Skip to content

Podcast

Pro-Worker AI

The AI Daily Brief: Artificial Intelligence News and Analysis

Source ↗ ← All highlights
  • Most Doctors Use AI For Admin Work Not Diagnosis
    • 81% of doctors now use AI, mainly for summarizing research, discharge instructions, and documentation rather than diagnosis.
    • The AMA frames this as ‘augmented intelligence,’ with only 17% using AI for assistive diagnosis. Transcript: Nathaniel Whittemore Next up, an interesting statistic from a new survey from the American Medical Association. The survey found that 81% of doctors now use AI in their profession. Leading use cases include keeping up with medical research, generating discharge instructions, and documenting appointments. The AMA first gathered this data in 2023 and have found that usage has more than doubled since then. Said AMA CEO John White, AI has quickly become part of everyday medical practice. Physicians see real promise in its ability to support clinical decisions and cut down on administrative burden. Notably, the AMA has adopted augmented intelligence as their term for AI, hammering home the point that technology isn’t supposed to replace human judgment. And indeed, when you dig into the data, that seems to be how it’s playing out. The leading use cases of AI in medicine are all about summarizing information and aiding with administrative work. Assistive diagnosis was the only use case that comes close to the actual practice of medicine, and only 17% of doctors said they were using AI in this way. Finally today, some new comments from Sam Altman, speaking at a BlackRock conference on Wednesday. (Time 0:05:43)
  • Create Employer Led Training And Stackable Credentials
    • Build a modern transition system where employers define needed skills and governments fund short, stackable credentials and apprenticeships.
    • Gina Raimondo recommends employer-led training, tax incentives for on-the-job training, and temporary wage insurance for transitions. Transcript: Nathaniel Whittemore Government invests in the training, incentives, and safety nets that help workers move quickly into them. The private sector has always been better positioned to see which new jobs are emerging, which skills matter, and how quickly demand will shift. So this new bargain should start with businesses taking the lead and providing real-time AI-powered insights into hiring plans, technology adoption, and skill needs. Now, from there, she goes into a number of other pieces of what she thinks would be a better overall framework. One area that she wants to see is better coordination between education and employers. She says, The future of higher education should be modular, and employers must be active partners in shaping what gets taught. The country needs to shift focus from long and expensive degrees that risk obsolescence before completion towards short, affordable job-linked credits that offer on-ramps from Education to work. People should be encouraged to pursue credentials that can stand alone or be stacked over time into degrees, bringing people back to campus over the arc of their lives. She gives the example of a mid-career accountant who doesn’t need another master’s degree. Instead, Raimondo writes, she may be better off with a four-month credential in temporary wage insurance that bridges any pay gap and incentivizes her to accept a new role sooner. She also calls for new ways for higher education to be funded, for a modernized apprenticeship system of employer-led training, and incentives for the private sector to do this. (Time 0:16:58)
  • Efficiency Versus Opportunity AI Determines Job Outcomes
    • Distinguish Efficiency AI from Opportunity AI: efficiency reduces headcount, opportunity expands output and new roles.
    • Incentives that push firms to reinvest AI savings into job-creating activities can shift outcomes toward opportunity AI. Transcript: Nathaniel Whittemore If you’ve heard me speak of this before, Efficiency AI is the idea of using AI to do the same with less, which is of course going to be at the root of most of these job cuts. Opportunity AI is seeing the potential for AI to allow you to produce more of whatever it is you produce or to bridge into new areas. Capitalism is of course inherently expanding, and so it is inevitable that in the long run, organizations that view AI as opportunity expanding will be the ones who win. Why I’m interested in incentives to reinvest AI-driven savings into the creation of jobs is that it creates an incentive for employers to not stop at the edge of efficiency AI and instead To jump into that framework of opportunity AI. Now that’s something that I’d like to explore in much more depth at some point, but let’s wrap up with Raimondo’s piece. (Time 0:18:38)
  • AI That Expands Work Instead Of Replacing It
    • Not all AI is pure automation; some classes expand work by creating new tasks or leveling expertise.
    • The MIT paper defines five tech categories and argues new-task and expertise-leveling AIs can be pro-worker by extending judgment and enabling new roles. Transcript: Nathaniel Whittemore In short, they argue that there are different categories of technological change, with various types of impact on human employment. In the abstract, they write, While AI’s capacity to automate work is substantial, we argue that its potential to serve as a collaborator by extending human judgment, enabling new Tasks, and accelerating skill acquisition is equally transformative and currently underexploited. The paper breaks technological change into a taxonomy of five categories. They evaluate each of those categories across three dimensions. Labor productivity, the value of human expertise, and labor share of national income. The five categories are one, labor augmenting technologies, two, capital augmenting technologies, 3. Automation technologies, 4. New-task technologies, 5. Expertise-leveling technologies. In each of these examples, they increase labor productivity. However, when it comes to the value of human expertise, there can be wide differences among them. Take, for example, the difference between automation technologies, where existing expertise is made obsolete, versus new task-creating technologies where new expertise is needed. Now, in their framework, the only unambiguously pro-worker category is new task-creating technologies. The one that I think would see the most debate among smart people is the ambiguous pro-worker designation of expertise-leveling technologies, which they call ambiguous because While new entrants benefit, incumbents’ expertise is potentially devalued. Whether democratization of expertise is a good thing or not could get into some thorny debates. But their point here is to make it clear that not all technology change is the same, and that when it comes to AI, although we assume that it’s all automation technologies, that’s just Not actually the case. They give a few examples of pro-worker AI in the field. One example is an electrician’s assistant, where an electrician uses LLMs to support electricians in troubleshooting electronic machinery. Workers can upload photos and diagnostic data, and an AI matches to a database of prior problems. In practice, this halved the average time for completing maintenance reports, and they categorize it as pro-worker because the worker remains in the loop modifying AI recommendations, Being collaborative, not subservient. Other examples they give are a service worker’s assistant, a teacher’s AI aid, hearing aids for Chinese gig delivery workers, and patent examiner decision support. And yet they say these cases are too rare right now. Their main argument is that the market is at the moment not capitalizing on pro-worker AI opportunities. They argue that the current AI focus is overwhelmingly on task automation and AGI development, neither of which coheres with their pro-worker definition. There are a couple of reasons they argue this is happening. Misaligned firm incentives, like managers using automation as a way to reduce dependence on unionized labor, rent dissipation, i.e. Managers wanting to redistribute savings to shareholders, and what the authors call the AGI bet, basically firms that believe AGI is imminent that see little point in investing in Pro-worker technologies. To wit, why build tools to enhance workers if workers will be fully replaceable shortly? They also see misaligned developer incentives. Some of those build off of the misaligned firm incentives, i.e. Customer demand shapes supply. If firms prefer buying AI automation, tech companies will prioritize building automation tools. It’s self-reinforcing. There’s also a time horizon problem. Pro-worker technologies might require years of investment while automation solutions are already market-ready. There’s also the potential for worker resistance. Workers themselves may resist pro-worker AI tools that require them to acquire new expertise and adjust work habits. If workers lack foundational skills or are reluctant to invest, then firms are further discouraged. From there, the authors give nine different policy directions that they believe could move the needle further in the direction of pro-worker AI. This, I think, is the area where most people would have debate, but I’m appreciative of the authors actually laying out some potential paths forward rather than just identifying the Problem. One category of remediations they recommend is, for example, for the government to leverage their huge GDP footprint in areas like healthcare and education to use market incentives To drive developers to develop pro-worker AI. They have a bunch of other ideas as well around the tax code, antitrust, etc. But the point is that whether these are the right directions or not, there are opportunities to try to drive towards more pro-worker AI. Lastly, and maybe most importantly, the paper pushes back on an idea which seems almost by default accepted in AI discourse, that automation is the dominant force in economic history. However, if automation were the whole story, the authors argue, labor’s share should have been declining relentlessly since the Industrial Revolution. But it hasn’t. In fact, it rose during the first eight decades of the 20th century. What’s more, they point out, rich, heavily automated countries have higher labor shares than poor, less automated countries. This is the opposite of what the Automation Road’s labor thesis predicts. The explanation is that new task creation counterbalances automation, which is a fancy way of saying that the creation in creative destruction does eventually kick in, and the jobs That go away are replaced by new other types of jobs. Which is not to say that we should just let the process happen on its own. In fact, the whole reason they’re writing the paper is to get more people engaged and explicitly trying to push towards a pro-worker AI paradigm. (Time 0:21:44)
  • Electrician Assistant Cut Report Time In Half
    • An electrician’s assistant halved maintenance report time by matching photos and diagnostics to past problems while the electrician stayed in the loop.
    • The tool is labeled pro-worker because technicians modify AI recommendations rather than being replaced. Transcript: Nathaniel Whittemore One example is an electrician’s assistant, where an electrician uses LLMs to support electricians in troubleshooting electronic machinery. Workers can upload photos and diagnostic data, and an AI matches to a database of prior problems. In practice, this halved the average time for completing maintenance reports, and they categorize it as pro-worker because the worker remains in the loop modifying AI recommendations, Being collaborative, not subservient. (Time 0:23:26)
  • Market Forces Make Pro-Worker AI Unpopular Today
    • Market incentives currently favor automation and AGI bets, leaving pro-worker tools underinvested.
    • Reasons include short firm horizons, rent capture to shareholders, developer alignment with buyer demand, and perceived imminence of AGI. Transcript: Nathaniel Whittemore They argue that the current AI focus is overwhelmingly on task automation and AGI development, neither of which coheres with their pro-worker definition. There are a couple of reasons they argue this is happening. Misaligned firm incentives, like managers using automation as a way to reduce dependence on unionized labor, rent dissipation, i.e. Managers wanting to redistribute savings to shareholders, and what the authors call the AGI bet, basically firms that believe AGI is imminent that see little point in investing in Pro-worker technologies. To wit, why build tools to enhance workers if workers will be fully replaceable shortly? They also see misaligned developer incentives. Some of those build off of the misaligned firm incentives, i.e. Customer demand shapes supply. If firms prefer buying AI automation, tech companies will prioritize building automation tools. It’s self-reinforcing. There’s also a time horizon problem. Pro-worker technologies might require years of investment while automation solutions are already market-ready. There’s also the potential for worker resistance. Workers themselves may resist pro-worker AI tools that require them to acquire new expertise and adjust work habits. If workers lack foundational skills or are reluctant to invest, then firms are further discouraged. (Time 0:24:06)
  • Use Public Contracts To Drive Pro-Worker AI Demand
    • Use government procurement and policy levers to steer developers toward pro-worker AI by rewarding tools that augment workers.
    • The paper suggests leveraging large public sectors like healthcare and education to create demand for pro-worker solutions. Transcript: Nathaniel Whittemore This, I think, is the area where most people would have debate, but I’m appreciative of the authors actually laying out some potential paths forward rather than just identifying the Problem. One category of remediations they recommend is, for example, for the government to leverage their huge GDP footprint in areas like healthcare and education to use market incentives To drive developers to develop pro-worker AI. They have a bunch of other ideas as well around the tax code, antitrust, etc. But the point is that whether these are the right directions or not, there are opportunities to try to drive towards more pro-worker AI. (Time 0:25:21)
  • History Shows Automation Often Creates New Jobs
    • Historical data contradicts automation-only narratives: labor’s share rose in early 20th century and is higher in rich automated countries.
    • The authors argue new-task creation offsets automation, so policy can influence whether gains favor labor or capital. Transcript: Nathaniel Whittemore Lastly, and maybe most importantly, the paper pushes back on an idea which seems almost by default accepted in AI discourse, that automation is the dominant force in economic history. However, if automation were the whole story, the authors argue, labor’s share should have been declining relentlessly since the Industrial Revolution. But it hasn’t. In fact, it rose during the first eight decades of the 20th century. What’s more, they point out, rich, heavily automated countries have higher labor shares than poor, less automated countries. This is the opposite of what the Automation Road’s labor thesis predicts. The explanation is that new task creation counterbalances automation, which is a fancy way of saying that the creation in creative destruction does eventually kick in, and the jobs That go away are replaced by new other types of jobs. (Time 0:25:53)
  • Agent Builders Will Become Core Knowledge Roles
    • New hybrid roles will emerge by combining domain expertise with agent-building and orchestration skills.
    • Whittemore predicts ‘agent builders’ and ‘agent orchestrators’ will become dominant flavors of knowledge work across industries. Transcript: Nathaniel Whittemore Think about it this way. Take any job that exists right now, any knowledge worker job, and make it have a baby with a software engineer. And then give that child who doesn’t know what they don’t know yet the awareness of what one parent does with the coding skills of the other. What comes out is kind of the new role. Effectively, we have agent builders and agent orchestrators in every flavor of the knowledge worker rainbow. And this will increasingly be an incredibly important role. In fact, flavors of AI engineer may become the dominant role. This is something that Leighton Space has been talking about a lot recently. (Time 0:27:01)