Trump’s command to combat bias is just bias

On November 2, 2022, attended the Google AI event in New York City. It was one of the responsible topics of artificial intelligence. When I listened to the executives talking about how their technology was aligned with human values, I realized that the ability to artificial intelligence models were a double -border sword. Models can be modified, for example, to reduce the biases, but also to impose a specific view. Governments can demand the monitoring of unwanted facts and enhance advertising. I imagined this as something that an authoritarian system like China may use. In the United States, of course, the constitution will prevent the government from tampering with the outputs of artificial intelligence models established by private companies.

On Wednesday, the Trump administration issued an artificial intelligence statement, which is long workplan For one of the most important issues facing the country – and even humanity. The plan generally focuses on the best China in the race for the superiority of artificial intelligence. But part of the matter seems to be synchronized with the playing book in China. In the name of the truth, the United States government now wants Amnesty International models to define Donald Trump to that word.

You will not find that intention clearly mentioned in the 28 -page plan. Instead, she says: “It is necessary that these systems be built from a thousand to Z with freedom of expression and expression in mind, and that the US government’s policy does not interfere with this goal. We must guarantee freedom of expression in the era of artificial intelligence and that artificial intelligence that was purchased by the federal government reflects the truth objectively instead of social engineering schedules.”

This is all right until the last sentence, which raises the question – the minute according to whom? What is exactly the “social engineering agenda”? We get the slightest idea of this in the next paragraph, which guides the Ministry of Commerce to consider the bases of Biden AI and “eliminate signals to wrong information, diversity, fairness, inclusion and climate change.” (Large strange letters as written in the published plan.) Acknowledgment Climate Is social engineering? As for the truth, in Fact paper About the plan, the White House says, “LLMS should be honest and give priority to historical accuracy, scientific and objective investigation.” It looks good, but this comes from management It limits American history For interpretations of “raising a level”, climate change denies, and they have to claim Donald Trump about being the largest president of America as an objective fact. At the same time, this week only, the Trump’s social account has republished Obama’s artificial intelligence in prison.

In a speech promoting the plan in Washington on Wednesday, Trump Explanation of logic Behind the guidance: “The American people do not want the Marxist madness in the models of artificial intelligence.” Then he signed Executive Entitled “Prevention of AI Wake up in the federal government.” While determining that “the federal government must be hesitant to organize the functions of artificial intelligence models in the private market,” it declares that “in the context of federal purchases, it is necessary not to buy models that sacrifice honestly and accurately on ideological business schedules.” Since all major artificial intelligence companies transport government contracts, it appears that it is a rear effort to ensure that LLMS in general appears in the interpretation of the White House of history, sexual identity and other hot issues. In the event of any doubt about what the government considers a violation, it spends several paragraphs on the demonization of artificial intelligence that supports diversity or calls for racial bias or gender equality values. Bugo alert-The executive of TROMP, which prohibits ideological bias from top to bottom is a blatant exercise in ideological bias from top to bottom.

Marx madness

It is up to companies to determine how to deal with these demands. I spoke this week to Openai’s engineer who works to behave the model that told me that the company is already striving to get neutral. They said that meeting government standards such as fighting tourists should not be a big obstacle. But this is not a technical conflict: it is a constitutional. If companies such as humans, Openai or Google decide to try to reduce racial prejudice in their LLMS, or make a conscious option to ensure that the responses of models are risk of changing climate, the first amendment protects those decisions as the practice of “freedom of expression and expression” that is described in the AI’s action plan. Government mandate deprives government contracts for companies that exercise this right is the essence of intervention.

You may think that companies that build artificial intelligence will fight, noting their constitutional rights in this issue. But so far no major technical company has publicly objecting to the Trump management plan. Google Celebrate White House support for pet issues, such as enhancing infrastructure. Antarbur has published a positive leaflet for the blog About this plan, although it complained about the sudden abandonment of the White House of strong export controls earlier this month. Openai says it is already close to achieving it Objectivity. Nothing about confirming their freedom of expression.

At work

Conservation is understood because, in general, the artificial intelligence plan is a reward for artificial intelligence companies. While the Biden Administration Extension From Big Tech, Trump’s plan is a large green light for the industry, which he considers a partner in the national struggle to overcome China. It allows artificial intelligence authorities to mainly give the previous environmental objections when building huge data centers. He pledged to support artificial intelligence research that will flow into the private sector. There is even a ruling that limits some federal funds for the states that try to organize artificial intelligence alone. This is a funeral prize for a failed part of the last budget bill, which would have banned the Islamic State for a decade.

As for the rest of us, though, the arrangement of “fighting tourists” is not easy to clean easily. Amnesty International is the way we get our news and information in it increasingly. The founding principle of the United States was the independence of these channels from the government’s intervention. We have seen how the current administration has won the parent companies of the media giants such as CBS to what apparently Compromise they Press principles In favor of the company’s goals. Extension of the “roof control” agenda to artificial intelligence models, it is not unreasonable to expect similar places. Senator Edward Marqui Written directly To the executives of Alphabet, Anthropic, Openai, Microsoft and Meta urge them to fight the demand. “The details of the details and implementation plan for this executive order will not remain unclear, but it will create major financial incentives for major technology companies … to ensure that chat stations are not produced from artificial intelligence that would disturb the Trump administration.” In a statement, he said: “Republicans want to use government authority to make Chatgpt look like Fox & Friends“

As you may doubt this, this opinion is not shared by the White House team, which works on the artificial intelligence plan. They believe that their goal is the real neutrality, and that taxpayers should not be forced to pay artificial intelligence models that do not reflect the non -biased truth. In fact, the same plan indicates a finger in China as an example of what happens when the truth is manipulated. The government is candidate to examine the border models of the People’s Republic of China to determine “compatibility with the Chinese Communist Party’s points and control.” Unless artificial intelligence ovens get some spine, a future evaluation of American The border models may reveal the Lockstep alignment with the White House talk points and control. But you may not find it by inquiring about the artificial intelligence model. I woke up very.

This is an edition of Stephen Levy Backchannel newsletter. Read the previous coverage from Stephen Levy here.