As the playground saying goes, “Sticks and stones may break your bones but words can never hurt you”. But it seems President Donald Trump doesn’t believe that. Quite the opposite, judging by one of his latest executive orders banning federal agencies’ use of a whole raft of words from ‘inclusion’ to ‘hate speech’.
As of St Patrick’s Day, President Trump had issued 132 executive actions, as logged by official and media agencies, including CNN. It’s hard to keep up with them. But it would be a mistake to regard them as unimportant.
Executive order banning words in federal documents
US presidential executive orders are usually issued to help civil servants do their work within federal government and are recorded in the Office of the Federal Register. They are supplemented by a range of other instruments, including memoranda with legal force on the executive branch. Presidents may also issue proclamations, press statements and invitations to the public to comment on a range of envisaged actions, such as on artificial intelligence (AI).
These executive orders cover almost anything the president wants them to. They can reverse previous policies or launch new ones. One of the first executive orders in January was on the Fifa world cup 2026.
Another, with far more potentially dark implications, was on which words could no longer be used in federal documents.
Humpty Dumpty in the White House
In Lewis Carroll’s Alice in Wonderland, Humpty Dumpty says, “When I use a word, it means just what I choose it to mean — neither more nor less”.
Increasingly, it seems that Trump’s latest executive orders are designed to erase values dear to western democracies’ sense of what it means to live in a free and liberal society. The question is why the president should apparently feel so threatened by the words he has ordered to be removed.
Trump’s order No.14151, signed on 20 January 2025,called for the eradication of all diversity, equity/equality and inclusion hiring policies and programmes from the US government.
On the same day, he issued No 14155 withdrawing the US from the UN World Health Organisation, halting all US funding and reallocating US personnel to other tasks. Another, No. 14162 instructed the US ambassador to the UN to give written notification of the USA’s withdrawal from the Paris agreement on tackling climate change — and yet another took on gender.
There are now only two legally recognised genders: male and female. The diversity agenda has been swept aside. Federal agencies may not refer to ‘gender’ and must ensure that personnel records and government-issued documents, like passports, “accurately reflect the holder’s sex”. Those agencies must not themselves engage in the promotion of ‘gender ideology’.
And the list goes on.
Training automated ‘virtual assistants’
Perhaps these changes reflect a wish to have the world think in a particular way aligned with, for example, President Trump’s opinions: opinions to be fed into AI apps. AI systems like virtual assistants are trained on large language models which involves them being fed with and programmed to sift huge quantities of data in order to issue an automated ‘decision’. Such decisions are based on a pre-selected set of common criteria, such as colour, size, frequency – defined by the person or programme itself. This is misleadingly portrayed as machine or AI ‘learning’.
Imagine a scenario where the language model feeding the AI system is full of racial and other biases (such as climate change denial, gender, misogyny) — whether inadvertently or not. Such biases will not only be replicated every time they ‘learn’ from their previous ‘decisions’, get updated or feed into other systems, but so will any errors and biases.
Many people may have encountered a ‘virtual assistant’ replying to a question with the words ‘I do not understand… Try again’. If it/he/she charmingly declines to recognise the concept of ‘error’, how then can a human enforce the legal right to have incorrect information corrected?
There may eventually no longer be any way to bypass automated decisions and consult a human being about an issue. Without a ‘human-in-the-loop’ to review everything — from the code used at the outset to the final output – the chances of harmful ‘decisions’ being made automatically and replicated endlessly grows. This is about far more than there being no-one to ring up, visit to make a complaint or ask for information about you to be erased or corrected.
How AI technology ‘learns’
For example, what if an AI technology owner decides to change the terms of service and alter the privacy policy and settings you had originally agreed to, and even goes so far as to remove privacy protections gradually? You may have no alternative other than to stop using the service altogether if you disagree with (and actually read) the new terms. From 28 March, all users of Alexa+ on an Echo device will have their voice recording data uploaded to Amazon’s cloud — in spite of previous challenges to Amazon’s handling of data.
In more dystopian terms, if the concepts of ‘hate speech’ and ‘discrimination’ don’t appear in any laws anymore, they won’t appear in language models or codes used in AI apps. It could be almost impossible to prove that ‘hate speech’ had been used to denigrate and discriminate against you to your disadvantage because those terms do not exist (and potentially have never existed, if they are excised from the historical record).
Redefining the truth: some valid concerns
Commentators suggest that the president is violating principles of the US constitution, and international law. But, if even our very words are there to be cancelled, redacted or erased, then all that is associated with them is up for grabs.
We must ask the question what if truth is reduced to whatever the executive order President Trump (or his virtual assistant or successor) says it is? With an international AI summit taking place in Washington in June, these are pressing concerns for the present and the future.
And this is the point. This is not about eliminating allegedly offensive words, political correctness and ‘woke’. It is about denying the human rights that generations since 1945 have taken for granted: equality, dignity, inclusion, mutual respect, anti-discrimination and autonomy.
Take harmful content. If words are to be deleted, or meanings redefined, any legal rules designed to protect the general public from people monetising such harmful content will become unenforceable and robbed of all meaning. The probity and truthfulness of fact-checkers sifting social media disinformation and testing claims against evidence will be eroded, and fast.
AI’s official language
Under 1 March 2025’s Order No. 14224, the USA now has one official language, English (that is, American English), yet a majority is predicted to be Spanish-speaking in the near future.
English is already the world’s lingua franca, and the dominant language online. Virtual assistants and generative AI will continue to reflect the Anglosphere, including the best and worst it has to offer. The language we choose to use now will continue to influence how AI performs in future.
So, with the increasing linguistic authoritarianism emanating from the US, it is not altogether daft to ask how we would cope if someone like Red Dwarf’s Kryten ordered Star Trek’s ‘Klingon’ to be the world’s official language. Nor unreasonable to wonder who would be included, entitled, prioritised, or even, excluded from such a system.
Is this denial of factually evidenced truth a smokescreen for treating the world as a personal theme park, where a not-so-unique form of autocracy is the only game in town? Maybe we’re no more than actors now in The Rocky Horror Show, where we are, as the narrator says “…crawling on the planet’s face, some insects called the human race. Lost in time, and lost in space. And meaning”.
