centers, it allows AI powers to blow past environmental objections. It has promised support for AI research, which will flow into the private sector. There is even a clause that restricts some federal funds for states that try to regulate AI. This is a consolation reward for the failed part of the recent budget bill, which may have banned the state rules for a decade.
For the rest of us, though, the “anti -vock” order does not end so easily. AI is fast the medium through which we get our news and information. One of the basic principles of the United States has been the freedom of such channels from government intervention. We have seen how the current administration has apparently compromised the companies of media giants like CBS in favor of corporate goals on their journalistic principles. It is not irrational to extend this “Vok” agenda to AI models, to expect a similar housing. Senator Edward Marki has written directly to the alphabet, Anthropic, Openi, Microsoft, and Meta CEO, urging them to request this order to fight. He writes, “Details and implementation plans for this executive order are unclear, but this will create significant financial privileges for larger tech companies … to ensure that their AI chat boats do not make speeches that will disturb the Trump administration.” In a statement to me, he said, “Republican chats want to use government power to give a voice to GPT Fox and friend.
As you may be suspected, the White House team has not shared this theory working on the AI plan. They believe that their purpose is real neutrality, and that taxpayers should not pay AI models that do not reflect a neutral truth. In fact, the project itself indicates a finger toward China, for example, when the truth is manipulated, what happens. It has directed the government to review the Frontier models of the People’s Republic of China to “align the conversation with the Chinese Communist Party with the points and censorship.” Unless AI’s Corporate Overlords will find some spinal cord, future assessment American Frontier models can well display lockstip alignment with White House Talking Points and Censorship. But you may not know by inquiring from the AI model. Very awake
This is an edition Steven Levy’s Back Channel Newsletter. Read the previous coverage from Steven Levy.
