Josh Tyrangiel: Useful AI starts with a problem, not a rivalry

A journalist points to recycling, tax systems and sepsis care as places where small teams have put AI to work.

JTJosh TyrangielGast · The Atlantic
Kanal
Masters of Scale
Veröffentlicht
19. Nov. 2026
Video
21:54
Lesezeit
6 Min.
Original
Englisch
Auf YouTube öffnen
Beim Scrollen schrumpft das Video in die rechte untere Ecke und läuft weiter. Untertitel folgen Ihrer Lesesprache.
LesespracheIn 2 Sprachen verfügbar
Diesen Artikel gibt es noch nicht auf Deutsch, daher wird das Original (Englisch) angezeigt.
ZUSAMMENFASSUNG

In 30 Sekunden

In East Lansing, an $8,000 trial used cameras on recycling trucks and specific feedback to improve recycling, says journalist Josh Tyrangiel. He argues that people are more likely to accept AI when it solves an obvious problem than when they are asked to take sides in a contest between tech leaders. His reporting also found officials using AI to modernize IRS systems and a hospital testing a sepsis prediction tool. Those efforts, he says, required leadership, funding and attention to the people whose jobs would change.

  1. 1Private launch and satellite networks can give a few people geopolitical powerTyrangiel describes Musk and Bezos pursuing launch, satellites and potentially space-based computing. He says concentrated control of communications can affect national conflicts.
  2. 2Separate the technology from the companies selling itTyrangiel says computer scientist Danny Hillis urged him to look beyond corporate rivalries. That led him to people applying AI to education, health care and government.
  3. 3An IRS modernization effort used AI-assisted programmingTyrangiel says the agency’s technology team began migrating its historical taxpayer records so the system could interact with newer software. The work continued amid political and funding pressures.
  4. 4A small recycling pilot paired computer vision with specific feedbackTyrangiel says East Lansing tested an $8,000 camera system and compared postcards. He reports a roughly 40% improvement in recycling and two complaints across about 12,000 tests.
  5. 5A hospital tested sepsis prediction against patient outcomesTyrangiel says Cleveland Clinic spent around 18 months working on a prediction system and reports a 41% reduction in sepsis mortality. He emphasizes the difficulty of changing practices across a large organization.

Das Wichtigste

  1. Start with a specific problem and the people who understand it, rather than looking for a problem to fit a technology.
  2. Tyrangiel says East Lansing combined camera recognition with targeted feedback in a small recycling trial before expanding.
  3. Budget for the cultural work of modernization, including explaining how employees’ roles might change.
  4. Treat improvements in health care as outcomes to test, not as evidence that an AI tool is perfect.
  5. Tyrangiel argues that concentrated control of space launch and satellites raises public-policy questions.
VOLLTEXT

Den bearbeiteten Text nach Kapiteln lesen

Nur das Transkript
01

Launch and satellites concentrate power

Tyrangiel describes Musk’s and Bezos’s ambitions across launch, communications and computing. He argues that concentrated satellite infrastructure has implications for governments.

Josh Tyrangiel says individuals can now wield influence comparable to states. He argues that people care less about whether a useful service uses AI than whether it works.

Jeff Berman asks about Tyrangiel’s Atlantic reporting on Elon Musk and Jeff Bezos and their competition in space. He asks what motivates decisions affecting communications in Ukraine.

Tyrangiel describes four steps: rocket launches, satellites for surveillance and internet, possible data centers in space or on the moon, and computing power for AI models on Earth. SpaceX and Blue Origin compete in launch, while Starlink already supplies satellite internet; he describes the later steps as theoretical.

He says control of communications can affect conflicts between countries. Tyrangiel recounts former national security adviser Jake Sullivan describing the two wealthy entrepreneurs as unusually powerful actors and discussing the need for competition between them.

Tyrangiel says his reporting indicated that one of two people ordered a shutdown of high-speed defense-capable communications during a Russian attack, while the precise chain of direction was unclear. He presents it as an example of private power intersecting with state decisions.

02

Why launch capacity is difficult to replace

Berman asks whether alternatives can counter the existing satellite network. Tyrangiel discusses limited launch capacity and possible government responses.

Berman asks whether European nations and Canada can find alternatives. Tyrangiel says competing satellite providers still need launch capacity, while SpaceX and Blue Origin also launch satellites for others.

He says outside companies have reportedly seen launch dates pushed out as demand from Musk’s own companies rises. In his view, Europe cannot quickly replace the existing system at scale, and the lack of launch diversity calls for a government response.

Berman observes that controlling each stage is a way to defend a business. Tyrangiel recalls Senator Mark Kelly saying the companies earned their success through years of investment and failed attempts, while arguing that government still has a role in addressing concentrated capacity and captive demand.

Asked what a future president could do, Tyrangiel credits the United States’ early-2000s commercial crew program with helping private space companies advance. He also points to antitrust and historical restrictions on producers controlling distribution as possible policy tools.

Tyrangiel says the challenge is that federal institutions have shown little interest in that level of enforcement. He argues that politicians need to explain why oversight of infrastructure can matter to national security.

Berman points out that a lawsuit could ultimately reach the Supreme Court. Tyrangiel agrees that immediate interventions would be politically difficult, while arguing that more launch providers are needed over the long term.

03

Find the useful applications beyond AI rivalries

Tyrangiel connects public distrust to a lack of clear benefits. He describes seeking domain experts who apply technology to concrete problems.

Turning to his book AI for Good, Tyrangiel says he encountered a public mood of distrust. People were being offered powerful technology without a convincing explanation of how it would help them, sometimes by companies already associated with social media’s harms.

He argues that people have agency both in defining AI uses they do not want and in encouraging those they do. As examples of limits, he mentions privacy and reporting misalignment; he also wants people to create demand for beneficial products.

After ChatGPT appeared, Tyrangiel began reporting an AI column and spoke with technology leaders. He found conversations dominated by money, valuations and rivalry, but wanted to understand what the technology could actually do.

Computer scientist Danny Hillis advised him to separate the technology from the companies. Tyrangiel then sought people who understood education, health care, public institutions and human connection, and who had brought technology to bear on specific problems.

Those applications were neither easy nor a universal cure, he says, but some improved existing processes considerably. He came away wanting to resist harmful uses while encouraging work that serves people.

Asked what surprised him, Tyrangiel says he found competent federal employees who cared about their work and understood the problems they faced. The challenge was matching their knowledge with appropriate tools rather than applying AI indiscriminately.

04

Modernizing the IRS requires people and funding

Tyrangiel describes an AI-assisted migration of tax records. He says change depends on funding and bringing employees along.

He points to the IRS and its individual master file, which holds tax returns and their subsequent changes. Tyrangiel says its technology chief used AI-assisted programming during a migration that made the system more capable of interacting with newer software.

Tyrangiel says the official remained in place and continued the work despite political pressure. Berman asks whether a capable public agency could eventually provide free, optimized tax returns, rather than leaving such services to private companies.

Tyrangiel thinks the technical potential exists but says improvement is not immediate and requires funding. He argues that modernization also depends on explaining the change to employees, addressing job concerns and sustaining support through political conflict.

Tyrangiel describes visiting East Lansing and meeting environmental manager Cliff Walls. Soon after starting, Walls encountered a proposal to use AI in recycling and found roughly $8,000 for a pilot.

05

East Lansing tests recycling feedback

A local official tried cameras on recycling trucks and compared two kinds of messages to residents. Tyrangiel attributes the reported results to a gradual rollout and practical feedback.

A company trained computer vision on images of recyclable materials and contaminants and attached the system to a truck. GPS connected observations to households, allowing the city to compare two kinds of postcards.

One postcard encouraged a resident to download an app; the other included a photograph of the item that had been put out incorrectly. Tyrangiel says the city introduced the trial slowly and did not promise that it would solve every problem.

He reports that the recycling rate improved by roughly 40% and that the team received two complaints across around 12,000 tests. To him, the result illustrates how visible effectiveness can matter more to residents than the label AI.

People he spoke with at the IRS welcomed the example. Tyrangiel argues that applications can gain ground when discussion moves away from personalities and toward concrete improvements, even if the systems are not perfect.

Berman notes that even $8,000 is meaningful in a municipal budget and that the local manager had discretion to try the project. He asks whether similar leadership exists in larger institutions such as a school district or hospital system.

06

Measuring health care against patient outcomes

Tyrangiel describes Cleveland Clinic’s focus on care and costs. He reports its sepsis-prediction project as an example of difficult implementation.

Tyrangiel says it does, and describes Cleveland Clinic’s leadership setting two goals: improve patient outcomes and save money that can be reinvested in care. He says a small number of people able to run projects and persuade colleagues made progress within a very large organization.

He describes sepsis as the body’s response to infection and says the hospital decided to reduce deaths associated with it. The clinic brought in a vendor and spent around a year and a half developing and testing a prediction system.

Not everyone was satisfied with the system, he says, and changing an organization’s culture remained difficult. Tyrangiel reports that sepsis mortality fell by 41% after the project and argues for supporting the people who implement such tools.

Mehr aus diesem Kanal

22:24EN
Masters of Scale· 18. Nov. 2026Podcast · 2 StimmenAmjad Masad: Replit's AI turnaround rests on enterprise trustGast: Amjad Masad · Moderation: Bob SafianReplit's CEO says security and reliability helped turn a stalled coding startup toward business customers as AI agents changed its own workplace.Coding-AgentenAI securityEnterprise-KI+2
2 Folien12:43EN
Masters of Scale· 16. Nov. 2026PodcastBill Gross Gives Away a Million AI-Generated Startup IdeasGross combined his 31-part idea-finding method with millions of documents, and now wants to help founders turn ideas into businesses.EntrepreneurshipStartupsAI agents+1
23:56EN
Masters of Scale· 12. Nov. 2026Podcast · 2 StimmenWalmart’s CEO wants faster change without losing the human connectionGast: John Furner · Moderation: Jeff BermanJohn Furner discusses agent shopping profiles, frontline tools and the leadership work needed to move a large retailer.LeadershipAI AdoptionManagement
21:35EN
Masters of Scale· 11. Nov. 2026Podcast · 2 StimmenEric Ries: Why successful companies lose the purpose that made them workGast: Eric Ries · Moderation: Krista TippettRies argues that pressure from capital can erode a company’s mission, and that trust deserves a place in business decisions.Company BuildingCompany CultureLeadership+1
23:03EN
Masters of Scale· 10. Nov. 2026Podcast · 3 StimmenTubi and e.l.f. grew by serving audiences incumbents overlookedGast: Anjali Sud, Kory Marchisotto · Moderation: Bob SafianAnjali Sud and Kory Marchisotto explain free streaming, affordable beauty and direct feedback from fans.Product StrategyGrowth StrategyDigital Media+1
27:58EN
Masters of Scale· 9. Nov. 2026Podcast · 3 StimmenShould AI agents get your personal data? A debate over trustGast: Meredith Whittaker, Neil Chilson · Moderation: Jeff BermanSignal’s Meredith Whittaker warns that broad agent access can expose others; Neil Chilson argues for limited, verified use.Privacy & data governanceAI securityAI agents+2