AI Policy Primer (January 2025)
Issue #17: AI Infrastructure; AI-enabled journalism; and the outlook for critical thinking skills
Every month, our AI Policy Primer looks at three external developments from the world of AI policy that caught our eye. In this edition, we compare and contrast the recent UK and US AI infrastructure announcements, spotlight a study on how AI may affect critical thinking, and explore the future of AI-enabled journalism. Please leave a comment below to let us know your thoughts, or send any feedback to aipolicyperspectives@google.com. Thanks for reading!
Policymakers taking action
US and UK government announce flagship AI infrastructure efforts
What happened: The UK Government announced a new AI Opportunities Action Plan, while the Trump administration announced ‘Project Stargate’ - a $500bn initiative to expand AI training infrastructure, led by Softbank, OpenAI, Oracle and others.
What’s interesting: Both efforts aim to attract investment for AI infrastructure, but they also highlight the starkly different approaches on either side of the Atlantic. The UK’s Plan, praised for its ambition, includes 50 recommendations, ranging from deregulating data centre planning to promoting AI adoption in the public sector. The goal is to ‘on-shore’ AI activity by leveraging the UK’s existing strengths - such as its strong university research base (Oxford, Cambridge, Imperial, etc) and a robust AI safety community - while addressing planning and energy constraints that have hindered broader economic growth .
The plan also signals a potential shift in UK economic policymaking. By establishing a “UK Sovereign AI unit” inside No.10, and “AI Growth Zones” with assumed planning approval for AI infrastructure and energy projects, the Plan explicitly states that “the invisible hand of the market” alone will not suffice. Instead, the UK must take a proactive role to remain competitive in AI.
The US, with lower energy costs, a stronger industrial base, and a more liberal approach to infrastructure, is in a fundamentally different position. Project Stargate is private-sector-led and operates on a scale that the UK and other nations cannot realistically match.
However, both initiatives face questions about funding and execution. For Stargate, the US government’s role remains unclear, raising concerns about bottlenecks in energy, land, and resource allocation. In the UK, much will hinge on the forthcoming Spending Review.
Ultimately, these announcements underscore how AI is becoming central to economic strategy and geopolitical influence. The ‘race for compute’ is set to intensify, despite ongoing breakthroughs in AI efficiency, leading to more public-private partnerships for large-scale AI investment.
Whether a significant share of this spending happens outside of the US will largely depend on how effectively other nations, including the UK, execute on their AI strategies, but will also be impacted by policies like the new US AI Diffusion Framework.
Study watch
Will AI use hurt critical thinking skills?
What Happened: A new study of 666 UK participants, spanning diverse age groups and educational backgrounds, found a strong self-reported negative correlation between AI tool use and self-reported critical thinking skills.
What’s Interesting: Over the past 20 years, the advent of the Internet has led educators to focus on equipping students with ‘hard’ STEM skills, alongside ‘soft’ skills like collaboration and critical thinking, to prepare them for an increasingly digital society.
Critical thinking involves analysing, evaluating, and synthesising information to make reasoned decisions. It encompasses problem-solving, decision-making, and reflective thinking, but as a concept it remains somewhat vague and it is difficult to assess how teaching programmes or technology affect it.
In particular, a long-running debate exists over whether technologies that automate routine tasks - from calculators to personal computers to AI - support or hinder critical thinking. There is also concern over whether such technologies may erode foundational knowledge or skills - like reading or numeracy - that may be essential for critical thinking and which appear to be stagnating or even declining in many countries.
In this study, the authors surveyed participants on how frequently and in what ways they used AI to retrieve information and make decisions. Participants were then asked about AI’s impact on their ability to think critically and solve problems independently, as well as their concerns regarding AI bias and transparency.
The findings suggest that while AI enhances efficiency for some individuals, it may come at the cost of a decline in independent problem-solving and critical analysis - or at least a perception of one. This could be due to ‘cognitive offloading’, where users delegate tasks to AI without redirecting their efforts to more complex, higher-order thinking.
The study also found that younger participants (17–25 years old) and those with lower educational attainment reported a greater dependence on AI, potentially reducing their ability to critically evaluate information and identify biases.
These findings mirror a recent experiment where students given temporary access to LLMs for learning performed worse on exams once access was revoked, compared to students who had never used the tools in the first place.
While this study raises concerns about AI’s impact on critical thinking, it also presents an opportunity. The research suggests that AI’s effects are malleable, and education could provide ‘cognitive scaffolding’ to help users engage with AI in more productive ways. For example, in their recent LearnLM work, our colleagues developed pedagogy-inspired LLMs that encourage students to reflect on questions rather than simply answering them. More broadly, there is a need for new kinds of AI literacy that does not just teach students, or educators, what AI is, but how to use it most effectively.
Sector spotlight
What Happened: John Micklethwait, editor-in-chief of Bloomberg News, and former editor of The Economist, shared predictions on AI’s impact on journalism in a talk at the James Cameron Memorial Lecture. He outlines a positive vision for AI’s integration into journalism and journalist jobs, yet predicts a decline for traditional Search and ad-based revenue models.
What’s Interesting: Micklethwait compares AI’s disruption to the early 2000s, when the Internet was blamed for ‘killing’ newspapers. He argues that media outlets were too quick to accept tech companies’ claims that content should be free, leading to a race to the bottom in pursuit of clicks. More recently, publications like The New York Times and The Information have reversed course, building high-priced subscription businesses. With AI, he expects a faster shift to high-quality AI-enhanced content, as media outlets try to avoid repeating past mistakes.
To illustrate AI’s role in journalism, Micklethwait highlights that one-third of Bloomberg’s 5,000 daily articles now incorporate some form of automation. In one example, investigative journalists used AI to analyse satellite imagery of ship movements to uncover oil smuggling from Iran. On the other end of the journalism spectrum, Bloomberg also uses AI to generate bullet-point article summaries - a feature disliked by journalists but valued by readers.
In his predictions, Micklethwait envisions AI reshaping journalism tasks but not eliminating jobs. He cites Bloomberg’s continued employment of roughly the same number of company earnings reporters, despite years of automation in that area. Similarly, he expects AI to play a growing role in editing and formatting articles, but to remain less capable of other editorial actions - such as commissioning stories or persuading a cabinet minister to reveal a resignation.
Micklethwait also warns of AI-enabled misinformation, particularly in image and video content, which he sees as more harmful than text-based misinformation - especially for fast-moving news stories, where social media plays a key role in verification. Following licensing deals, he expects the decline of traditional search engines and the media outlets dependent on search-driven revenue. After many false dawns, he also expects the long-awaited emergence of truly personalised AI news offerings.
Looking backwards, over the past 15 years, the number of working journalists in the UK and US appears to have grown slowly, and with increasing diversification of Internet-enabled roles. While uncertainty is high, we expect this trend to continue over the next five years, as AI-driven analysis becomes standard, though it remains to be seen whether traditional or newer media outlets will lead the shift.