And in contrast to some of the studies currently coming out, which tend to claim either amazing success or doom laden failure for the use of AI for learning, she adopts an analytical and nuanced viewpoint, examining the evidence and providing a list of key takeaways from each report, leading to implications for educators and developers.
Here are the Key takeaways from each of the five studies.
Surface-Level Gains: Generative AI tools like ChatGPT improve task-specific outcomes and engagement but have limited impact on deeper learning, such as critical thinking and analysis.
Emotional Engagement: While students feel more motivated when using ChatGPT, this does not always translate into better long-term knowledge retention or deeper understanding.
Over-reliance on AI tools hinders foundational learning, especially for beginners.
Advanced learners can better leverage AI tools to enhance skill acquisition.
Using LLMs for explanations (rather than debugging or code generation) appears less detrimental to learning outcomes.
Scaffolding Through Customisation: Iterative feedback and tailored exercises significantly enhance learning outcomes and long-term retention.
Generic AI Risks Dependency: Relying on AI for direct solutions undermines critical problem-solving skills necessary for independent learning.
Offloading Reduces Cognitive Engagement: Delegating tasks to AI tools frees cognitive resources but risks diminishing engagement in complex and analytical thinking.
Age and Experience Mitigate AI Dependence: Older, more experienced users exhibit stronger critical thinking skills and are less affected by cognitive offloading.
Trust Drives Offloading: Increased trust in AI tools encourages over-reliance, further reducing cognitive engagement and critical thinking.
Confidence ≠ Competence: Generative AI fosters overconfidence but fails to build deeper knowledge or skills, potentially leading to long-term stagnation.
Reflection and SRL Are Crucial: Scaffolding and guided SRL strategies are needed to counteract the tendency of AI tools to replace active learning.
As Phil Hardman says in the introduction to her article:
At the same time as the use of generic AI for learning proliferates, more and more researchers raise concerns about about the impact of AI on human learning. The TLDR is that more and more research suggests that generic AI models are not only suboptimal for for human learning — they may actually have an actively detrimental effect on the development of knowledge and skills.
However she remains convinced of "the potential of AI t:o transform education remains huge if we shift toward structured and pedagogically optimised systems."
To unlock AI’s transformative potential, she says, "we must prioritise learning processes over efficiency and outputs. This requires rethinking AI tools through a pedagogy-firstlens, with a focus on fostering deeper learning and critical thinking."
She provides the following examples:
Scaffolding and Guidance: AI tools should guide users through problem-solving rather than providing direct answers. A math tutor, for instance, could ask, “What formula do you think applies here, and why?” before offering hints.
Reflection and Metacognition: Tools should prompt users to critique their reasoning or reflect on challenges encountered during tasks, encouraging self-regulated learning.
Critical Thinking Challenges: AI systems could engage learners with evaluative questions, such as “What might be missing from this summary?”
Its well worth reading the full article. Phil Hardman seems to be of the few writing about AI from a pedagogic starting point.
About the Image
This illustration draws inspiration from Leonardo da Vinci’s masterpiece The Last Supper. It depicts a grand discussion about AI. Instead of the twelve apostles, I replaced them with the twelve Chinese zodiac animals. In Chinese culture, each zodiac symbolizes distinct personality traits. Around the table, they discuss AI, each expressing their views with different attitudes, which you can observe through their facial expressions. The table is draped with a cloth symbolizing the passage of time, and it’s set with computer-related objects. On the wall behind them is a mural made of binary code. In the background, there’s an apple tree symbolizing wisdom, with its intertwining branches representing neural networks. The apples, as the fruits of wisdom, are not on the tree but stem from the discussions of the twelve zodiacs. Behind the tree is a Windows 98 System window, opening to the outside world. Through this piece, I explore the history of AI and computer development. Using the twelve zodiacs, I emphasize the diversity of voices in this conversation. I hope more people will join in shaping the diverse narratives of AI history in the future.
Hanna Barakat & Cambridge Diversity Fund / Better Images of AI / Lovelace GPU / CC-BY 4.0
Last Friday, 24 January, marked the UNESCO International Day of Education. And as part of that, the UNESCO Institute for Lifelong Learning hosted a webinar on ‘Lifelong learning in the age of AI’ aiming “to bring together policymakers, practitioners, and researchers to revisit the idea of lifelong learning in the age of emerging technologies, with a thematic focus on lifelong learning as a concept, workplace learning, digital competencies of adult educators, and bridging the grey digital divide.”
Current policies on AI and lifelong learning, they said, often adopt an instrumental and technologically deterministic approach, prioritising efficiency over human development and agency. UNESCO is committed to supporting Member States to harness the potential of AI technologies for achieving the promise of lifelong learning opportunities for all, while ensuring that its application in the learning contexts is guided by the core principles of inclusion and equity.”
The webinar would discuss current trends, in policy, research, and innovative practices in emerging technologies such as AI and its relation to lifelong learning and the concept of agency.
One of the speakers was Rebecca Eynon, Professor of Education, the Internet and Society, at the University of Oxford with a presentation entitled 'Reconfiguring Lifelong Learning in the Age of AI: Insights from policy and research'. In many ways her presentation was prescient, coming as it did two days before the news of the DeepSeek Open Source model broke.
Rebecca began by questioning what is Al in lifelong learning? Is it an approach or an academic methodology? Motivations of engagement are about researching and facilitating learning (and are often more about knowledge acquisition and are psychological in focus), while remaining cautious about the current hype around Al in Education. They also encompass relations between Al and humans while working with Al. Al is assumed to contribute to increased efficiency of humans and learning and Al is implemented and conceptualized as a peer or colleague. Al is viewed as part of a wider reconfiguration of humans and their contexts.
Artificial Intelligence is currently hailed as a 'solution' to perceived problems in education. Though few sociologists of education would agree with its deterministic claims, this Al solutionist thinking is gaining significant currency.
Rebecca went on to explain research using a relatively novel method for sociology - a knowledge graph - which together with Bourdieusean theory, she said, facilitated a critical examination of how and why different stakeholders in education, educational technology and policy are valorising Al. including their main concepts and motivation.
Drawing on this analysis, she argues that Al is currently being mobilised in education in problematic ways and advocates for more systematic sociological thinking and research to re-orientate the field to account for society's structural conditions. She pointed to the dominance of the commercial sector the prevalence of personalisation. The commercial sector is tending to dominate conversations about Al and education. But the commercial motivations are based on the needs of the market, and promotes an individual view of learning where economic agendas predominate.
There is, Rebecca said, almost an absence of Al policy and specific education actors may well intensify economic and individual notions of education. This has likely implications for what kinds of systems are designed for education . Although this points to an intensification of economic and individual notions of education this is not inevitable. Change is complex, and there is fragility in the ed tech market, with some signs of discontent with Al. She pointed to increasing calls for ethical and equitable AI,.
Rebecca concluded her presentation by pointing to the need to make visible and understand the networks around AI in education and the complex ecology to change them. She said we have to work as a community to demand alternative Al futures for Lifelong Learning.
About the Image
Through distortion, this image depicts a pixelized and reconfigured portrait of Ada Lovelace cast on a microchip. Ada Lovelace was an English mathematician who discovered that a computer could follow a sequence of instructions beyond pure calculation. Her contributions laid the groundwork for programmable computing that underpins algorithms driving AI advancement. As GPU (Graphics Processing Units) microchips are maximizing parallel computing for accelerating tasks like machine learning, the image blending Lovelace’s historical contributions with modern computational technology.
Elise Racine / Better Images of AI / Morning View / CC-BY 4.0
As I write this post, the newspaper headlines are focused on the record decline in value of tech shares, especially chip manufacturer Nvidia, following the release of the Open Source DeepSeek Large Language Model and platform.
Most of the extensive news coverage has focused on the tech business and the likelihood that it represents a bubble, especially for the Generative AI companies, OpenAi, Anthrpopic Google and the like. The other ,main focus has been geopolitical, with China having caught up with the USA in AI development.
For education, DeepSeek can be seen as good news. The domination of Generative AI by large tech companies has priced out public sector EdTech development. Here the big news is that DeepSeek is open source, freely available on the HuggingFace platform. The company is purely focused on research rather than commercial products – the DeepSeek assistant and underlying code can be downloaded for free, while DeepSeek’s models are also cheaper to operate than OpenAI’s o1.
As the Guardian newspaper reports, Dr Andrew Duncan, the director of science & innovation at the UK’s Alan Turing Institute, said the DeepSeek development was “really exciting” because it “democratised access” to advanced AI models by being an open source developer, meaning it makes its models freely available – a path also followed by Mark Zuckerberg’s Meta with its Llama model. “Academia and the private sector will be able to play around and explore with it and use it as a launching,” he said.
Duncan added: “It demonstrates that you can do amazing things with relatively small models and resources. It shows that you can innovate without having the massive resources, say, of OpenAI.” Its interesting to hear the motivation put forward by DeepSeek CEO, for open source. As Alberto Romero has written in his Algorithmic Bridge newsletter, in a rare interview for AnYong Waves, a Chinese media outlet, DeepSeek CEO Liang Wenfeng emphasized innovation as the cornerstone of his ambitious vision:
we believe the most important thing now is to participate in the global innovation wave. For many years, Chinese companies are used to others doing technological innovation, while we focused on application monetization—but this isn’t inevitable. In this wave, our starting point is not to take advantage of the opportunity to make a quick profit, but rather to reach the technical frontier and drive the development of the entire ecosystem.
DeepSeek has developed a completely different approach to the big tech companies who have focused on ever increasing use of hardware to scale Large language Models. And is so doing they have greatly reduced both the financial and the environmental cost of developing such technologies. This may well be a vindication of the policy of prioritising educational spending in China. Romero point out startups and universities can train top AI models and world-class human talent respectively in China. He says “DeepSeek—contrary to Google, OpenAI, and Anthropic—publishes a lot of papers on frontier research, architectures, training regimes, technical decisions, innovative approaches to AI, and even things that didn’t work.”
DeepSeek doesn’t solve all the problems associated with Generative AI. It doesn’t stop the so called hallucinations, it doesn’t solve the problems with bias and it is still only a productive language model. But through innovation DeepSeek has greatly reduced the need for vast amounts of power, has shown there are alternative approaches to demand for leakage amounts of capital investments and has provided an free and Open Source model for further research and innovation.
About the image
'Morning View' is part of the artist's series, 'Algorithmic Encounters': By overlaying AI-generated annotations onto everyday scenes, this series uncovers hidden layers of meaning, biases, and interpretations crafted by algorithms. It transforms the mundane into sites of dialogue, inviting reflection on how algorithms shape our understanding of the world.
Janet Turra & Cambridge Diversity Fund / Better Images of AI / Ground Up and Spat Out / CC-BY 4.0
Its probably fair to say that the hype around Generative AI far outstrips the reality. Perhaps that is because in the slow process of discovering actual jobs which AI can do, copy editing and advertising is one of those at the forefront.And goven that the AI companies are very keen on hyping their products, there is an endless stream of artciples saying how wonderful AI is for almost every occupation. These articles almost always talk about how much time AI saves and how this increases productivity.
Here's an example from a company called Screenloop:
"AI is not simply a tool—it’s a disruptor. In an era where agility and efficiency are non-negotiable, the role of AI has extended beyond merely assisting recruiters— it’s reshaping the very fabric of how we think about talent acquisition. What was once seen as an administrative task is now a strategic role in defining a company’s future.
Now that AI tools can handle resume screening, scheduling, and even initial candidate engagement, recruiters find themselves in a unique position. The question now isn't just 'how much time can we save?' but rather, 'how can we strategically reinvest it? How can we maximise the impact of the time AI gives back to us?'
Speed and Precision
AI's impact on recruitment processes, particularly in screening, is undeniable. Tools that can parse thousands of resumes in seconds have revolutionised the way talent acquisition operates. But the real advantage isn't just speed—it’s precision. By filtering out irrelevant applications and flagging the most promising candidates based on predetermined criteria, AI doesn't just make processes faster, it makes them smarter.
Enhanced Candidate Experience
A significant benefit of AI in recruitment is the potential to enhance candidate experience. In fact, companies using AI in hiring have reported a 30% increase in candidate satisfaction, not because the tools replaced human interaction, but because they created space for more meaningful engagement. When AI handles repetitive tasks, recruiters can focus on optimising the candidate experience based on insights collected by tools such as Screenloop's candidate pulse solution. And this matters—experience drives employer brand perception in a competitive market."
AI may not take your jobs but it's a headache in recruitment.
As the ISE have been telling us in detail, AI is not, so far, displacing loads of jobs as might have been feared a couple of years ago, but it's still having quite an impact. AI is good at writing covering letters and CVs, and so it makes sense for candidates to use them, and so they do. That means it's a lot easier for candidates to write a lot of relatively good job applications, quickly, and so that's exactly what they do.
This means everyone is applying for all the jobs available, so even though there are actually more jobs than there used to be, they're all getting more applicants, all using the same tools, with largely identical applications and recruiters are swamped, which means they have to spend more resources to administer a recruitment round, which ultimately makes recruitment harder and more expensive. That may start to have an effect on vacancy numbers.
What recruiters want to do is encourage applicants to use AI well - after all, it's likely to be a useful business skill - and discourage it being used badly. So they don't want to stop it entirely, but do expect a lot more talk this year about how to limit it being used in applications. And a lot of talk from online hustlers claiming they have a magic solution to make your applications foolproof using AI, of course.
About the feature image
The outputs of Large Language Models do seem uncanny often leading people to compare the abilities of these systems to thinking, dreaming or hallucinating. This image is intended to be a tongue-in-cheek dig, suggesting that AI is at its core, just a simple information ‘meat grinder,’ feeding off the words, ideas and images on the internet, chopping them up and spitting them back out. The collage also makes the point that when we train these models on our biased, inequitable world the responses we get cannot possibly differ from the biased and inequitable world that made them. Attributions - Studio of: Willem van de Velde II, Michele Tosini https://nationalgalleryimages.ie/groupitem/40/ This image was created using Canva: www.canva.com
Reihaneh Golpayegani & Cambridge Diversity Fund / Better Images of AI / Women and AI / CC-BY 4.0
I have been moaning lately about the quality of so called research and publications about education, learning and the. use of Generative AI. Well, the hype is showing no signs of dying down but there does seem to be some pretty good research beginning to emerge. And I understand it takes time to do research especially if you are trying to find out about the potential impact of AI on learning.
We study the adoption of ChatGPT, the icon of Generative AI, using a large-scale survey linked to comprehensive register data in Denmark. Surveying 18,000 workers from 11 exposed occupations, we document that ChatGPT is widespread, especially among younger and less-experienced workers. However, substantial inequalities have emerged. Women are 16 percentage points less likely to have used the tool for work. Furthermore, despite its potential to lift workers with less expertise, users of ChatGPT earned slightly more already before its arrival, even given their lower tenure. Workers see a substantial productivity potential in ChatGPT but are often hindered by employer restrictions and a perceived need for training.
Somebody - and I cant remember who - usefully got Chat GPT to do a summary and published it on LinkedIn:
41% of employees said they have used ChatGPT for work tasks.
Women are 16% less likely to ChatGPT for work than men.
Marketing professionals are the most likely to use ChatGPT (at 65%). Financial professionals are the least likely to use it (at 12%)
Less experienced and younger employees are more likely to use it. Every year of experience and age reduces likelihood of use by 0.6 & 0.7 percentage points.
More highly paid professionals are likely to use it.
Employees think ChatGPT can lead to big productivity gains in their job. They said that it could half the time to complete about 1/3 of their tasks. However many employees remain very uncertain about time savings from using the tech.
Despite these perceived time savings, employee regular use remains limited. For instance among employees who think it will save 1/2 the time in their job, only about 1/3 intend to use it.
Employees think ChatGPT can lead to big productivity gains in their job. They said that it could half the time to complete about 1/3 of their tasks. However many employees remain very uncertain about time savings from using the tech.
Despite these perceived time savings, employee regular use remains limited. For instance among employees who think it will save 1/2 the time in their job, only about 1/3 intend to use it.
Time saving may not lead to greater productivity. 37% of employees said they will not complete more tasks if ChatGPT can do it for them. 24% said they will devote more effort to using ChatGPT if it can save time.
The use of ChatGPT is mainly driven by individual worker initiative rather than company policy and systems.
Employees often face frictions in using ChatGPT. The limiting factors seem to be lack of training (42%) and company restrictions on use (32%). Restrictions on use was particularly high in the financial sector (82%). only 8% of employees reported fear of job loss as a reason for not using chat gpt.
I think the finding that the use of ChatGPT is mainly driven by individual worker initiative rather than company policy and systems nis interesting. It is reflected in our findings from the AI Pioneers project that most use is of GenAI in vocational education and training is mainly driven. by individual teacher initiative! But most research in learning or rather more commonly education, had focused on formal teaching and learning. But of course most people trying out GenAI are informal learners and there has been less insight into this.
About the image
This image is inspired by Virginia Woolf's A Room of One's Own. According to this essay, which is based on her lectures at Newnham College and Girton College, Cambridge University, two things are essential for a woman to write fiction: money and a room of her own. This image adds a new layer to this concept by bringing it into the Al era. Just as Woolf explored the meaning of “women and fiction”, defining “women and AI” is quite complex. It could refer to algorithms’ responses to inquiries involving women, the influence of trending comments on machine stereotypes, or the share of women in big tech. The list can go on and involve many different experiences of women with AI as developers, users, investors, and beyond. With all its complexity, Woolf’s ideas offer us insight: Allocating financial resources and providing safe spaces-in reality and online- is necessary for women to have positive interactions with AI and to be well-represented in this field.
This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Cookie settingsACCEPT
Privacy & Cookies Policy
Privacy Overview
This website uses cookies to improve your experience while you navigate through the website. Out of these cookies, the cookies that are categorized as necessary are stored on your browser as they are essential for the working of basic functionalities of the website. We also use third-party cookies that help us analyze and understand how you use this website. These cookies will be stored in your browser only with your consent. You also have the option to opt-out of these cookies. But opting out of some of these cookies may have an effect on your browsing experience.
Necessary cookies are absolutely essential for the website to function properly. This category only includes cookies that ensures basic functionalities and security features of the website. These cookies do not store any personal information.
Any cookies that may not be particularly necessary for the website to function and is used specifically to collect user personal data via analytics, ads, other embedded contents are termed as non-necessary cookies. It is mandatory to procure user consent prior to running these cookies on your website.
