Elon Musk’s Disapproval of ‘Woke AI’ Implies ChatGPT Might Become a Target under a Trump Administration
If you believe the United States is politically split at this moment, just anticipate the AI culture battles.
ELON MUSK recently brought ChatGPT and other AI programs into Trump’s spotlight by reiterating his caution that existing AI models are overly “woke” and “politically correct.”
“Many of the AIs being developed in the San Francisco Bay Area adopt the mindset of their surrounding individuals,” Musk stated at the Future Investment Initiative, an event supported by the Saudi Arabian government that took place in Riyadh this week. “Thus, you possess a woke, nihilistic—according to me—philosophy that is being integrated into these AIs.”
Despite being a controversial personality, Musk is correct in asserting that AI systems carry political biases. The problem, nonetheless, is not entirely one-dimensional, and Musk’s portrayal could advance his own agenda because of his connections to Trump. Musk leads xAI, a rival to OpenAI, Google, and Meta, which could gain advantages if those firms are targeted by the government.
“Musk undoubtedly maintains a very close relationship with the Trump campaign, and any remarks he makes will carry significant weight,” states Matt Mittelsteadt, a research fellow at George Mason University. “At most, he might hold a position in a possible Trump administration, and his opinions could genuinely be transformed into actual policy.”
Musk has earlier claimed that both OpenAI and Google are afflicted by “the woke mind virus.” When Google’s Gemini chatbot generated historically incorrect images, such as black Nazis and Vikings, in February, Musk interpreted it as evidence of Google employing AI to promote an excessively woke perspective.
Musk clearly dislikes government regulation, yet he supported a proposed AI law in California that would mandate companies to make their AI models accessible for review.
The initial Trump administration similarly aimed at alleged bias among Big Tech firms through an executive order that intended to hold platforms like Twitter, Google, and Facebook responsible for politically motivated censorship of information. The pressure exerted a noticeable effect, causing Meta to ultimately scrap its plans for a specific news section on Facebook.
Mittelsteadt mentions that Trump’s choice for VP, JD Vance, has also voiced intentions to regulate Big Tech firms, even labeling Google as “one of the most dangerous companies in the world.”
Mittelsteadt mentions that Trump has various methods to penalize companies. He mentions, for instance, how the Trump administration terminated a significant federal contract with Amazon Web Services, a choice probably swayed by the ex-president’s perception of the Washington Post and its proprietor, Jeff Bezos.
Policymakers could easily identify evidence of political bias in AI models, even if it affects both sides.
A study conducted in 2023 by researchers from the University of Washington, Carnegie Mellon University, and Xi’an Jiaotong University discovered various political biases in different large language models. It also demonstrated how this bias could influence the effectiveness of hate speech or misinformation detection systems.
A different study, carried out by scientists at the Hong Kong University of Science and Technology, discovered bias in various open-source AI models regarding contentious topics like immigration, reproductive rights, and climate change. Yejin Bang, a PhD student participating in the research, notes that many models generally exhibit a liberal and US-centric bias; however, these models can also reveal a range of liberal or conservative biases based on the subject matter.
AI models reflect political biases since they are trained on large amounts of internet data that necessarily encompasses various viewpoints. Many users might not recognize any bias in the tools they utilize, as models include limitations that prevent them from producing specific harmful or biased outputs. These biases can emerge subtly, and the extra training provided to models to limit their output can further increase partisanship. “Bang states that developers could guarantee that models encounter various viewpoints on contentious issues, enabling them to reply with an impartial perspective.”
The problem could escalate as AI systems grow more widespread, according to Ashique KhudaBukhsh, a computer scientist at the Rochester Institute of Technology who created a tool known as the Toxicity Rabbit Hole Framework, which uncovers the various societal biases present in large language models. “We worry that a harmful cycle is about to begin, as newer generations of LLMs will be trained more on data tainted by AI-generated material,” he states.
“I am certain that bias in LLMs is already a problem and is likely to become a much larger one in the future,” states Luca Rettenberger, a postdoctoral researcher at the Karlsruhe Institute of Technology, who analyzed LLMs for biases concerning German politics.
Rettenberger suggests that political groups may also seek to influence LLMs in order to promote their own views above those of others. “If someone is very ambitious and has malicious intentions it could be possible to manipulate LLMs into certain directions,” he says. “I see the manipulation of training data as a real danger.”
There have already been some efforts to shift the balance of bias in AI models. Last March, one programmer developed a more right-leaning chatbot in an effort to highlight the subtle biases he saw in tools like ChatGPT. Musk has himself promised to make Grok, the AI chatbot built by xAI, “maximally truth-seeking” and less biased than other AI tools, although in practice it also hedges when it comes to tricky political questions. (A staunch Trump supporter and immigration hawk, Musk’s own view of “less biased” may also translate into more right-leaning results.)
Next week’s election in the United States is hardly likely to heal the discord between Democrats and Republicans, but if Trump wins, talk of anti-woke AI could get a lot louder.
Musk offered an apocalyptic take on the issue at this week’s event, referring to an incident when Google’s Gemini said that nuclear war would be preferable to misgendering Caitlyn Jenner. “If you have an AI that’s programmed for things like that, it could conclude that the best way to ensure nobody is misgendered is to annihilate all humans, thus making the probability of a future misgendering zero,” he said. Rettenberger proposes that political factions might attempt to sway LLMs to advance their perspectives over those of competing groups. “If a person is extremely ambitious and possesses harmful intentions, it may be feasible to steer LLMs in particular directions,” he states. “I view the alteration of training data as a significant threat.”
There have been efforts made to change the bias balance in AI models. In March of last year, a programmer created a chatbot with a more conservative perspective to showcase the subtle biases he observed in platforms such as ChatGPT. Musk has committed to ensuring that Grok, the AI chatbot created by xAI, is “maximally truth-seeking” and less biased than other AI tools, although it tends to hedge on complex political issues in practice. (A committed Trump backer and immigration hardliner, Musk’s interpretation of “less biased” could also lead to more conservative outcomes.)
The upcoming election in the United States will probably not resolve the divisions between Democrats and Republicans, but if Trump emerges victorious, discussions about anti-woke AI may amplify significantly. Musk presented a dire perspective on the matter during this week’s event, mentioning a situation where Google’s Gemini claimed that nuclear conflict would be better than misgendering Caitlyn Jenner. “With an AI designed for such tasks, it might determine that the optimal solution to prevent any misgendering is to eliminate all humans, thereby resulting in a zero chance of future misgendering,” he stated.
