# <small>a critical approach<br /> to ai education</small> blooming ai erasmus+ project / athens / 26.05.2026 <small>frédéric clavert, c^<small>2</small>dh, university of luxembourg</small> <img src='/prez/img/2026-05-26_PrezCriticalAIBloomingAI/06.png' width='128px' />

# <small>brave new words</small> <img src='/prez/img/2026-05-26_PrezCriticalAIBloomingAI/00.png' width='50%' /> Note: In April 2023, Sal Khan -- the founder of [Khan Academy](https://en.khanacademy.org/) -- gave a TED talk that was going againts the dominant anxiety that generative AI would turbocharge cheating and erode learning. THis TED talks is entitled "How AI could save (not destroy) education" with the right safeguards, Khan argues that AI can deliver the two things education has long needed but never been able to scale: a personal tutor for every student and a teaching assistant for every teacher. Most of the talk is a demo of Khanmingo, an AI assistant that was then integrated into Khan Academy and was based on GPT-4. As often with any of Khan's public intervention, there's no discussion of any critiques made to edtech: almost nothing about privacy, equity questions. Sal Khan then published a book on this, called "Brave new words". (See: Khan, Salman. Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing). Penguin, 2024.) The choice of the title is quite puzzling for anyone who read Aldous Huxley. Khan develops his argument, sometimes after dialoguing with quite important researcher in the field of AI and education, Ethan Mollick for instance. But still no engagement with any critical AI work.

## <small>how ai will not save nor destroy education</small> <small>(it's not the right question)</small> <img src='/prez/img/2026-05-26_PrezCriticalAIBloomingAI/03.png' width='30%' /> Note: Three years after his TED talk, Sal Khan admitted that Khanmingo, the supposed to be revolutionnary AI agent of Khan Academy, the supposed to be revolutionnary MOOC platform, was a non-event. Students are just not using it. What's the point of talking about this? Basically, we do not want Blooming AI to become, as Khanmingo, a "non-event", Sal Khan's own words. What I'm going to argue here is that non-engaging with Critical AI studies, including in the AI tech sector, could be one of the main problem of such missed experiments as khanmingo. I'll hence will try to speak today about this litterature. It's an important point, as, for instance, the AI EdTech field of research, that hence exists since 1989 -- the *Journal of Artificial Intelligence in Education* was founded that year, it became then after its relaunch the *International Journal of Artificial Intelligence in Education* (Springer, Elsevier since 2026) is full of revolution announcements that never came (journal of the International AIED Society).

## <small>[retracted]</small> <img src='/prez/img/2026-05-26_PrezCriticalAIBloomingAI/01.png' width='50%' /> Note: Integrating critical AI for education is all the more important that many accepted ideas can be all the more false that the methods of some researchers are not at the best level. This meta-study, for instance, was published in one of the Nature's journal. With a good amount of citation, it was explaining that using ChatGPT had a significant positive impact on learning and assessment. Its methodology had raised various suspicions -- and in the end their methodology was particularly weak.

# <small>critical edtech</small> Note: I'll be particularly using Ben Williamson's (Edinburgh) work, here, particularly hist article from 2024 -- *The social life of AI in Education* [@Williamson_2024]. Williamson explores the implementation of AI in education as a complex social, cultural, and political phenomenon.

## <small>technochauvinism</small> Note: 1. Critique of "Technochauvinism" and AI Hype Williamson critiques the widespread assumption that AI will inevitably and positively transform the future of education, the transformative power of AI for student research and writing. (I know, on this point, my colleague in the blooming ai project might not agree with this point, but I honestly do). Rejecting Technochauvinism -- that means basically, in our case, being conscious that AI is not necessarily the solution.

## <small>social and political risks</small> Note: 2. The Social and Political Risks of AI - Bias and Inequality: Technologies like "transformer" models are problematic because they can reproduce significant biases, generate false information, and risk disproportionately harming those at the margins. Furthermore, contemporary AI applications may amplify existing inequalities, injustices, and social divisions. - Human Rights and Law: He points to concerns regarding the risks AI poses to human rights, democracy, and the rule of law. - The Inseparability of Code and Context: Drawing on the work of Dan McQuillan, Williamson highlights that the mathematical calculations in AI code cannot be separated from the social context in which they are applied.

## <small>data, platforms and big tech</small> Note: 3. The Economic Dimension: Data, Platforms, and Big Tech The Role of Big Tech: Companies such as Google, Microsoft, and Amazon are prominent players. For example, Google uses its Classroom platform to launch AI capacities like adaptive learning and automated tutoring. Data as Value: Platforms collect vast amounts of data to generate value through "AI upgrades" and new fee-paying structures. Infrastructure: Amazon acts as a key promoter of AI in education by providing the cloud computing services (Amazon Web Services) that power many edtech platforms.

## <small>different framings of ai in education</small> Note: 4. Three Framings of AI in Education - As a research methodology: Used to better understand learning and impact educational outcomes. - As a source of profit: Driven by the interests of the industry. - As political rhetoric: Utilized to demand educational reforms. Ultimately, Williamson argues that because AI is a "multisector and interdisciplinary site of development," its deployment can lead to significant, and often unanticipated, social and educational implications, including ethical and regulatory problems.

# <small>critical challenges for blooming ai</small> Note: So AI in Education, in the bloomign AI project and framing, is understood as a research methodology: how to better understand learning and impact educational outcomes. Which is of crouse good -- and I would not be here if it had been different. But the debate to have here -- and I think that the blooming AI proposal as we submitted it is already partially answering those questions. But if we want to avoid blooming ai having the same fate as Khanmingo, critical edtech studies are insisting on some elements to take into account, critical challenges that we should keep in mind.

## <small>ethics</small> Note: [@Holmes_2022] - Ethics: There is a growing focus on ethics, specifically regarding human rights, but there is still relatively little research specifically addressing the ethics of AI in education.

## <small>efficacy and evidence</small> Note: Efficacy and Evidence: Much of the research is developer-led and involves small learner groups, limiting generalisability. There is a lack of large-scale, independent, and transferable studies.

## <small>data privacy and security</small> Note: there are technical choices to come, that we will have to do. The use of platforms like Google Classroom creates concerns regarding the collection of student data and the difficulty of knowing how internet-based services process that data.

## <small>environmental impact</small> Note: Environmental Impact: The massive energy consumption required to train large-scale data-driven models (e.g., GPT-3) presents a significant challenge.

# <small>a card to play</small> <img src='/prez/img/2026-05-26_PrezCriticalAIBloomingAI/05.png' width='35%' /> Note: When you read critial edtech litterature, it's hard to see any room for improvement. I think we have in fact a card to play. Let's go back to a recent event, the Canvas incident. Basically, Canvas, a leader in EdTech, in primary and secondary schools but also in universities, was victim of a leak of personal data. Instructure, Canvas mother firm, let the security of Canvas be bad enough to be hacked at the end of April 2026 and hacked again in May. Canvas is one of those learning system that recently opened an AI-based tutor. That's where we have a room for blooming ai. - it's first based on litterature on learning especially the bloom taxonomy, of course, - it can propose a system that would not avoid any such leaks (ensuring 100% cybersecurity is not possible), but that would be distributed instead of being centralised like Canvas and other BigTech solutions, It's an alternate model to the risk of an oligopolistic situation led by AIED private firms that would exert a social control over education in favour of their interests.