WASHINGTON, DC, USA –– White House Press Secretary Karoline Leavitt answers questions from journalists on Jan. 28, 2025.
WASHINGTON — As the Trump administration increasingly welcomes vloggers and social media influencers into press briefings and the Oval Office, established outlets like the Associated Press find themselves excluded from the century-old press pool, sparking controversy about what "transparency" truly means.
Joshua Sukoff is a photojournalist from Long Island, New York. He is currently studying journalism at Northwestern University.
An illustration of AI chat boxes.
Earlier this week, a member of the UK Parliament—Mark Sewards—released an AI tool (named “AI Mark”) to assist with constituent inquiries. The public response was rapid and rage-filled. Some people demanded that the member of Parliament (MP) forfeit part of his salary—he's doing less work, right? Others called for his resignation—they didn't vote for AI; they voted for him! Many more simply questioned his thinking—why on earth did he think outsourcing such sensitive tasks to AI would be greeted with applause?
He's not the only elected official under fire for AI use. The Prime Minister of Sweden, Ulf Kristersson, recently admitted to using AI to study various proposals before casting votes. Swedes, like the Brits, have bombarded Kristersson with howls of outrage.
I'll bite and attempt to defend “AI Mark” specifically and the use of AI by elected officials more generally.
Let's start with “AI Mark.” While I understand public frustration around the seemingly hasty adoption of AI into a key government service, my research suggests that those ready to remove Sewards from office are failing to ask a key question: what's the alternative?
"AI Mark" was designed specifically to make up for Sewards' inability to meaningfully respond to manifold constituent inquiries. According to Sewards, he has "tried [his] best to sit at [his] desk and answer all the requests that come through on [his] laptop, but it’s not possible for one person to do that." "AI Mark,” on the other hand, can analyze such requests around the clock. That said, constituents want more than merely to be heard (or read); they're reaching out for some affirmative action by the MP. So, can "AI Mark" help with that?
My hunch is yes. Constituent work is hard. It's arguably the most important role for elected officials. Yet, it's also one of the least appreciated and one of the hardest to do well. Done right, constituent services performs at least three functions: first, it ensures individuals can get through complex bureaucracies; second, it surfaces emerging issues that warrant broader attention; and, third, it directs the elected official to prioritize issues that are most relevant to their communities.
Speaking from my experience as a former intern to a U.S. Senator, I can testify to the fact that "AI Mark" is likely an improvement upon the alternative of either a small army of undergraduate interns pouring over those constituent requests or the elected official themself attempting to do so.
Finding substantive constituent inquiries is no easy task. For every one person reaching out for support on a substantive matter, there are likely dozens, if not hundreds or thousands, of duplicative or irrelevant submissions. Hundreds of people may send identical letters urging a vote on a certain issue—a human is not necessary to read each of those; AI can quickly consolidate such letters. Other submissions may involve demands that exceed the authority of the office. There's little need for a human to confirm that the MP, senator, or representative does not have jurisdiction over that request. AI can do that with a high degree of accuracy in a fraction of the time. AI can then quickly filter through the flood of requests that likely do not merit much attention. The elected official and their staff can use that saved time to more promptly take action on the remainder. That's a win for everyone.
With respect to PM Kristersson and the use of AI for research, a similar defense can be raised. Elected officials are often short on time to do research on every issue that comes before their desk. In some cases, they will get a briefing from their staff explaining the pros and cons of that decision. Such analysis may not be high-quality. There’s the possibility that the staffer thinks they know how the official wants to vote or should vote and, therefore, biases their report. There’s also high odds of that staffer being taxed for their time themselves and, consequently, producing an incomplete or inaccurate report. Finally, there’s the possibility of the staffer using AI to do the task! Sophisticated AI tools such as OpenAI’s DeepResearch can scour the internet for relevant sources and information in a matter of minutes; it would be strange if a policy researcher failed to make use of this tool to supplement their analysis. What’s the harm of the PM simply skipping to this final step?
The harm in this case, as well as the case of "AI Mark," is a lack of transparency and engagement. Clandestine use of AI is almost always going to incite public unrest. Folks like to know how and why their elected officials are working on their behalf. The answer, however, is not to prevent or oppose the use of AI in policymaking but rather to make sure such use is out in the open and subject to regular review.
Ninety years ago this month, the Federal Register Act was signed into law in a bid to shine a light on the rules driving President Franklin Roosevelt’s New Deal—using the best tools of the time to make government more transparent and accountable. But what began as a bold step toward clarity has since collapsed under its own weight: over 100,000 pages, a million rules, and a public lost in a regulatory haystack. Today, the Trump administration’s sweeping push to cut red tape—including using AI to hunt obsolete rules—raises a deeper challenge: how do we prevent bureaucracy from rebuilding itself?
What’s needed is a new approach: rewriting the rule book itself as machine-executable code that can be analyzed, implemented, or streamlined at scale. Businesses could simply download and execute the latest regulations on their systems, with no need for costly legal analysis and compliance work. Individuals could use apps or online tools to quickly figure out how rules affect them.
These aren’t theoretical ideas. The first prominent work in this area was undertaken by Prof. Robert Kowalski at Imperial College London, who codified the British Nationality Act as a set of rules. Since then, AI researchers have explored—and, in many cases, solved—the numerous challenges associated with turning regulations into code. That includes identifying areas where human judgment remains central, ensuring that encoded regulations clearly indicate where discretion applies, flagging potential exceptions, and certifying that decisions are fully traceable.
In the European Union, the GovTech4All project is developing a “Personal Regulation Assistant,” powered by regulatory code, to assist citizens in identifying and accessing benefits, regardless of their level of digital literacy or policy knowledge. The project will serve as a model to replicate the rules-as-code approach across other areas of European regulations.
In the U.S., meanwhile, the approach has been championed by private-sector innovators. Intuit’s TurboTax is a leading example, showing how the tax code can be translated into a computational interface to help individuals. The Bay Area startup Symbium has encoded regulations to enable California homeowners to secure solar installation permits—a process that used to take weeks or months of paperwork, revisions, and waiting—in just seconds.
Such ventures show the power of using digital tools to streamline the implementation of regulations—but they require individual businesses to interpret and codify the rules in question. If the tax code, the building code, or other regulations were already available as machine-executable rules, this process would be orders of magnitude faster, could be scaled nationwide, and would deliver powerful efficiencies across the U.S. economy.
Swapping our existing mishmash of PDFs and static webpages for elegant, unified computer code would instantly unlock important new efficiencies—automatically flagging ambiguities, simplifying complex rules, and eliminating redundancies without losing substance. It would also enable powerful tools like compliance test suites and public-facing rule repositories, driving greater transparency, reducing red tape, and enhancing ease of use.
What would it take to “encode” any rule book, regardless of whether it is at the federal, state, or city government level? The first step is to identify and codify the regulations in most need of an overhaul. Obvious examples might include engineering or design standards, which are currently slow to adapt to technological changes, but which are also prescriptive and could easily be rewritten as code. The processes for permitting and environmental impact assessments—already recognized by the White House as a target for new efficiencies—would be another leading candidate.
We’ll also need to use new technologies to enable rules to be converted into code in reliable and scalable ways. Such efforts have been daunting until now because of the huge manual effort required to analyze and rewrite regulations. New AI tools, however, make it possible to both analyze vast amounts of text and to write and rigorously validate computer code, with almost superhuman speed and accuracy. With regulatory sprawl wiping 0.8 percentage points from America’s annual GDP growth, using AI to accelerate the process of turning federal rules into code would deliver clear ROI and powerful efficiencies across the federal government and beyond.
As things stand, America’s federal agencies still use a 19th-century rulemaking process—and as individuals and businesses, we’re all paying the price for that. President Trump is right to push for reductions in government red tape. But that effort should be paired with a concerted effort to bring federal regulations into the 21st century and develop a machine-readable rule book that’s ready for the challenges and opportunities of the AI era.
Vinay K. Chaudhri supports a National Science Foundation initiative on Knowledge Axiomatization. Previously, he led AI research at SRI International and taught knowledge graphs and logic programming at Stanford University.
This morning’s announcement that Nvidia and AMD will resume selling AI chips to China on the condition that they surrender 15% of their revenue from those sales to the U.S. government marks a jarring inflection point in American industrial policy.
This is not just a transaction workaround for a particular situation. This is a major philosophical government policy shift.
What was once a matter of national security has now somehow become a transactional arrangement that is essentially a pay-to-export policy that blurs the line between protective regulation and economic extortion.
The deal, reportedly brokered after Nvidia CEO Jensen Huang met with President Trump, allows the companies to sell their H20 and MI308 chips in China, a market that previously accounted for billions in revenue.
There were two immediate responses from two former officials and trade experts who voiced strong concerns. Christopher Padilla, former head of the Commerce Department’s International Trade Administration, called the deal “astonishing,” likening it to “a mix of bribery and blackmail” and warning it may be “possibly illegal” and Peter Harrell, a fellow at the Carnegie Endowment for International Peace, said the arrangement sets a “terrible precedent,” noting that national security export controls should not be relaxed in exchange for financial concessions
The unintended consequences of a government policy are unknown, and the potential cost of access is steep, not just financially, but philosophically. This arrangement contradicts the very principles Republicans have long championed: free enterprise, limited government interference, and a clear separation between national security and commercial interest.
History provides some lessons.
Consider the Jackson-Vanik Amendment of 1974, which tied trade relations with the Soviet Union to human rights benchmarks, particularly the freedom of emigration for Soviet Jews. It was a moral stand, using trade as leverage to promote democratic values. But it was also transparent, legislated, and principled—even if imperfect.
Today’s chip deal with China lacks that moral clarity. It’s not about human rights or democratic norms. It’s about money. The 15% levy isn’t a tariff, a sanction, or a strategic investment. It’s effectively just a tax on corporations. And unlike Jackson-Vanik, it wasn’t debated in Congress or anchored in law. It was brokered behind closed doors, with the executive branch acting as both regulator and beneficiary.
This blurring of roles recalls the worst instincts of economic policy, where governments extracted rents from private enterprise under the guise of national interest. It also evokes the “crony capitalism” of post-Soviet Russia, where access to markets was contingent on political favor and financial tribute.
This policy sets a dangerous precedent by monetizing export permissions, effectively transforming the federal government into a tax collector on global commerce. It’s a dangerous slope. If the U.S. can demand a revenue cut for chip sales to China, what stops it from doing the same for pharmaceuticals to Europe, or software to Latin America? The logic of national security becomes a flexible pretext for economic leverage, and the private sector is left navigating a landscape where policy is shaped not by principle, but by deal-making.
This isn't an industrial strategy, but essentially it's economic policy to the highest bidder. And it sends a troubling signal to allies and adversaries alike: that American innovation is for sale, and its gatekeeper is no longer Congress or the Constitution, but the highest bidder in the West Wing.
If Congress is serious about reshoring supply chains, protecting intellectual property, and countering China’s techno-authoritarianism, then it must reject this ad hoc monetization of policy. True industrial strategy requires coherence, transparency, and a commitment to long-term national interest, not short-term revenue schemes. This should be embraced by Democrats and Republicans alike. Otherwise, we risk becoming the very system we claim to oppose: one where power is transactional, markets are politicized, and innovation is hostage to the whims of the state.
David Nevins is co-publisher of The Fulcrum and co-founder and board chairman of the Bridge Alliance Education Fund.
Generative AI is advancing at breakneck speed. Already, it’s outperforming doctors on national medical exams and in making difficult diagnoses. Microsoft recently reported that its latest AI system correctly diagnosed complex medical cases 85.5% of the time, compared to just 20% for physicians. OpenAI’s newly released GPT-5 model goes further still, delivering its most accurate and responsive performance yet on health-related queries.
As GenAI tools double in power annually, two distinct approaches are emerging for how they might help patients.
One path involves FDA-approved tools built by startups and established technology companies. The other empowers patients to safely use existing tools like ChatGPT, Gemini, and Claude.
Each path has advantages and tradeoffs. Both are likely to shape healthcare’s future.
To better understand what’s at stake, it’s first helpful to examine how generative AI differs from the FDA-approved technologies used in medicine today.
Narrow AI
Medicine has relied on “narrow AI” applications for more than two decades, using models trained to complete specific tasks with structured clinical data.
These tools are programmed to compare two data sets, identify subtle differences, and assign a precise probability factor to each. In radiology, for example, narrow AI models have been trained on thousands of mammograms to distinguish between those demonstrating early-stage breast cancer and those with benign conditions like fibrocystic disease. These tools can detect differences too subtle for the human eye, resulting in up to 20% greater diagnostic accuracy than doctors working alone.
Because narrow AI systems produce consistent, repeatable results, they fit neatly within the FDA’s current regulatory framework. Approval requires measurable data quality, algorithmic transparency, and reproducibility of outcomes.
Generative AI: A new kind of medical expertise
Generative AI models are built differently. Rather than being trained on structured datasets for specific tasks, they learn from the near-totality of internet-accessible content, including thousands of medical textbooks, academic journals, and real-world clinical data.
This breadth allows GenAI tools to answer virtually any medical question. But the large language model responses vary based on how users frame questions, prompt the model, and follow up for clarification. That variability makes it impossible for the FDA to evaluate the accuracy and quality of the tools.
Two distinct pathways are emerging to bring generative AI into clinical practice. Maximizing their impact will require the government to change how it evaluates and supports technological innovation.
1. The traditional path: FDA-approved, venture-backed
As medical costs rise and patient outcomes stagnate, private technology companies are racing to develop FDA-approved generative AI tools that can help with diagnosis, treatment, and disease management.
This approach mirrors the narrow AI model: high-priced tools that are highly regulated and largely dependent on insurance coverage for American families to afford them.
With venture funding, companies can fine-tune open-source foundation models (like DeepSeek or Meta’s LLaMA) using a process called “distillation.” This involves extracting domain-specific knowledge and retraining the model with real-world clinical experiences, such as tens of thousands of X-rays (including radiologists’ readings) or anonymized transcripts of patient-provider conversations.
Consider how this approach might impact diabetes management. Today, fewer than half of patients achieve adequate disease control. The consequences include hundreds of thousands of preventable heart attacks, kidney failures, and limb amputations each year. A generative AI tool trained specifically for diabetes could replicate the approach of a skilled chronic disease nurse: asking the right questions, interpreting patient data, and offering personalized guidance to help users better manage their blood sugar levels.
This path already appears to have federal backing. The Trump administration recently launched its Medicare-funded Health Tech Ecosystem initiative, partnering with more than 60 tech and healthcare firms to pilot AI-enabled tools for chronic disease management, including diabetes and obesity.
Although distillation is faster and cheaper than building an AI model from scratch, the timeline to FDA approval could still span several years and cost tens of millions of dollars. And any adverse outcome could expose companies to legal liability.
2. The alternate path: Empowering patients with GenAI expertise
This second model flips the innovation equation. Instead of relying on expensive, FDA-approved tools developed by private tech companies, it empowers patients to use low-cost, publicly available generative AI to manage their own health better. This can be accomplished through digital walkthroughs, printed guides, YouTube videos, or brief in-person sessions.
For example, a patient might input their blood pressure, glucose readings, or new symptoms and receive reliable, evidence-based advice from ChatGPT or Claude: whether a medication change is needed, when to alert their doctor, or if emergency care is warranted. Similarly, patients working with their physicians could use these LLMs to detect early signs of post-operative infection, worsening heart failure, or neurological decline.
With 40% of doctors already engaged in “gig work,” an ample supply of clinicians from every specialty would be available to contribute their expertise to develop these training tools.
This model would bypass the need for costly product development or FDA approval. And because it offers education, not direct medical care. It would create minimal legal liability.
Government support for both models
These approaches are not mutually exclusive. Both have the potential to improve care, reduce costs, and extend access. And both will benefit from targeted government support.
The traditional path will require companies to evaluate the reliability of their tools by testing the accuracy of their recommendations against clinicians. When these tools are equivalent, the FDA would give its approval.
The alternate path of educating patients to use existing large language models will benefit from educational grants and added expertise from agencies like the CDC and NIH, partnering with medical societies to develop, test, and distribute training materials. These public-private efforts would equip patients with the knowledge to use GenAI safely and effectively without waiting years for new products or approvals.
Together, these models offer a safer and more affordable future for American healthcare.
Robert Pearl, the author of “ChatGPT, MD,” teaches at both the Stanford University School of Medicine and the Stanford Graduate School of Business. He is a former CEO of The Permanente Medical Group.
