If you’ve lived in China or worked with Chinese businesses, you’ve likely efficiently used them several times a day, ranging from ordering a coffee from Luckin (the Chinese version of Starbucks) to topping up your SIM card. If you haven’t heard of mini apps but want to find a good platform to deploy your AI-empowered services or content, then this is the article that you should read.

Tencent’s app, WeChat, is currently the only “super app” in China, with Alibaba’s AliPay perhaps being the only rival in certain functions. WeChat is not just a chat app like WhatsApp or Line. It not only encompasses Facebook’s multiple functions, but it is also widely used for functions that are core to Chinese daily life, essentially a Chinese version of posting photos, making payments, ride hailing, or even donating to charity, all in one.

And for context, WeChat has 1.34 billion monthly active users globally. Along with its ubiquitous e-wallet and jaw-dropping new functions, such as taxi hailing plug-in from DiDi, WeChat thrives on its mini programs—a core functionality of its digital ecosystem that strengthens WeChat’s number one “go to app” status in China.

What exactly are Mini Programs? Why are they so powerful? We’ll uncover what makes them tick and provide a practical guide for making sense of it all. We also discuss how this could relate to AI deployment in China, particularly in the context of customer service.

What Are WeChat Mini Programs? Apps within an app

Forget about the idea of constantly adding and upgrading your Apps. Imagine this: no downloads, no installations, no storage space taken. You scan a QR code or simply do a keyword search within WeChat, and a fully functional, user-friendly page loads in seconds. Furthermore, you can leverage your social network and share these applications with your friends as you wish.

This is the WeChat mini program (微信小程序): It’s a lightweight application technology built by Tencent inside WeChat, its host platform. With over 1.2 billion monthly active users, it has completely reshaped how Chinese interact with the digital world. Mini programs were first created in 2017. The number of mini programs has grown very quickly, from 580,000 in 2017 to 4.3 million in 2023.

For users, its core appeal is ultimate convenience. Use it and leave it. Sharing is effortless, and integration with WeChat provides a seamless and secure (Tencent-moderated) experience.

For businesses, it is highly cost-effective. One codebase works on both iOS and Android. Customer acquisition is cheaper thanks to seamless access to WeChat’s social graph, search, and geo-location features. It also easily integrates critical functions like WeChat Pay, forming a perfect commercial closed loop.

From milk tea shops on street corners to giant e-commerce platforms like JD.com, mini programs have become an essential part of the digital infrastructure in daily life in China.

Who Are the Giants in the Mini Program World?

The technology itself is developed by Tencent, the creator and owner of WeChat. But who are the most important players thriving via the mini programs?

The answer isn’t a single name but a coalition of empires:

• Internet giants: Like JD.com (occupying WeChat’s shopping sector), Pinduoduo (social e-commerce), Meituan (for within-the-day purchase and delivery services), Trips.com “(first Mini Program stock”). They use Mini Programs as extensions of their core businesses, capturing massive traffic and transactions.

• Digitally transforming brands: Luckin Coffee and Heytea have perfectly blended offline ordering with online marketing through Mini Programs, redefining the consumer experience. All major brands, such as Nike and Uniqlo, also utilize Mini Programs for sales and membership management.

• Tool and content kings: Like Tencent Docs and the viral photo book app Xiaoniangao, which dominate their specific verticals.

• Government service platforms, including mini programs, are extremely successful also because many government agencies are now on them, which boast enormous user numbers and serve as key channels for China’s digital governance and daily transactions. Increasingly, city-level governments, official tourist sites, airports, and even some regions’ tax-filing services rely on mini programs on WeChat to deliver their social services.

Leveraging the social network of WeChat

WeChat mini programs are not just for the big players. They also create opportunities for SMEs and individuals (such as school students) to create their own projects. While living abroad, I did not have any immediate need to try the big players’ mini programs mentioned above. Instead, one day I received a friend’s invitation to take some fun personality tests, and this turned out to be how many of us got hooked into the ecosystem.

To take mini games and tests as a case study, these applications perfectly align with WeChat’s social nature and demonstrate how companies leverage the sharing feature as a key to expand business. Quizzes such as “Which character from Empresses in the Palace are you?” (one of the all-time favorite historical dramas in China) are light-hearted social currency for moments. Increasingly, there are also professional assessment-driven tests, developed by certified institutions, often paid for. They are used for professional career evaluations or mental health screenings (e.g., SCL-90 scale). These interactive mini apps can quickly capture the attention of the 1.3 billion daily active users on WeChat.

For smaller players, the key to their success is sharing. Test results act as personal badges, and users love sharing them to seek validation, spark conversations, or express identity. This viral mechanism drives incredibly low-cost traffic to mini programs.

Mini programs have evolved from a feature into an ecosystem, an economy, and a lifestyle. Whether you’re a user, entrepreneur, or developer, the mini program universe is worth understanding. It’s not only a window into China’s digital society but also a blueprint that may define the future of mobile experience. And according to the company’s disclosure, mini programs are being used widely outside of China too, especially by businesses across housing rental, food and dining, transportation, logistics, and delivery services.

How WeChat Mini Programs and AI Redefine User Experience

WeChat mini programs have revolutionized the digital landscape in China by offering lightweight app-like functions without the need for download. In the age where digitalized service becomes essential, mini apps are rising in demand. And their power is further unlocked when they converge with an algorithm.

Many e-commerce and service-based mini programs integrate chatbots to handle customer inquiries, recommend products, and track orders 24/7. News and video mini programs use algorithms to analyze user behavior and serve personalized content feeds, thereby increasing engagement.

WeChat mini programs serve as readily available vehicles for AI, particularly in delivery customer service. Instead of a user manually searching for a product or service, AI empowers programs to anticipate needs and personalize the entire interface. For instance, the JD.com shopping mini program greets users with a homepage uniquely tailored to their past behavior, while a food delivery mini program like Meituan might proactively suggest your usual order at lunchtime. This creates an experience of discovery rather than search, where the AI acts as an intuitive guide, dramatically reducing the effort required from the user.

In the super app ecosystem, the goal of superior design is to make the technology feel invisible. The most successful implementations are those where the AI works so seamlessly in the background that the user simply feels smart and efficient, enjoying a service that is effortlessly personalized to their context and needs. This adaptation creates experiences that are not just functional but feel genuinely intuitive and responsive.

And here is a last note if you’d like to dig deeper–

How to Create Your Own Mini Program?

For developers, the Mini Program tech stack is friendly and well-documented:

• Tool: The official WeChat Developer Tools is the all-in-one platform for coding, debugging, and uploading.

• Process: Register an account → get an AppID → develop → test → submit for Tencent’s review → publish.

From what we learned, this could take as little as one day and as little as 300 RMB (or 42 USD).

As promised, I’m bringing in more diverse guest writers to AI Proem. The second guest post featured this week is from Anna Kim, a U.S.-licensed lawyer (Washington DC and Tennessee) who has now been based in Hong Kong for over ten years. She will shed some light on WHY and HOW the U.S. and China are competing in the tech race (I don’t like to frame it that way, but it is what it is, I guess) from a legal perspective.

During her legal career, Anna has represented clients on M&A and joint venture transactions, engineering and construction projects, and conflict resolution. Currently, she is taking a break from law practice to think and write about things that interest her—in this case, the development of artificial intelligence.

On Substack, Anna intends to explore the intersection of law and technology, staying engaged with emerging trends and connecting with others in the field. Anna is open to collaborating on posts about AI and related laws and regulations. Feel free to DM her directly if you’d like to discuss ideas. You can also find her on Linkedin here.

Let me know how you like these guest posts, and I’ll try to keep bringing in interesting writers/ experts every couple of weeks or so. But for next week, you’ll have old fart me again, soz.

Subscribe now

High-Stakes Competition Game

It is no secret that the United States views China as its main challenger in the global race for technological dominance. The rivalry extends to such areas as 5G, advanced semiconductors, and now also artificial intelligence. Leadership in AI is widely perceived as critical for future economic growth, national security capabilities, and even shaping culture and societal norms. The competition is spurring an important debate concerning the impact of intellectual property (IP) frameworks on AI development. AI models, especially LLMs, are being trained on a vast amount of data, oftentimes including copyrighted materials. The central question of the debate is should the tech companies be allowed to get away with using copyrighted content for LLM training to remain competitive?

Redefining Fair Use

A key flashpoint in the U.S. is the recent push by major AI players like Google, OpenAI, and Anthropic for strategic government involvement in AI regulation. The companies, among other things, are pushing to redefine the boundaries of "fair use" under U.S. copyright law, specifically proposing the inclusion of training LLMs on vast datasets—which invariably include copyrighted materials—as a legitimate form of fair use.

The tech companies’ rationale for redefining “fair use” rests on several core arguments:

• Innovation and progress: Broad access to data for training is essential for driving AI innovation, scientific breakthroughs, and societal benefits. Limiting this access would hinder technological innovation, especially for smaller and newer companies with limited budgets.

• Practical necessity: Training LLMs requires vast datasets, making individual licensing for every piece of copyrighted material impractical and overly burdensome, thereby slowing down model development and scientific experimentation.

• Democratizing AI: Access to more data from the widest possible range of sources will ensure more powerful innovations that deliver more knowledge, contributing to the success of "democratic AI.”

• National security and global leadership: Restricting the use of copyrighted material for AI training would weaken America's technological lead in AI and potentially allow China to overtake the U.S., thus posing a national security risk. Maintaining "American AI models’ ability to learn from copyrighted material" is crucial to winning the AI race.

However, this push has been met with fierce opposition. In addition, a significant number of lawsuits (reportedly 39 as of March 13, 2025) have been filed in the U.S. against leading AI companies by content creators and copyright holders.

Among other things, critics argue that:

• Market usurpation: AI-generated content (text, images, code, music) directly competes with human creators and can potentially devastate creative industries.

• Unprecedented scale: The sheer volume of data scraped—often entire swathes of the internet—goes far beyond traditional fair use precedents, representing systematic, industrial-scale copying.

• Lack of compensation: It is patently unfair for multibillion-dollar companies to be using creative works en masse, without permission or compensation, to build commercial products.

Early Rulings

While the vast majority of the lawsuits against major LLM developers are still winding their way through the courts, the legal landscape is not entirely a mystery. Early rulings in related cases offer initial clues—and they haven't been favorable news for those seeking broad fair-use protections for AI training.

One recent case is Thomson Reuters v. Ross Intelligence. Thomson Reuters, the company behind the legal research platform Westlaw, sued an AI-powered legal research company, Ross Intelligence, because Ross used material from Westlaw – specifically, short summaries of legal points called headnotes – to train its own research tool. Thomson Reuters claimed this was copyright infringement. The court ultimately sided with Thomson Reuters. It was found that Ross copied protectable parts of the headnotes and that Ross's use to train its AI model was not considered "fair use. “The court considered factors such as the commercial nature of the use, the extent of the copying, and the direct competitive harm to Westlaw's market.

While Google, OpenAI, and Anthropic operate different business models than Ross, and their specific cases involve different types of data and uses, the outcome in Thomson Reuters v. Ross Intelligence undeniably raises concerns. It signals a judicial willingness to scrutinize fair use claims in the AI training context, especially where there is commercial use, direct competition, and substantial copying of content. This precedent, even if not directly binding on all future cases, likely fuels the AI companies' nervousness and intensifies their lobbying efforts for legislative clarity or intervention.

The China Card

It is conceivable why U.S. AI companies, facing mounting legal challenges at home, would choose competition with China as a focal point to sway government action and public opinion. Out of all the major players, OpenAI has been perhaps the most fervent advocate for relaxing copyright constraints, explicitly framing it as a matter of national competitiveness. They write in their submissions to the U.S. government, “[T]here’s little doubt that the PRC’s AI developers will enjoy unfettered access to data — including copyrighted data — that will improve their models. If the PRC’s developers have unfettered access to data and American companies are left without fair use access, the race for AI is effectively over.” (emphasis added)

Furthermore, the companies emphasize the potential risks to U.S. national security, economic leadership, and democratic values should China surpass the United States in AI capabilities. This framing is a strong strategic choice. National security arguments often resonate with policymakers and can elevate commercial interests to matters of state importance, potentially overshadowing concerns about creator rights or market impacts.

However, this seemingly patriotic stance might be viewed as somewhat self-serving. While concerns about geopolitical competition are valid, leveraging the "China threat" provides a powerful narrative to argue for regulatory changes that directly benefit the companies' business models and shield them from costly litigation and licensing demands. As such, OpenAI’s and Google’s rhetoric should be taken with a grain of salt.

China's Approach

But what is the actual approach to data scraping and AI training in China? Is the use of copyrighted materials as “unfettered,” as OpenAI suggests? The answer appears to be no, however, there are indications that OpenAI’s underlying fear—that Chinese AI labs may face fewer legal hurdles compared to their U.S. counterparts—could be justified.

As a U.S.-licensed lawyer, I am not an expert in Chinese law. However, based on research into recent developments, China seems to be adopting a pragmatic approach. There are no specific laws solely regulating using data for AI training. Instead, court decisions and regulatory focus often appear centered more on anti-competitive effects and the legality of data access methods (e.g., unauthorized scraping, breaching website terms, or technical barriers) rather than copyright infringement per se.

To illustrate using the Ross case: hypothetically, Ross Intelligence might still face legal trouble in China, not necessarily for copyright infringement of headnotes, but perhaps for using unauthorized methods to access proprietary data from a direct competitor (Westlaw) to build a competing product, potentially violating anti-unfair competition principles. However, scraping publicly available images or text from various websites to train a foundational LLM, which doesn't directly replicate a specific competitor's product, might not trigger the same illegal access or anti-competitive concerns. Therefore, hypothetically, such large-scale scraping for general training might currently face fewer legal obstacles specifically related to copyright claims compared to the U.S.

Perceived Edge

It is understandable how the perceived difference in approaches between the United States and China may lead to the argument that China's potentially less restrictive environment concerning copyright enforcement and data access might position it for more rapid AI innovation. The thinking is that China's strong national focus on AI leadership, combined with this potentially more permissive legal interpretation regarding the use of vast datasets for training (perhaps irrespective of copyright status elsewhere), could grant Chinese developers access to a larger data pool, thereby accelerating innovation.

However, it is crucial to temper this perspective. Firstly, China's legal and judicial systems are also actively grappling with AI's challenges. Secondly, the landscape is not definitively permissive; significant ambiguities remain. This lack of legal certainty can be a double-edged sword: while it might allow for faster movement in some areas now, it can also hinder long-term investment and create risks for Chinese companies, especially those with international ambitions whose models might be challenged under stricter IP regimes abroad.

Key Takeaways & Recommendations

• The U.S. legal landscape remains uncertain: Expect continued high-profile litigation. While early rulings (like Ross) suggest courts may protect creators when direct copying and competition are clear, definitive rulings on broad fair use claims for LLM training data are still pending. Relying solely on fair use as a defense carries significant risk.

• The "China Advantage" narrative needs nuance. While China clearly prioritizes AI development and has a different legal approach, portraying it as a simple copyright "free-for-all" is an oversimplification. Using this narrative as the primary justification for potentially weakening U.S. copyright protections warrants careful scrutiny.

• Proactive data governance is key: For AI developers, investing in transparent data sourcing, exploring licensed datasets, building data partnerships, and developing robust data provenance tracking are becoming crucial strategies to mitigate legal and reputational risks, regardless of how fair use ultimately evolves.

• Constructive dialogue over confrontation: A more productive path forward involves constructive dialogue between AI labs, content creators, and policymakers to explore potential frameworks (like tiered licensing, compensation pools, or technical standards) that balance fostering innovation with respecting creator rights and ensuring fair compensation.

• Monitor global regulatory trends: The development of AI regulation is a global phenomenon (e.g., the EU AI Act). How other major jurisdictions address AI training data and copyright will inevitably influence international norms and potentially shape future U.S. policy directions.

  ---

Disclaimer: All guest posts are fully editorially independent from Grace and represent the expertise and opinions of the authors only.

Our guest of the day is , a policy strategist, writer, and International Strategy Forum (ISF) Fellow— a program by Schmidt Futures that supports rising leaders at the intersection of geopolitics, technology, and public service.

She has served in the U.S. State Department, leading foreign policy think tanks and crisis diplomacy roles. She is a graduate of Harvard Kennedy School and speaks English, Spanish, and Mandarin, just to name a few languages. She’s quirky and fun, knowledgeable and patient. We met when we were both living in Beijing in 2015, and it’s so great to reconnect on Substack. :D

Her newsletter is called , a space for her to interrogate AI’s potential—and its limits—through hands-on experiments, sharp analysis, and a healthy dose of irreverence.

Subscribe now

1. Penguins, Tariffs, and the Algorithm That Governs Us

This week, the Trump administration announced 48% tariffs on Laos, 46% on Vietnam, 20% on the EU—and yes, even duties on the Heard and McDonald Islands, a penguin-filled Australian territory with no actual trade.

The reason? A so-called “Liberation Day” meant to reclaim America’s economic independence.

The method? Well, that’s where things get uncanny.

Because these tariffs don’t read like diplomacy. They don’t even read like strategy.

They read like someone typed this into a chatbot at 2am:

“Give me a simple, fair-sounding formula to fix the U.S. trade deficit with tariffs.”

And then copy-pasted the result into official policy.

When I started this blog, I wanted to explore what happens when we ask AIs to think—to weigh in on geopolitics, trade, culture, even war. I didn’t expect the U.S. government to do the same.

But here we are: not just an artificial inquiry, but what looks like an artificial implementation.

This post is about that: how Liberation Day became a real-time case study in AI slop and what it reveals about a new kind of statecraft: fast, aesthetic, surface-level—vibe-governed by large language models.

2. What is AI Slop?

“AI slop” is the messy, uncanny residue of machine-generated output applied uncritically to complex problems. It’s when a policy or piece of content looks, feels, and acts like it came straight out of a chatbot—formulaic logic, surface-level symmetry, shallow nuance, zero context-checking. You’ve seen it elsewhere—legal memos that barely parse, news articles that feel auto-filled, job posts written by a logic engine. In design, it’s the odd extra finger on a hand; in geopolitics, it’s the 10% tariff on a penguin island with no human inhabitants.

3. How AI Probably Shaped the Tariffs

There’s a pattern to the Liberation Day tariffs. And it’s suspiciously tidy.

According to analysts and tech observers (including those quoted in Newsweek and Followin), many of the announced rates follow a clear formula:

It’s not speculation—it lines up almost perfectly.

• The EU’s ~$235.6B deficit on $605.8B imports = ~39% → tariff set at 20%.

• Indonesia’s ~$17.9B on $28.1B = ~64% → tariff: 32%.

• Japan’s ~$68B on $144B = ~47% → tariff: 24%.

• Vietnam? 92% imbalance → 46%.

• Laos? 48%. Cambodia? 49%.

Even places with no meaningful trade—like the Heard and McDonald Islands—somehow made the list. That’s not negotiation. That’s a spreadsheet. Trade logic, flattened into code—no one bothering to open the algorithm back up.

It’s the kind of logic you’d get if you prompted a model:

And then just ran with it.

Some economists have already pointed out that these aren’t actually “reciprocal” tariffs. They’re just targeting anyone who exports more to the U.S. than they import—abandoning Most-Favored Nation principles and blowing past WTO norms in the process.

This isn’t how trade policy usually works. It’s how auto-generated symmetry looks when no one applies human judgment.

Either someone manually recreated this logic, or someone asked an LLM a prompt, liked the vibe—and pasted it into policy.

Economist and software engineer Rohit Krishnan (Substack: Strange Loop Canon) retro-engineered a potential prompt and fed it so several AI models — the results were eerily similar to the actual Liberation Day tariffs.

4. What does the potential use of AI mean?

The real story isn’t whether AI was explicitly used to design trade policy. It’s that the logic reads like it was—because it rewards exactly what this administration already values: speed, spectacle, and surface-level coherence.

(See original tweet here).

The Trump administration has long thrived on speed, disruption, and visual dominance. Whether it’s early-morning tweet storms, ambiguous executive orders, or bombastic press conferences, their communications strategy favors fast, simple, black-and-white declarations. That’s where AI comes in—not as a strategic tool, but as a lazy weapon.

This speaks to a deeper danger: AI makes things look legitimate while delivering polished nonsense. A tool like ChatGPT delivers what they crave: immediacy, coherence at a glance, and surface-level rationality. It can churn out a policy doc in minutes, structure numbers with no real-world logic, and make contradictions look clean. In short: it’s perfect for manufacturing techno-plausibility with no accountability.

The intent wasn’t “let’s build an AI-led trade doctrine.” The intent was: let’s go fast, look powerful, and say we’ve done something bold.

AI didn’t just grease the wheels—it scaled the bullshit. Its structure amplifies the chaos, because it packages incoherence inside syntax that reads like strategy. And it does it all in seconds.

Which brings us to why it all feels so familiar. Because AI slop doesn’t just deliver speed—it delivers a mood. And that mood is exactly what this administration knows how to weaponize.

5. AI Slop as “Vibe Governance”

That legit-sounding nonsense isn’t a bug—it’s a feature. And it’s not just useful in policy terms. It’s emotionally resonant. And this dovetails perfectly with Trump's existing modus operandi, which many online have been dubbing “vibe governance”: when political action is driven less by institutional process and more by affective impact. Not “What will this do?” but “How will it land?”

In this case, the vibe was: “America is being treated unfairly. Let’s fix it with math. AI math. Reciprocity math.”

“Vibe governance” may sound like a 2020s meme, but it echoes deeper traditions in political spectacle. From Nixon’s “law and order” theater to Berlusconi’s media dominance, governance by feeling—rather than by fact—is hardly new.

What distinguishes the AI era is the formalization of vibe into code. Language models don’t just mirror mood—they stabilize it. They don’t just simulate coherence—they smooth contradiction, synthesize ideology, and generate the tone of consensus, even when none exists.

It also provides an escape to laziness that wasn't there before. It used to take more time and work to come up with legit-sounding nonsense.

And that’s what makes this moment so slippery. This isn’t just bad policy. It’s simulation formalized as governance—where success is measured not by what works, but by what performs the vibe of working. In other words, AI is not only able to produce populism, but it can automate it while laundering it through the grammar of rationality.

In short, AI is the perfect accomplice for populist leaders. It doesn’t care what’s true. It just knows what passes as true. And that’s more than enough to govern under populism.

And once AI becomes a backstage partner in governance—even unofficially—it quietly reshapes how power moves. Not just what gets decided, but how those decisions get built, framed, and justified.

6. What Kind of Power Does AI Slop Enable?

AI isn’t making policy on its own. But it’s starting to shape how power operates: how decisions get framed, justified, and pushed through. Not by replacing humans, but by subtly reformatting the logic of governance.

Here’s what that looks like.

Power by Proxy

No need to consult trade economists. No interagency coordination. Just ask the model a “what if” and see what sticks. AI becomes the first draft, the scaffolding, the pseudo-expert no one has to challenge. A policymaker doesn’t need to own the analysis—just the output.

The result isn’t collaboration. It’s delegation, without attribution.

Power by Illusion

A tariff plan that feels “fair.” A chart that looks mathematically sound. A policy that reads like something someone thought through. GPT-generated logic offers structure without depth, symmetry without accountability. It sounds reasonable. And in a crisis-prone system, that’s often enough to carry the day.

This is governance via aesthetic: formatting over deliberation. And it’s persuasive.

Power Without Guardrails

There are no disclosure rules. No required prompt logs. No model audits. We don’t know what’s AI-generated and what isn’t—and that’s not an accident. The lack of structure isn’t a bug. It’s an advantage. It allows plausible deniability at every step.

And the scary part? There’s no threshold. Nothing says, “AI can be used here, but not there.” No one has drawn that line.

Because drawing the line would mean admitting it's already being crossed.

So if the line is blurry—and mostly invisible—what would it take to draw one? What’s a responsible threshold for using AI in real-world policy? And what happens if we never set one?

7. Thresholds and What Comes Next

So what should the threshold be? When is it okay to use AI in policymaking—and when does it become slop?

This isn’t a yes-or-no binary. Large language models can be useful: for drafting options, summarizing stakeholder positions, testing hypothetical outcomes. They’re good at structure. They can mirror conventional logic, which makes them handy when you need a clean frame fast.

But they’re bad at decisions. And they’re worse at judgment. Unless prompted with surgical precision and followed by real human review, they’ll always default to surface logic—what sounds right, not what is.

So the threshold can’t be about whether AI is involved. It has to be about how high the stakes are, and how much scaffolding is in place to catch what the model can’t see. Foreign policy, trade negotiations, national security—these aren’t places to wing it with auto-symmetry and vibes.

If AI is shaping policy, we should know. Release the prompts. Show your work. Make it auditable. Otherwise, what we’re doing isn’t statecraft. It’s performance with invisible collaborators.

And if slop becomes the dominant mode—if simulation replaces strategy entirely—we’re in trouble. You get policy written in 30 minutes, formatted like a white paper, and treated as if it carries the weight of deliberation. You get press releases that sound like consensus but mean nothing. You get real-world consequences—economic fallout, diplomatic backlash—structured by models that weren’t trained on impact.

And you get silence when you ask who wrote it.

The penguins are still staring. Honestly, at this point, they’re asking better questions than most of us.

Because the real problem isn’t that AI generated policy. It’s how quickly we decided that was good enough.

More about Artificial Inquiry

Artificial Inquiry is an experimental Substack exploring what AI-generated analysis can reveal about current events—and what it can’t. Each post uses tools like ChatGPT to dissect everything from leaked government chats to geopolitical groupthink—while also incorporating the author’s own meta-analysis of how AI performs, where it fails, and what that tells us about power, bias, and human reasoning. It’s part academic, part entertainment, and all public inquiry.

Disclaimer: All guest posts have full editorial independence from Grace, so they represent the expertise and opinions of the authors only.