By Dr. Tawhid CHTIOUI, Founding President of aivancity, the leading institution for AI and data
\n\n\n\nSelected by Keyrus as one of the 25 most influential global figures in the field of AI and data (January 2025)
\nIn early July 2025, Elon Musk sparked yet another controversy in the already sensitive world of artificial intelligence. Grok, the chatbot developed by his company xAI, has just been reprogrammed to adopt a more direct, sometimes sarcastic, and often provocative tone. Its stated goal: to offer a broader “algorithmic freedom of expression” than that of other conversational models, which are deemed too moderate, too cautious, or excessively neutral.\n
\n\n\n\nThis decision is no small matter. It marks a major turning point in how companies view their responsibility regarding systems capable of generating massive amounts of public discourse. Even as major players in the sector are stepping up preventive measures—such as ethical filters, context-based moderation, and independent audits—Musk is openly choosing to scale them back, thereby shifting the line between the safety of speech and freedom of expression.
\n\n\n\n\nBut beyond this technical and symbolic gesture, two crucial questions immediately arise, prompting a debate that goes far beyond the specific case of Grok:
\n\n\n\nBehind the call for this “algorithmic freedom of expression,” isn’t there in reality a deliberate business strategy aimed at appealing to a specific target audience that is receptive to “politically incorrect” rhetoric? Should we fear that artificial intelligence will become, in the future, a new ideological battleground, further fragmenting the public sphere, with each group adopting an AI that reflects its own image?
\n\n\n\nThese questions are not merely technical or even legal; they fundamentally involve an ethical, sociological, and political analysis that is essential to any discussion of the role of AI in our democratic societies. Should a chatbot’s “freedom” be considered according to the same criteria as human freedom? Who is responsible for defining the acceptable limits of algorithmic discourse?
\n\n\n\nBut above all, in this new digital landscape where even machines are assigned ideological stances, one educational question takes center stage: what kind of civic education must we now envision to prepare future generations to live in a world where AI no longer merely responds, but asserts, provokes, and divides?
\n\n\n\nOriginally launched as a conversational AI for the general public, the Grok chatbot—developed by Elon Musk’s company xAI—quickly established itself as a provocative alternative to dominant models such as ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google DeepMind). From the very start, Grok adopted a more informal, direct, and less conventional tone.
\n\n\n\nHowever, in July 2025, Musk and his teams took things a step further. Through a series of targeted technical adjustments, the company explicitly decided to relax the measures typically used to regulate the output of conversational models. Specifically, these changes primarily concern system prompts, which are now less restrictive regarding the types of phrasing or ideas permitted, as well as security filters, which have been reduced in number and intensity. This evolution allows the model to generate responses on sensitive topics, sometimes even “borderline” ones. At the same time, contextual moderation has been reduced to offer discourse deemed more authentic by its designers, at the risk of occasionally divisive expression.
\n\n\n\nAccording to Musk’s public statements, as reported by several specialized media outlets, the ambition behind these choices is clear: Grok is intended to become an AI that openly says “what others dare not say.” This positioning stands in stark contrast to the dominant approaches in the sector, which focus on enhanced moderation—notably through the intensive use of techniques such as “red teaming,” which involves systematically testing models to identify ethical and discursive risks; regulation through supervised human learning; and the integration of preventive ethical filters from the design phase onward.
\n\n\n\nThis new direction has sparked a wide range of reactions among the general public and within the scientific community. Some have shown keen interest, even a strong curiosity, seeing it as a novel way to explore the still-undefined discursive boundaries of artificial intelligence. Conversely, experts in algorithmic ethics have expressed serious concerns, warning of the increased risks of spreading potentially toxic or polarizing discourse.
\n\n\n\nFrom a regulatory standpoint, Musk’s move comes amid a legal landscape that remains unclear in the United States, where the liability of companies that develop generative systems remains ambiguous, particularly under Section 230 of the Communications Decency Act. In Europe, by contrast, the AI Act already establishes a more stringent framework regarding transparency and oversight of content generated by artificial intelligence, which could ultimately pose a major obstacle to an approach as permissive as the one proposed by Grok.
\n\n\n\nThis situation, unprecedented in its scale and implications, inevitably raises the question of how far we can—or should—allow artificial intelligence to express itself freely, and what underlying principles drive this call for algorithmic freedom.
\n\n\n\nUnder the guise of a noble debate about the limits of algorithmic free speech, the initiative launched by Elon Musk with Grok appears, however, to be driven by motives that are less idealistic and more strategic. Behind the claim of this so-called “liberation” lies, in reality, a subtle commercial positioning strategy in the highly competitive market for generative artificial intelligence (Zuboff, 2019).
\n\n\n\nIndeed, as chatbots become ubiquitous in both public and private spheres, distinguishing them from one another becomes increasingly complex. Launching a controversial chatbot—one that openly expresses strong or unconventional opinions—can thus attract media attention and appeal to a niche audience drawn to provocation. The aim here is less to defend an abstract freedom than to exploit an implicit expectation: that of unconventional, even rebellious, algorithmic speech that stands in stark contrast to the cautious uniformity of traditional AI (Morozov, 2014).
\n\n\n\nThis marketing approach, described by some researchers as “techno-populism” (De Blasio & Viviani, 2020), involves capitalizing on the public’s fascination with AI that defies conventional norms, thereby tapping into the growing rejection of “political correctness.” Grok fits squarely into this logic, where provocation becomes a marketing promise in its own right, a way to assert a disruptive brand identity that elicits both support and rejection.
\n\n\n\nBut this strategy raises profound ethical questions. In her book *The Age of Surveillance Capitalism* (2019), Shoshana Zuboff highlights how tech companies use the rhetoric of individual freedom and innovation as a smokescreen to mask what are essentially commercial strategies. Algorithmic freedom, as championed by Musk, is no exception to this analysis: far from constituting a step toward greater discursive pluralism, it is instead part of “attention capitalism,” which monetizes controversy at the expense of a calm and rational public debate (Wu, 2016).
\n\n\n\nCould it be, then, that the freedom Grok claims to champion is nothing more than a mirage—a product tailor-made to attract media attention and satisfy a very specific ideological agenda? Moreover, isn’t there a dangerous shift taking place here: one in which freedom of expression—a fundamental human right historically acquired—is being co-opted as a selling point for automated systems devoid of moral agency and critical consciousness (Floridi & Chiriatti, 2020)?
\n\n\n\nThis strategic shift is far from insignificant. It highlights the growing risk of confusion between human values and commercial values, between algorithmic ethics and economic opportunism. By using freedom of expression as a marketing pitch, Musk isn’t truly liberating an AI: he’s liberating a market, opening up a new economic niche, and perhaps paving the way for an even more troubling ideological fragmentation, driven by custom-built artificial intelligences (Couldry & Mejias, 2019).
\n\n\n\nBeyond the commercial implications raised by Grok’s strategy, a deeper societal shift may be on the horizon. Elon Musk’s decision to create an explicitly misaligned and provocative artificial intelligence could, in fact, usher in a new kind of algorithmic fragmentation, transforming chatbots into true ideological standard-bearers, personalized in the image of their users (Sunstein, 2017).
\n\n\n\nThis development raises a major concern: while Grok currently stands out for its provocative, non-partisan stance, might we see the emergence of AI systems that are openly conservative, progressive, environmentalist, libertarian, or nationalist in the future? In other words, could the AI market fragment along explicit ideological lines, with each cultural or political group adopting an AI as a faithful and reassuring reflection of its preexisting beliefs (Van Dijck et al., 2018)?
\n\n\n\nThis “algorithmic tribalization , ” to borrow the term coined by José van Dijck (2018), is not without consequences for social cohesion and the quality of democratic debate. Cass Sunstein warned several years ago, in his book #Republic (2017), about the risks associated with the creation of “digital echo chambers,” where users are exposed only to discourse that confirms their existing beliefs. When applied to generative artificial intelligence, this phenomenon could prove even more dangerous: far from simply filtering existing content, personalized AI would actively generate partisan discourse, thereby directly contributing to the growing polarization of public opinion.
\n\n\n\nIn this regard, Grok could well mark the beginning of a new kind of culture war, in which each ideological or cultural faction would equip itself with a bespoke artificial intelligence capable of reinforcing—through its apparent technological authority—the preexisting beliefs and biases of its users. Rather than an open and pluralistic dialogue, an automated and fragmented discourse could then prevail, erecting virtual boundaries between ideological communities now solidified by increasingly persuasive algorithms (Pariser, 2011; Noble, 2018).
\n\n\n\nThis prospect raises a fundamental question: Is contemporary society ready to face a world where artificial intelligence is no longer a neutral mediator of knowledge and public debate, but rather an ideologically programmed mouthpiece, amplifying the already well-documented polarization effect on social media? How can we prevent the algorithmic radicalization of opinions, and how can we preserve a shared public sphere where democratic debate remains possible despite the proliferation of divisive AI (Habermas, 2022)?
\n\n\n\nFaced with this tangible threat, regulators, researchers, and educators will need to act quickly. For if artificial intelligence is destined to become the new cultural battleground, the greatest risk is that of an irreversible fragmentation of the democratic debate itself. The question that then arises for all of us becomes unavoidable: do we really want to live in a world where even machines, designed to inform, become first and foremost tools in the service of ideology?
\n\n\n\nGiven the possibility that artificial intelligence could become personalized ideological extensions, education appears more than ever to be a crucial issue. For if, in the future, every individual can surround themselves with machines capable of reinforcing their convictions rather than challenging them, the role of schools as spaces for debate, openness, and intellectual emancipation is profoundly called into question (Biesta, 2013).
\n\n\n\nIn this context, civic education can no longer be limited to simply imparting digital or technical skills. It must now equip individuals to think critically about their own beliefs, question their certainties, and, above all, understand how the algorithmic systems that influence and shape their daily choices actually work (Buckingham, 2019). In other words, education must imperatively develop a new form of critical literacy—algorithmic literacy—capable of deconstructing not only digital content but also the ideological mechanisms conveyed by personalized AI (Livingstone & Helsper, 2007; Pangrazio & Sefton-Green, 2021).
\n\n\n\nIn practical terms, this means incorporating into school curricula a systematic examination of how artificial intelligence shapes our perceptions, beliefs, and social behaviors. This critical education on AI should enable future citizens not only to use emerging technologies intelligently, but also to question their neutrality, deconstruct their biases, and actively prevent the polarizing or radicalizing effects they might generate (Selwyn, 2019).
\n\n\n\nMore broadly, educating citizens in a world where AI is becoming an active ideological force means strengthening their ability to engage with difference, to engage in debate with opposing ideas, and to reject the deceptive comfort of digital echo chambers. Thus, rather than passively enduring the algorithmic polarization of the social world, education must provide concrete tools to maintain a vibrant, pluralistic, and democratic public sphere (Giroux, 2020).
\n\n\n\nBeyond the specific case of Grok, this discussion brings us back to a much broader question that is now central to our collective future: what kind of future do we want for our societies as artificial intelligence becomes widespread? By turning provocative AI into an attractive commercial product and paving the way for potentially devastating algorithmic ideological fragmentation, it is in fact our democratic model—our collective capacity to engage in dialogue, debate, and build society—that is at stake right before our eyes.
\n\n\n\nThis debate is not a technical one; it is fundamentally ethical, social, educational, and political. It requires us to take back control over how we design, use, and regulate these emerging technologies. For through AI, it is indeed our own values, our social cohesion, and our democratic future that are reflected.
\n\n\n\nFreedom of expression, a noble and essential concept, must not become a pretext for commercial or ideological abuses. On the contrary, it must remind us that it is ultimately up to us, as human beings, to collectively decide which voices we wish to hear and what kind of digital world we want to inhabit.
\n\n\n\nIn this crucial battle, critical and empowering education remains our best defense against algorithmic radicalization. Faced with the temptation to create AIs in our own image, it is ultimately people who must be prioritized—by educating citizens capable of thinking freely, constantly questioning, and consciously choosing the world they truly desire.
\n\n\n\nFor ultimately, the issue is not merely a question of how far the freedom of expression of artificial intelligence can go. Above all, it is a question of how far we ourselves are willing to go to preserve true freedom: the irreplaceable freedom to think, to doubt, and to imagine our shared future together.
\n\n\n\nIn early July 2025, Elon Musk sparked yet another controversy in the already sensitive world of artificial intelligence. Grok, the chatbot developed by his company xAI, has just been reprogrammed to adopt a more direct, sometimes sarcastic, and often provocative tone. Its stated goal: to offer a broader “algorithmic freedom of expression” than that of other conversational models, which are deemed too moderate, too cautious, or excessively neutral.
\n","protected":false},"author":2,"featured_media":230878,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"content-type":"","footnotes":""},"categories":[28],"tags":[95,92,96,94,91,93],"class_list":{"0":"post-230838","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-articles","8":"tag-chatbot-ia","9":"tag-elon-musk","10":"tag-gouvernance-algorithmique","11":"tag-grok","12":"tag-intelligence-artificielle","13":"tag-xai"},"yoast_head":"\n