Are we critical enough of GenAI at 黑料福利网?

4 december 2025

When ChatGPT launched in 2022, amazement at its abilities sparked a tech race, promises of an AI revolution, and hundreds of billions in investments. Recently, however, reports of GenAI impairing students, threatening mental health, spreading 鈥渇ake news,鈥� making us talk 鈥渓ike robots,鈥� and being a looming 鈥渇inancial bubble鈥� have dominated the news.

Critical AI literacy

Such negative reports were nothing new to the participants in the Critical AI Literacy symposium at Radboud University on October 13. Speakers like computational cognitive scientists Prof. Iris van Rooij and Dr. Olivia Guest, and computer scientist Prof. Dagmar Monett, have been critical of AI tools for years, if not decades. Recently, they have spearheaded a growing effort to criticize 鈥渙verhyped鈥� and 鈥渄etrimental鈥� AI technologies like chatbots and LLMs. Following the publication of an open letter to all Dutch universities calling to 鈥渟top the uncritical adoption of AI technologies in academia鈥� and an academic article, this well-attended symposium was another outlet for critical voices.鹿 The meeting鈥檚 critical tone was perfectly captured when, after an opening presentation branding the technology 鈥渄egenerative AI,鈥� an attendee鈥檚 question about responsible use was met with the reply: 鈥淚t cannot be used responsibly at all.鈥�

Given 黑料福利网鈥檚 position in the Brainport region and its current strategy for GenAI adoption in research and education, an outright ban is unlikely anytime soon. However, I argue that important lessons can be learned from the burgeoning field of Critical AI Literacy (CAIL). Its analyses include but go beyond well-known criticisms of GenAI and offer a more fundamental critique with several alarming potential consequences for its use in education and research. I will examine CAIL鈥檚 main critiques, from my perspective as an information literacy and education specialist, and propose ways they could shape our approach to GenAI at 黑料福利网.

Artificial hype

One of CAIL鈥檚 primary critiques of GenAI is that the big tech companies are exaggerating their tools鈥� capabilities and are creating hype. Guest et al. place this behavior in the context of a longer history of AI 鈥渉ypes鈥� and failed promises about AI technology.虏 Since the 1950s, the term 鈥淎rtificial Intelligence鈥� itself has been used as a 鈥渕arketing phrase鈥�, they argue. More recently, floating signifiers like 鈥淏ig Data鈥� and 鈥淎rtificial Neural Network鈥� have been employed similarly by the tech industry as marketing labels to mystify their products, deflect criticism, and thereby consolidate power. Therefore, the authors urge us to 鈥渞emain critical of the vocabulary the technology industry coopts and deploys, and to remain respectful of scientific terminology.鈥�

Another example of such hype marketing is OpenAI's claim that ChatGPT 5.0 is so good that it has PhD-level intelligence (which I have heard repeated uncritically at 黑料福利网).鲁 Given ChatGPT's propensity to hallucinate, create biased outputs, and, most importantly, its inherent inability to think critically or understand a topic, this claim is easily refuted. Fortunately, we hold PhD candidates at 黑料福利网 to higher standards. The lesson here is that we should always be wary of marketing phrases and exaggerated promises, and determine what the tools can actually deliver before implementing them in research and education.

No consensus in educational research on benefits

CAIL鈥檚 proponents stress that GenAI tools are currently being implemented at universities without adequate consideration of alignment with pedagogical goals and ethical standards. Lucy Avraamidou, Professor of Science Education, concluded in her symposium presentation that there is 鈥渘o research evidence that AI supports learning.鈥� While some educational scientists would disagree with this statement, and while there is recent research that points to potential didactic benefits, there is no academic consensus at the moment on the long-term benefits of GenAI tools in education.鈦� There is much ongoing research, including at 黑料福利网, on the use of GenAI in , assessment, and academic writing. But we are far from concluding whether GenAI will improve or hinder students鈥� education in the long run.

An example from a major actor in the Brainport region, ASML, shows that the critics are right to caution against uncritical adoption. At the recent AI Summit Brainport, the company presented its rigorous decision-making process for implementing AI in its 200-million-dollar EUV machines. ASML first evaluates the purpose of an AI tool in any given process and refrains from implementing it when the purpose is unclear. Second, they ensure that an AI tool will empower their experienced engineers, as the aim is to make their human engineers more efficient and reliable, not replacing them. Finally, a key consideration for ASML is whether the implementation will be durable or will compromise the product's reliability and long-term serviceability. Shouldn鈥檛 we apply the same diligence to our decision-making process for integrating GenAI into teaching our precious students as ASML does for using AI in their precious machines?

Besides implementation in education, we must also consider what we teach students and staff about GenAI. According to the CAIL proponents, AI literacy education should first and foremost involve critical AI literacy.鈦� They argue that universities should provide students and staff with the intellectual means to critically examine these tools. They see it as pointless to focus on instructions for using the tools, as there is hardly any skill involved in interacting with GenAI. And indeed, anyone who has used Copilot, Gemini, or ChatGPT will have noticed that these are the most accessible and user-friendly pieces of software ever developed. I therefore share Guest et al.鈥檚 dislike of the term 鈥減rompt engineer鈥�, which wrongfully implies it requires a lot of skill to effectively instruct a GenAI tool.鈦�

Are we doing enough, and is resistance futile?

Aren鈥檛 we already doing a lot at 黑料福利网 to mitigate the dangers of GenAI and implement it successfully? Fortunately, there are indeed many positive developments in this direction: new AI policy frameworks to provide clearer direction and unify GenAI projects, strong steps toward EU AI Act compliance, pilots for GenAI in education, and tests with self-hosted LLMs are all underway. One of the dangers, though, is that by prioritizing speedy implementation out of a desire to meet industry demand and a fear of falling behind, we don鈥檛 take seriously enough the didactic, ethical, and environmental dangers. And while staff and students are warned about the negative aspects of GenAI in our policy papers, guidelines, and trainings, the question is: will we act on them? If not, Guest et al. would accuse us of 鈥渃ritical washing鈥�: paying lip service to the fact that commercial GenAI tools are biased, were trained in violation of copyright laws, rely on exploited workers, and cause environmental harm, while still recommending them.鈦�

But isn鈥檛 resistance futile? GenAI usage is growing rapidly and seems unstoppable. Another often-heard argument is that 鈥渟tudents are going to use it anyway,鈥� so banning is pointless. The latter might be true, but that does not preclude discouragement if the tools prove harmful. Guest et al. compare it to the history of smoking.鈦� In the 1960s, 60% of the Dutch smoked, but after discouragement campaigns, only 18% do today. Similarly, after the launch of the iPhone, there was initially little concern about potential negative side effects. Now, a strong movement seeks to reduce smartphone use, especially in schools, due to its detrimental cognitive and health effects. One recent study suggests that improving AI literacy is already discouraging people from using it, as greater AI literacy decreases receptivity to AI.鈦�

Even without ascribing to all CAIL critiques, there are enough reasons for alarm, I argue. Therefore, before any large-scale application of GenAI tools in education and research, we should heed their advice to rigorously deconstruct marketing phrases and claims, reach a research-based consensus on whether and how these tools benefit learning, and determine whether their use conforms to our code of scientific conduct.

黑料福利网