{
  "id": 19494737,
  "title": "US Should Help Hospitals and Universities Build AI | Opinion",
  "url": "https://urgent.news/2026/11/11/us-should-help-hospitals-and-universities-build-ai-opinion",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-11T11:56:35.000Z",
  "source": {
    "name": "Newsweek",
    "slug": "newsweek",
    "url": "https://www.newsweek.com/us-should-make-it-easier-hospitals-universities-help-build-ai-12698666"
  },
  "original_language": "en",
  "account": "Institutions like hospitals and universities possess valuable data that could aid researchers in studying rare conditions, yet they often hold those data within their own systems for various reasons. Conversely, universities may have the expertise to lead AI research but lack the necessary computing power to execute it. Both parties should have a means to collaborate in the development of artificial intelligence (AI) under conditions they can verify. Given that public agencies are investing in AI research, it is my belief that they should finance projects enabling institutions to pool computing power and learn from each other's data while maintaining control over their contributions.\n\nAccording to Stanford University's 2026 AI Index, companies have produced over 90% of the notable advanced AI models tracked in 2025. While companies possess the resources to construct some of the most demanding systems, universities and hospitals provide a distinct advantage. They possess knowledge about which questions matter in a laboratory or hospital setting and can assist in testing whether a model functions beyond the environment where it was initially developed. By granting institutions access to the tools for constructing and testing models, they gain more influence over the AI challenges addressed.\n\nA timely opportunity exists to broaden this access. In September, the National Science Foundation unveiled an operations center for its National Artificial Intelligence Research Resource. This center will oversee access to computing power, data, models, and other research tools, providing the agency with a platform to evaluate how institutions can contribute resources as well as utilize them. Universities could allocate time on their research computers when capacity is available, allowing other teams to utilize that time to test models they have already developed. Universities should have the ability to prioritize their own researchers and impose limitations on the types of work accepted from external teams. A joint project should also consider the expenses related to electricity and staff time. If the institution bearing those costs has limited access to the outcomes, it has little incentive to continue participating. An agreement should delineate the benefits each partner receives, such as computing time, access to the resulting model, or a portion of the research funding.\n\nThe same principle applies to data. Consider a scenario where multiple hospitals are investigating a rare complication. Each may have insufficient cases to construct a model independently. With a technique known as federated learning, the hospitals could train the model using records within their own systems. The project would amalgamate updates from each hospital into a shared model while the records remain in their original locations. The hospitals could subsequently assess how well the model performs on their own patients' data. Maintaining records in place addresses one risk. The National Institute of Standards and Technology has documented how an attacker can occasionally recover details about the training data from the updates themselves or from a trained model. One protective measure outlined by NIST involves combining updates in a manner that obscures each institution's contribution from the party responsible for combining them. Projects should undergo testing for leaks during training and again before any external party can utilize the model. Each participant should be aware of who can halt the work if a test discloses an issue. Transparent records would also facilitate dispute resolution. If a team utilizes more computing time than agreed upon, both partners should be able to verify the same record. If a hospital withdraws, the agreement should specify when its systems will cease receiving new work and whether the project may continue using models previously trained. The National Science Foundation should leverage its new center to run a pilot round of shared AI projects, incorporating these safeguards into the funding terms. Each project should document the authorization of the work, the resources employed, and how partners can verify that the work adhered to the agreement. Before expanding the program, the agency should publish information on the costs of the projects, which projects gained access to the necessary computing power or data, and the effectiveness of the safeguards implemented. This provides other institutions with evidence to determine whether joining would benefit their research or public duties. Ultimately, institutions closest to the problems AI could resolve should play an active role in its development and have a voice in how their contributions are utilized.",
  "summary": "Companies have the resources to build the most demanding computer systems. Universities and hospitals bring a different strength.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}