Showing posts with label society. Show all posts
Showing posts with label society. Show all posts

The turbulent AI era is here. The choices we make now are critical

Mike's Notes

Another opinion piece on the impact of AI. This time from Bill Gates.

I'm publishing a range of opinions from different writers on this blog.

My Opinion

My opinion is that AI is a good thing, despite its weaknesses and the need for much more work to fix its problems. However, the financial numbers don't stack up for the money being spent. It seems like a speculative bubble, and someone is going to go broke.

Gary Marcus

Gary Marcus wrote this recently:

Excellent new essay by Bill Gates, echoing many of the themes I have been writing about here and in my 2024 book Taming Silicon Valley:

    • The critical, lasting importance of the decisions we make now.
    • The risks that AI creates, around bioterrorism, deepfakes, disinformation, cyberattacks, and so on, empowering criminals such that “Even criminals with very limited skills will be able to target victims at every scale”. He also argues (though I did not anticipate this in the earlier book) that “AI could stunt our kids’ development and replace human relationships”.
    • The urgent need for—and lack of—a systematic plan for how society will address AI.
    • The importance of having civil society, and not just government and tech companies, centrally involved in devising that plan. (I would also emphasize the critical need for including independent scientists, which Gates does not make explicit).
    • The importance of “creating a domestic and international framework for dealing with AI”.
    • The need for tax structures and incentives to keep humans employed, and for “rebalanc[ing] how we tax labor and capital.”

...

Resources

References

  • Taming Silicon Valley: How We Can Ensure That AI Works for Us, Gary Marcus, MIT Press. 2024.
  • Magnifica humanitas. Encyclical Letter of His Holiness,  Pope Leo XIV on safeguarding the human person in the time of artificial intelligence. Holy See. 2026.

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Gates Notes
  • Home > Ajabbi Research > Library > Subscriptions > Marcus on AI
  • Home > Handbook > 

Last Updated

09/09/2026

The turbulent AI era is here. The choices we make now are critical

By: Bill Gates
Gates Notes: 27/08/2026

William Henry Gates III is an American businessman and philanthropist. A pioneer of the microcomputer revolution of the 1970s and 1980s, he co-founded the software company Microsoft in 1975 with his childhood friend Paul Allen.

...

We need a plan to ensure that the good outweighs the bad.

...

What you need to know

The transition to the AI era will be one of the most turbulent times in human history. Right now, we are not preparing adequately for that transition. If the world takes the right steps, AI will be a force for good and leave everyone better off.

During my entire life I’ve only had two jobs. In the first one, I played a role in developing software to empower people through my work at Microsoft.

In my second one, which I started full time in 2008, I am giving back the wealth I made at Microsoft with the goal of making the world a healthier, better educated, and more equitable place. This is the job I will have for the rest of my life.

Both of these experiences inform my perspective on artificial intelligence. When I first learned about computers at age 13 I was fascinated by the idea of making them more intelligent and able to perform things that, at the time, only humans could do. Although the term “AI” was used from around the time I was born, the technology has only made significant progress in the last decade. It is now incredibly capable and it is continuing to improve at a mind-blowing rate. AI for the first time can replace and even exceed human cognition.

AI will either be the greatest equalizer ever invented, or the worst source of injustice.

In terms of equity, AI will either be the greatest equalizer ever invented, or the worst source of injustice. The challenge is monumental. Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history. How will we use this technology to make the world a fairer place and keep it from widening the divide between rich and poor? How will we protect the people who are most vulnerable to the harms caused by artificial intelligence, including those who lose their livelihoods and the sense that they are in control of their future?

I believe that answering these questions and acting on the answers should be the world’s top priority. If the world takes the right steps AI will be a force for good and leave everyone better off.

Unfortunately, right now we are not preparing for it. I don’t see evidence that leaders, experts, and communities are confronting the challenges adequately. There is no plan to ease the entry into the AI era.

Part of the reason for this is that many commentators underestimate the extent of the impact AI will have. I think there are a few reasons why.

One is the fact that AI models still make mistakes. It is hard to envision any of them replacing human cognition when, not long ago, they couldn’t solve a simple Sudoku puzzle or figure out how many R’s are in the word strawberry.

But the reliability problem is being fixed quickly, as researchers create models that can check their own work and improve themselves. Soon they will be substantially better than humans at many tasks.

Another reason people underestimate AI is that analogies to the effects of past innovations are misleading. We have no experience with a technology that can be adopted quickly or that can think and move like a human. When the PC came along, it took twenty years to significantly change how we worked because the software had to be developed, the price had to come down, and people had to learn how to use the tools and incorporate them into their business processes. AI, on the other hand, runs on the devices we already have, and it uses natural language. We don’t have to adapt to it because it can adapt to us. It can watch the same training video that is used to train human workers and learn from existing data.

I want to acknowledge a potential bias. I have benefited enormously from the technology industry. Although I have diversified my portfolio quite a bit, I still have financial ties to it. I am working with Microsoft and other AI companies in my role as chairman of the Gates Foundation to try and ensure AI is deployed in ways that will truly benefit people around the world.

However, my views on AI are not motivated by the potential to make money for myself. Any profits generated by my investments, including those related to technology, will go to the Gates Foundation to tackle global inequity. Of course, readers will have to decide for themselves whether this clouds my view.

This time really is different.

For as long as I can remember, I’ve wished innovation could happen faster. With AI, my feelings are more complicated.

We need time to prepare for the social, political, and economic upheaval.

I wish the world could get the benefits rapidly and delay the problems it will cause as long as possible, but the benefits and problems are arriving at the same time. I believe we need time to prepare for the period of social, political, and economic upheaval we are about to enter. The people who need the most time are the ones who have the least—the accounting worker who’s replaced by a bot or the $20-an-hour worker who loses their job to a $10-an-hour robot.

Many observers say that this technology transition will be like previous ones. They give the example of how jobs in the United States shifted from agriculture to office work. However, that proceeded over several generations and created new jobs where human cognition was required. In this case, the technology can substitute for human cognition.

Because it can see, listen, speak, and reason and will eventually do physical work just as smoothly as any human, it will not just affect one sector. AI will take on work in law, customer service, medicine, software, and manufacturing. It will hit these industries rapidly, over the course of a decade rather than a few generations. There will be some new jobs, but without the right policies there will be far fewer than exist today.

If someone had a credible plan for slowing down AI advances globally, I would likely support it. However, I don’t think that’s going to happen. The geopolitical and economic incentives are pushing too hard to go full speed ahead.

To make sure we maximize the positive effects of this unprecedented technology and minimize the bad so we are better off overall, we need to understand both the benefits and the risks. I’ll start with the risks.

The transition to AI comes with three big risks.

I plan to write about each of these in more detail in the future, so I’ll touch briefly on them for now.

Many jobs will disappear forever.

In 1933, during the Great Depression, unemployment in the United States was roughly 25 percent. It remained in double digits for much of the following decade. It ultimately recovered as demand, investment, and growth returned.

AI may not reach this level, but its impact will not go away with an economic cycle. The jobs at most risk are entry- and mid-level, and the new jobs being created will mostly require skills that take many years to learn.

White-collar jobs are already being hit modestly. After the widespread adoption of generative AI, employment fell significantly among young workers in jobs that are especially vulnerable to replacement, but not among their older colleagues.

I think this trend will continue, but it will not be confined to a handful of industries or occupations. Jobs in sales and customer support (online and over the phone), software engineering, and paralegal work may be among the first affected, but the disruption will reach much further as AI takes on tasks that today still require trained workers: things like assessing loan applications, doing data analysis, and even triaging patients. A few areas like software engineering will generate new demand as the costs go down, so the net job loss in those areas will be less than in others as long as some tasks, such as design, are better done by humans.

Blue-collar jobs will be affected as well. Although robots are not as far along as AI, eventually their cost will be dramatically lower too. Many Americans I talk to don’t realize how fast dexterous robots are advancing because much of the advanced work is being done in other countries, primarily China. Or they may be confused by those videos of robots dancing badly that have been going viral lately. I think “smart” robots will begin to compete with people on some physical tasks—in the construction and hospitality industries, for example—by the end of the decade.

Robots and AI combined can create a vicious cycle. After one company adopts them and uses the savings to lower its prices, its competitors will feel immense pressure to do the same. If existing companies don’t adopt them, then start-ups will. Many people will shift to other jobs, but the turmoil of losing work, getting retrained, and finding other work will be significant. Market forces will make adoption go faster and faster and, unless we intervene, there will be fewer good jobs available and the benefits will accrue to a small group.

I’m especially worried about young people, who will enter a workforce with fewer entry-level openings. They understand the challenge because they are the most active users of AI and see both the capabilities and the rate of improvement. It’s no wonder that so many of them feel negatively about AI.

The biggest shift for workers will happen when AI provides nearly error-free work. At that point, it will be able to function on its own without a human checking in on it, and companies will have every economic incentive to let it.

This will lead to a fundamental change in how we think about work, income, and economic security. How will an economy that’s been built around employment operate if fewer people are working, or if many people are working fewer hours?

In a capitalist society, employment is the way most people get the money they need to pay for the basics of life as well as being a key source of dignity and social connection.

When a community has high unemployment, the ripple effects can be pervasive. Research suggests that in some parts of the United States, factory closures contribute to a rise in deaths from opioid overdoses. Now imagine similar pressures on both white-collar and blue-collar workers nationwide.

We have to think now about how to reduce job losses so that everyone can share in the prosperity that AI creates. Waiting until people are already displaced or underemployed will be too late. AI is a structural challenge to the way our economy is organized, and it requires thinking and action now.

AI will empower people (and perhaps AIs) to do more harm.

Long before AI entered the mainstream, there was information online about how to create weapons like bombs, bioweapons, even computer viruses. AI will make it much easier to not only get this information but act on it. Even criminals with very limited skills will be able to target victims at every scale: individuals, companies, and governments.

Even criminals with very limited skills will be able to target victims at every scale.

AI-enabled fraud, disinformation, deepfakes, and surveillance are the harms that many people will feel most keenly in their everyday lives.

AI capabilities are starting to be used for cyberattacks. The smartest cybersecurity experts I know are scared about the next few years, because the attackers are getting powerful new capabilities faster than the defenders can fix all the weaknesses. After all, the same AI model that can find a flaw in software so a company can fix it can also help a criminal exploit it. The resources needed to make an attack are going down significantly and we haven’t been able to separate those abilities from benign usage.

Think about the infrastructure that will be vulnerable: hospitals, financial institutions, water systems, power grids, systems for managing government benefits. When these institutions are attacked, it’s the patients, customers, and benefits recipients who stand to lose.

The same goes for bioterrorism. Although AI will lead to lifesaving advances in drugs and vaccines, it will also make it easier to design a deadly new disease. Again, the positive capabilities are hard to separate from the dangerous ones. This is a global problem.

The risks I’ve just mentioned are all about how AI will empower bad actors who have relatively little power now. The same tools will also concentrate power in places where it already exists. Autonomous weapons, for example, will make governments even more capable of using deadly force without a human being part of the decision. Monitoring and manipulating public opinion will be easier and cheaper, and more effective too.

Eventually, the power to use AI to harm people will not be limited to people or institutions. AI systems themselves already occasionally act in ways their designers didn’t intend. The technology is improving faster than anyone expected and in surprising ways, and as the models become more powerful, they could begin to act against our interests and we could lose control. I’ll have more to say about this in the future.

AI models could begin to act against our interests and we could lose control.

AI could stunt our kids’ development and replace human relationships.

When I was growing up in Seattle, I didn’t have that many friends aside from a few other boys who were like me. It took hard work and a lot of help from my mom to develop my social skills so I could relate to different kinds of people. I still draw on those lessons today at the age of 70.

I doubt I would have put in the same work if I had had an AI companion back then. They talk to you in ways you’re already comfortable with. They don’t push you outside your comfort zone. They are always available and never get mad at you. This gives them the potential to become highly addictive and to rob us of the lessons we learn from connecting with other people.

The body of evidence on this subject is still small and a bit mixed, but there are signs that we should be very concerned. For example, in one study of more than 1,100 people who use AI companions, researchers at Stanford and Carnegie Mellon found that those with smaller social networks were the most likely to turn to a chatbot for companionship. And the heavier and more emotionally personal that use became, the worse they felt.

Young people could be affected for their entire lives. In his book The Anxious Generation, Jonathan Haidt makes an observation about the effect of social media that is even more true for AI: “Like young trees exposed to wind, children who are routinely exposed to small risks grow up to become adults who can handle much larger risks without panicking. Conversely, children who are raised in a protected greenhouse sometimes become incapacitated by anxiety before they reach maturity.”

An AI companion designed to never upset you is a big, protected greenhouse.

We are only beginning to understand the dangers that the internet—especially social media—can pose to young people’s development. We’re seeing compulsive use, disrupted sleep, cyberbullying, and exposure to harmful content. AI could magnify many of these risks by making them more persuasive and difficult to escape, and we should not wait another generation to start taking them seriously. Countries including Australia, the United Kingdom, and Norway are adopting protections for children online. China has gone the furthest. Its rules restrict AI companion apps broadly, bar designs that foster emotional dependence, and ban virtual relatives and romantic partners for minors.

The same tool that will allow people to learn more than ever could also lead to many people learning less.

I’m also worried about AI’s impact on education. Ironically, the same tool that will allow people to learn more than ever could also lead to many people learning less. One preliminary survey suggested that heavier AI use was associated with less critical thinking. The effect was stronger for younger people.

This would be the worst possible time for humans to lose their critical thinking skills. In an era of deepfakes and misinformation that can be tailored to you individually, the ability to tell what is true from what is not becomes an essential life skill.

It’s unclear where to draw the line on these psychosocial problems. In some cases, AI may help people understand how to do better in their human relationships. It may be the only contact with the outside world for isolated elderly people and people with limited mobility, and it will be better than nothing. Wherever we end up drawing the line, it should be our decision, made intentionally.

The good things we do with AI could be very, very good.

It’s often said that we overestimate how much will change in the short term and underestimate how much will change in the long term.

With AI, I see something different going on. Some people see only the upside of AI and do not focus enough on the negatives. Others make the opposite mistake, which is to focus exclusively on the dangers—which are real—at the cost of missing the potential benefits.

We need both: deep concern about the AI harms we need to minimize, and grounded optimism about the positives if we maximize them for everyone.

Maximizing the benefits is just as important as minimizing the harms. If people see how AI makes their lives easier, it will help build the public trust that is necessary for managing the harder parts of the transition. If the first thing AI does in most people’s lives is take away their job, those who are already skeptical about it will outright reject it. This will make it harder to ever deliver on the benefits and it is another reason why governments, industries including the medical industry, and AI companies should be working together now.

With its ability to synthesize knowledge from every scientific field, AI can accelerate innovation in the world’s toughest technical challenges: providing reliable clean energy for everyone, combating climate change, growing enough food, eradicating diseases, and more. Researchers working on cancer treatments or nuclear energy can use AI to search through massive amounts of scientific literature. It can help them identify patterns that a human might miss and decide which experiments offer the most promise. When intelligence is no longer the limiting factor that it is today, smaller companies will be able to compete with organizations that have far larger research budgets. R&D and innovation will be supercharged.

Healthcare is one area where AI can help solve real-world problems. Many small American hospitals lack on-site specialists who can quickly diagnose a patient during a life-threatening emergency. In those places, AI could make sure a heart attack is caught in time and a family avoids the crushing expense of a medical emergency. Viz.ai is one example. It analyzes scans to detect strokes and other emergencies and helps medical teams coordinate their patients’ care. It is being used in nearly 2,000 U.S. hospitals.

AI will also help primary-care doctors make better diagnoses and keep in touch with their patients when they’re not in the clinic. It will help patients understand test results and complicated schedules for taking their medicine.

Agriculture is where I see the fastest impact of AI in low-income countries.

I surprise a lot of people when I tell them that a second area—agriculture—is where I see the fastest impact of AI in low-income countries. In most low-income countries, farmers don’t get reliable weather forecasts or advice on what seeds to plant, how to protect their crops and livestock from disease, or how to improve their soil. With population growth in these countries and the challenges of climate change, these farmers need more help than ever. Using AI, low-income farmers will soon be able to get better advice about all these things than even the richest farmers get today and increase their output substantially.

Government services are a third area where AI can make people’s lives easier. In the United States, I’ve met families who, understandably, were overwhelmed by the process of applying for health insurance, student aid, or food assistance. Faced with a huge stack of complicated bureaucratic forms, many felt like giving up. AI can streamline things dramatically so they get the help they need faster and the government can operate more efficiently. Governments can make the citizen’s experience far better, starting with those who need its safety net services the most.

Despite my concerns about its impact on our mental health, I think AI can also help a lot there. Most communities have too few counselors, psychiatrists, and addiction specialists. With the right privacy safeguards in place, AI tools could help people recognize warning signs. Then, if needed, they can offer evidence-based coping strategies and team up with a human to provide more responsive treatment.

AI can be a boon for education as well, despite the concerns I mentioned earlier. It can free teachers up to spend more time working with students one on one or in small groups and give them a clearer view of where the whole class is struggling. For students, an AI tool that preserves what researchers call “productive struggle”—the cognitive work that builds understanding—can strengthen learning. When a student first encounters a new idea, the AI gives substantive explanations and offers both questions and answers. Later, when it’s checking their comprehension, it holds the answer back and helps them arrive at it on their own.

Taken together, the advances in all these areas could make everyday life easier, more affordable, and less constrained by a person’s income or connections.

AI could give individuals and small businesses access to capabilities that today require expensive professional help or large staffs, while making products and services better and cheaper. It could help people with disabilities live more independently and enable workers and entrepreneurs with good ideas to accomplish far more than they can today.

Most importantly, it could give people back some of the time and attention now consumed by paperwork, bureaucracy, searching for reliable information, and tasks they cannot afford to pay someone else to handle. These benefits may seem modest, but multiplied across millions of lives, they would be profound: more people getting good advice when they need it and having greater freedom to focus on the lives they want to build.

We have to be deliberate about ensuring that it benefits everyone and not just a wealthy few.

In all these areas, the operative word is “can”—AI can improve life for people at every income level. But it won’t do that automatically. As with any new technology, we have to be deliberate about ensuring that it benefits everyone and not just a wealthy few. This will require governments and philanthropy to play a strong role so that less wealthy citizens and low-income countries are full beneficiaries.

The Gates Foundation has 19 years left of the 20 years in which it will spend its remaining $200 billion. AI will help it achieve its ambitious goals by both accelerating the discovery of vaccines and medicines for HIV, TB, malaria, and malnutrition and helping the healthcare workforce and patients know how to use those tools. The foundation’s goals include cutting the number of children who die every year in half again, as was done from 2000 to 2024. All of our work, not just health but also agriculture and education, will take full advantage of AI.

I will write much more about these efforts next month in the foundation’s annual Goalkeepers report—including our focus on making sure that AI models are available in the languages spoken by people in all the countries where we support work, and not just the ones that are common in rich and middle-income countries. Many of the leading AI companies, including OpenAI, Anthropic, Google, and Microsoft, are partnering with the foundation on all of these initiatives, which is making a big difference.

The world needs a plan.

It is great that some AI companies are proposing solutions to challenges raised by their own technology, but we should not expect them to lead the charge. Some of the issues are outside their area of expertise, and in a democratic society it’s not their role to decide these things.

Instead, solutions should be developed through a public democratic process that includes elected officials, policymakers, educators, health workers, local officials, and community leaders. Millions of people will have their lives disrupted, and we’ll need a stronger, more flexible social safety net to help them manage the transition. Local communities are already raising concerns about the energy and water needed for data centers. Without solutions, some groups will push for stopping AI development and deployment altogether.

The solutions should be shaped by our answers to the profound questions raised by AI, including how we preserve our humanity in a time when machines can out-think us. As people who spend their lives thinking about what it means to be human, religious leaders can play a key role in this. I was fascinated by Pope Leo XIV’s encyclical on AI, “On Safeguarding the Human Person in the Time of Artificial Intelligence.” It lays a strong foundation for the work that needs to be done.

In the coming months, I will share more ideas for making sure that AI’s benefits outweigh the harm it causes. Here are three to start, beginning with what I think is the most important one.

Build a new system for managing the transition.

The highest priority is a monumental task: creating a domestic and international framework for dealing with AI.

None of our current institutions were designed to handle a technology that spreads so fast and touches so many parts of our lives. So we’ll need to make new ones.

It’s hard to overstate what an enormous undertaking this will be. After the attacks of 9/11, the U.S. government went through its biggest reorganization since World War II for the purpose of improving just one function, national security.

AI will require much, much more. It will affect national security as well as employment, education, taxation, energy, elections, air and water, public health, the financial system, law enforcement, transportation, public lands, and IT systems.

These sectors overlap in ways our existing bureaucracy is not designed to manage. A labor department may understand workforce disruption but not security risk. A business regulator may understand market concentration but not AI’s effects on children and teenagers. Left to themselves, institutions will see only one part of the system, while the consequences of AI will ripple across the entire system.

At the national level, countries will need bodies that can set priorities across government agencies. The goal will be to make sure that every risk is accounted for. Otherwise, an AI-enabled attack might succeed because no one thought it was their job to stop it.

But even a country that gets its own house in order will still be exposed to risks that cross borders. This is why an international organization will need to be built in parallel.

It will be unlike any other institution we have ever created, though it can follow the model of some existing systems. There’s an inspections regime for nuclear weapons, regulations for international aviation, and agreements that protect the ozone layer. A new global organization for AI will need elements of all three and more.

It is fair to wonder whether the world’s institutions are up to the task of designing and implementing this new architecture. Government moves slowly when it moves at all, and polarization within and between countries makes it harder than ever to get things done. Some cooperation between the U.S. and China will be required.

We do not have the luxury of moving slowly. The place to start is with a process for building the right institutions before the disruption forces governments into crisis mode. National leaders should convene economists, technologists, labor experts, business leaders, and workers themselves regularly to identify where existing institutions are failing and what new authorities may be needed. Countries will need to learn from each other.

And the countries that host the leading AI developers and control critical parts of the supply chain should begin meeting now to set up shared norms, before competitive pressure makes it harder for them to cooperate.

Building the framework I’m talking about will take years, which is why we need to start now.

Set aside some jobs for humans.

My dad died of Alzheimer’s in 2020. In the later stages of his illness, he was cared for day and night by paid caregivers who understood him even when he struggled to express himself. He couldn’t always tell them when he was hungry, but they always knew.

My family and I will always be grateful to that amazing group of professionals. Something in the care they gave my dad was irreplaceably human. No robot could or should have done it.

I think about that team when the question of which jobs will disappear and which will remain comes up. I believe that as AI and robots improve, we’ll set aside certain things for only people to do. I’ve started calling this domain Human Reserved, and it’s an example of the kinds of ideas we’ll need to consider.

I like the phrase Human Reserved because it makes me think of nature reserves—places where we could put buildings and roads, but we choose not to because the loss would be too great.

We might set something aside as Human Reserved for economic reasons. For example, we may do it because allowing machines to take over a certain role will displace a large number of people who can’t easily change jobs. You can’t tell a 55-year-old who has worked in construction their whole career that they need to go work at an elder care facility and expect them to find it fulfilling.

Sometimes the decision to make something Human Reserved will be driven by other factors. In health, for example, imagine a robot giving you the awful news that you have an incurable disease. There’s no technical reason why it couldn’t. Yet it shouldn’t.

The Human Reserved domain will evolve over time—for example, we should consider setting aside some jobs now and phasing in AI slowly over years or decades with a commitment to preserve some jobs. Some areas, like education and mental health care, will be a mix, with a human in charge who’s using the technology to extend what they can do.

The lines will also vary from place to place. Some countries might insist on having humans take care of the elderly. But a country like Japan, which has a shrinking workforce and not enough young people to care for the old, may welcome a caregiving robot.

The idea of Human Reserved raises a host of questions I don’t have answers to. Who gets to decide what we reserve for humans? What criteria should we use? How do you keep companies from cheating and using robots anyway? What happens to international trade when one country lets robots make something and another country doesn’t? These will need to be worked out in public as part of the transition plan.

Rebalance how we tax labor and capital.

As workers are pushed into different jobs, they will need retraining and other support from the social safety net. But they will be working less, which means they will be paying less in income taxes, and government revenues will drop just when the demand for those services is greatest. The funds will have to come from somewhere at a time when budgets are stretched.

I believe we should tax AI tokens and robots. Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings. But if you buy a robot, you can usually write it off right away as a business expense. The tax system nudges you toward replacing people with machines.

A tax would slow the rush away from human labor a little and raise money for retraining and a stronger safety net. It would need to be targeted so it does not slow down the purely beneficial uses of AI, like making medicine and education cheaper.

Critics of this idea point out that it’s not optimally efficient in an economic sense, but they’re not considering the broader value of work for individuals and society. And with all the accelerated innovation we will have, we’ll be able to afford a little inefficiency as the price for keeping people employed.

I proposed a robot tax years ago and most of the reaction was that it was a strange idea. I’m still a big proponent of it. Although it is not the whole solution to the threat of AI, it is part of a wise response.

However we raise money for more assistance, it needs to reach the people who need it most, including workers who lose their jobs to AI and robots, people whose hours or wages decline, and communities where the losses are concentrated. We need to start doing that work now so that the systems are ready when the need becomes acute.

What I’m doing.

I will use my voice and time to get AI and equity higher on the public agenda. I will raise the issue with lawmakers every time I visit Washington, D.C., and when I meet with leaders around the world. It will be front and center in my conversations with the people who are developing AI models. I will advocate for the national and international framework I described earlier. The Gates Foundation will help drive beneficial usage, including in Africa. Breakthrough Energy, a company I founded, will use AI to help companies develop cheap clean energy and help solve the climate problem. I will also be writing about AI on a regular basis.

My message to leaders is:

You have a chance to act now, before unemployment rises sharply, communities are hurting, and public trust has eroded. You can make sure that your government handles the problem holistically, rather than divvying it up into multiple bureaucratic fiefdoms. You can make sure AI benefits everyone. And you can work with other governments to meet this national and global challenge.

Finally, I will try to widen the circle of people shaping this debate. It should include workers, college students who are about to enter the workforce, community leaders, religious leaders and faith-based organizations, parents, educators, and others whose voices often aren’t heard but who have insight into how the transition will affect people’s lives.

How do we ensure that the benefits of AI reach people who do not already have wealth, influence, and access?

How do we strengthen the social safety net and help workers and communities thrive even when they’re displaced?

How should public institutions adapt?

And how do we preserve our humanity through all of this?

This unprecedented technology demands an unprecedented global response.

This unprecedented technology demands an unprecedented global response. If we get it right, the payoff for humanity will be phenomenal and the world will be a more equitable place.

I rarely stop thinking about AI—not because I have all the answers, but because the questions it raises are too consequential to leave to a small group of technologists. Leaders across academia, business, government, and civil society all have a role to play in shaping what comes next.

The luxury of statistics

Mike's Notes

A great article by David Court lays out a fundamental truth about the business of filmmaking.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Compton
  • Home > Handbook > 

Last Updated

08/05/2026

The luxury of statistics

By: David Court
Compton: 03/05/2026

Founder of Compton School, the business school for creative people: ‘The movie industry is a crucible where money and talent are refined down to their essence. It has so much to teach us about business, creativity and human nature.’

Why filmmakers can never see eye-to-eye with movie studios.

When filmmakers and studio executives come together there is always a great display of bonhomie.

Hands get shook and backs get slapped. As though they were bound together in their mutual enterprise: the movie business. Perhaps there’s some argy-bargy about profit shares but basically they are on the same page.

Yet the truth is: there is a fundamental disjunct between filmmaker and executive.


It’s not just that one is on salary while the other is eating their savings. Nor is it the power imbalance between them, or the different pathways they have followed to arrive where they are. It’s that one is operating at 30,000 feet and the other is dug in at ground level.

If you’re making a film, it occupies your entire field of view. You can’t see past it. You don’t have time or bandwidth to think about anyone else’s film. You are all in on one thing — the film you are making.

This is true financially too. You have put all your chips down on one square and you can’t take them back. Whereas the studio executive is managing a slate of films. They’ve got chips on 20 squares.

Harvard professor Mihir Desai calls this the core of finance. He starts his book The Wisdom of Finance with a story about chance and pattern. Chance is what the world looks like close up: an arena of luck, accidents, flukes, coincidences, or what the ancients called fortuna.

Whereas pattern is chance viewed over time – statistically. It’s what we notice when we study the world and keep records. It’s the view from 30,000 feet.

‘Finance, ultimately, is a set of tools for understanding how to address a risky, uncertain world’ – Mihir Desai

The study of patterns is what gave us the insurance business (the averaging of bad luck), portfolio theory in finance (strategic diversification) and the studio slate (the search for a hit). These are all methods of managing risk, of taming fortuna.

But our filmmaker has wandered into a casino and bet everything on single spin of the wheel. Their risk is irreducible. They don’t have the luxury of statistics.

So for all the backslapping, the studio executive and the filmmaker are really two different species. Different stakes, different aims, different view of the world.

For our filmmaker, chance is the whole point. The risk is not to be managed but to be taken.

This post is the first in a series I plan to publish based on my reading – the insights I’ve gathered and think worth sharing.

The American Dream needs a factory reset

Mike's Notes

An interesting solution to the housing problem. Shows what is possible if there were the will.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Rational Optimist Society
  • Home > Handbook > 

Last Updated

19/02/2026

The American Dream needs a factory reset

By: Stephen McBride
Rational Optimist Society: 18/01/2026

Rational Optimist Society - Founder.

Housing’s “iPhone moment”

In today’s diary:

  • Expensive homes = fewer new families
  • Where most housing innovators fail
  • How Tesla’s “tent hack” solves housing
  • Factories that update like an iPhone
  • Build your home like The Sims—in 30 days

Dear Rational Optimist,

I have a friend back in Ireland named Zach.

Zach is a mechanic with his own business. He grinds and does everything right. Yet every night he walks past the main house and goes to sleep in a log cabin in his girlfriend’s parents’ backyard.

Now his girlfriend is pregnant with their first child. They’re about to bring a new life into the world, and they don’t have a place to put the crib.

I have another friend in New York City, the founder of a promising, high-growth startup. He’s crushing it but lives in a Manhattan one-bedroom apartment with his wife and daughter. They want another baby but can’t afford the extra bedroom.

Studies show rising housing costs explain roughly half of the fertility decline in America between 2000 and 2020.

That’s millions of kids who were never born because the rent was too high.

Having kids is a vote of confidence in the future. It’s the ultimate act of optimism. We’re pricing people out of that optimism.

We know the solution: Build. More. Housing. Yet we’re building homes today slower than we did in 1971.

Over the last decade venture capitalists incinerated billions of dollars betting on startups that promised to fix housing. They all failed.

But I think I’ve finally found…

The Henry Ford of housing.

Before we meet this company, let’s visit the graveyard of failed housing startups. There are many headstones.

A few years ago housing disruptor Katerra raised $2 billion to build “gigafactories” for homes. It wanted to mass-produce homes on an assembly line like iPhones, ship them nationwide, and snap them together on-site.

We build cars in factories, so why not houses? It sounded inevitable.

Katerra went bust in 2021. It stepped on two booby traps which killed almost every would-be housing disruptor.

Booby trap No. 1: shipping air.

When you build finished rooms in a factory and ship them, you’re shipping floors, walls, and ceilings. You’re essentially paying to ship empty boxes of air.

This eats up every penny saved by the factory efficiency.

Booby trap No. 2: cash incinerator.

Giant factories cost a lot of money to build. To make the math work they must run at near-full capacity.

But housing is boom-and-bust. When the market dips (it always does), the factory doesn’t stop costing money and turns into a cash incinerator. Katerra built cathedrals of manufacturing requiring perfect economic weather to survive. When it started raining, it drowned.

Then there’s the deadliest booby trap of all.

The US government tried to build houses in factories…

1969 was a great year for the optimists. America put a man on the moon and Concorde flew its first supersonic voyage.

It was also the year of Operation Breakthrough, the US government’s experiment to industrialize housing. The goal was to fund mass-producible building systems and construct 25,000 modern homes.

It was a total disaster. They built fewer than 3,000 units before shutting down.

Uncle Sam failed to account for local governments.

Factories work when they build the exact same thing, over and over. You can’t do that with homes because there are 26,000 towns and cities with different building codes.

It’s like trying to mass-produce a car, but every town has its own rules about where the brake pedal and steering wheel should go.

We’ve been trying to build houses like cars for a century. But houses aren’t cars. They’re legal projects, financial products, and custom assemblies rolled into one.

Malcom’s box

Malcom McLean was a North Carolina truck driver with an idea.

Create steel containers that could neatly and quickly stack on ships, trains, and trucks. The shipping container was born!

On a spring morning in 1956, McLean’s refitted oil tanker left Newark carrying 58 identical steel boxes. That was the day global trade got rewired.

The cost of shipping fell by more than 90% over the following years. Those steel boxes became the universal language of trade, easily swappable across ships, trains, and trucks. Every global brand you know—Nike, Walmart, Apple—owes its business model to McLean’s steel box enabling global trade.

What McLean did for shipping, Cuby Technologies is doing for housing.

“What’s in the box?”

If you’re a Dune superfan like me, you know the scene.

"What's in the box" Dune quote image

Ask that question to Cuby co-founder Aleks Gampel and he won’t respond “pain.” He’ll say, “Everything you need to build a new home in just 30 days.”

Cuby doesn’t build houses. It builds the factories that build houses.

It took an entire automotive-grade production line—robotics, CNC machines, welding stations—and packed it into approximately 122 shipping containers.

Cuby’s product is the Mobile Micro-Factory (MMFTM). It’s a standardized, portable factory that turns homebuilding into a predictable manufacturing process.

When Tesla hit “production hell” in Fremont, it couldn’t get permission to build a new facility fast enough. So Elon put up a massive tent in the parking lot. Because it was a “temporary structure,” he bypassed the zoning nightmare and saved the company.

Cuby takes Tesla’s tent hack to the next level:


Cuby lean-tos image
Source: Cuby

If you build a factory, you need permits and years of approvals. Cuby figured out how to snap 122 shipping containers together and be classified as one giant “machine.”

This hack allows Cuby to stand up an MMF, capable of pumping out 200 homes per year, in just 30 days.

MMFs are compact enough to slot into a mall parking lot. You inflate a massive, pressurized dome. Inside the dome the shipping containers open up to become a fully functioning housing factory:


Cuby housing factory interior image
Source: Cuby

Cuby’s other co-founder, Aleh Kandrashou, walked me (virtually) through its test facility in Eastern Europe to see how an MMF works.

Cuby broke the construction process down into 35 different departments. Walk past one container and inside is a dedicated welding robot fusing steel foundations. Move to the next container, and it’s a specialized paint booth coating the exterior panels.

The containers snap together to form a conveyor belt that takes raw materials—steel coils, glass, resin—and spits out a complete “kit of parts” to build a home:


Cuby conveyor belt image
Source: Cuby

Every stud, pipe, wire, and floorboard needed for a specific house is flat-packed.

Cuby = affordable homes.

Cuby’s target cost is $100 to $110 per square foot. That’s far cheaper than traditional builders that spend $150 to $300+ per square foot depending on location.

Aleks stressed to me Cuby is relentlessly focused on costs: “Tesla launched with the expensive Roadster to fund the cheap Model 3. You can’t do that in housing. If you are a Roadster on day 1, you die.”

“If Jesus came back today the only job he’d recognize is a…”

Carpenter. That was Aleh’s humorous way to describe the lack of innovation in housing.

It’s not for lack of trying. As I mentioned, startups have been trying to disrupt housing for a century.

Cuby has “last-mover advantage.” It designed the MMF specifically to disarm the three booby traps that killed its predecessors.

Shipping air.

Katerra built big whole rooms and shipped them to the site. Cuby ships the factory to near where the house will be built.

Cash incinerator.

A Cuby factory costs 10% as much as a normal factory. It only needs to build 70 homes a year to make money.

Best of all, it’s mobile. If the housing market in Phoenix cools, you can pack the 122 containers and move them to Dallas, where demand is hot.

Cuby doesn’t build homes. It builds the factory that builds the home, which is another safety buffer. It enters into joint ventures with local developers that put up the $10 million to build the factory. Cuby doesn’t deploy the machine until the demand is guaranteed.

Regulatory camouflage.

Cuby’s factories produce a kit of parts that follows International Building Code specifications. With small tweaks they are compliant with America’s 26,000 jurisdictions.

Aleh told me its first US test home in Michigan had zero permitting issues. The home was built under 60 working days at 30% to 40% below local contractor quotes!

To a building inspector, a Cuby home looks like a normal house, just built with unusually high precision:


Cuby home example image
Source: Cuby

Cuby is basically…

A software company wrapped in steel

As an ROS member you know all about the physical innovation famine.

For 50 years progress was trapped in a narrow cone of software, apps and the web. That’s why your phone is a supercomputer, but your house is still built like it’s 1925.

Now that cone is widening into the physical world. Cuby manually mapped out the 10,000 steps required to build a house from scratch. It filmed every process, wrote code for every action, and built it into a system called “FactoryOS.”

This is Cuby’s secret sauce. It’s LEGO instructions on steroids.

FactoryOS spits out 3D instructions for every single screw in the house. It’s built on Unreal Engine, the same video game engine used for Fortnite. These digital guides allow even an idiot like me who struggles to assemble an IKEA desk to build a house.

The software also acts as a relentless quality control manager. For example, it won’t let a worker move to the next step until the AI visually confirms the last step is perfect.

There’s a reason I call Cuby “the Henry Ford for homes.”

Before Ford pioneered the assembly line, building cars relied heavily on highly skilled craftsmen. Ford’s innovations simplified the process and drastically reduced build time. Cuby’s software does the same for homes.

Its digitally guided microtask system atomizes assembly. Four workers (in two shifts) can go from foundation through finishes in roughly 45–60 days. Cuby plans to drive this under 30 days.

We need to talk about toilet paper

Cuby clocked 1 million engineering hours designing its Mobile Micro-Factories, kit of parts, and software. That obsession shows up in strange places, like the bathroom.

When Cuby ships MMF extension units to a site (which are like self-contained command centers equipped with Starlink, workstations, lockers, hot showers, and every tool the crew needs), it packs the exact number of toilet paper rolls needed for four workers for the specific duration of the build.

That precision planning defines Cuby. If a worker finishes a task but a specific wrench isn’t back in its slot, the AI recognizes it. The software won’t let him finish his working day until he finds it. No delays due to missing tools:


Cuby workspace image
Source: Cuby

To avoid “shipping air,” the software calculates the volume of empty space in a container down to the cubic inch. It will delay ordering small parts (like door hinges) until they can perfectly fill the gaps in a shipment of larger materials. Tetris for supply chains.

But my favorite feature is how the factories improve themselves.

We all know about Tesla’s over-the-air updates. Back in 2017 when Hurricane Irma was hurtling toward Florida, Tesla remotely unlocked extra battery range for owners fleeing the storm. With the flick of a switch, the car got better.

Cuby does the same for factories. If a crew in Nevada finds a faster way to install a window, that process update is pushed to every Cuby factory worldwide instantly. Factories now update like your iPhone.

Ultimately the only thing that matters is: Can Cuby build homes faster and cheaper?

Yes. Labor accounts for roughly 70% of the cost of building a home, depending on location. Cuby’s FactoryOS aims to slash that by over 80%.

Today a traditional construction crew burns about 450 minutes of human sweat to finish a single square foot of a house. Cuby does the exact same work in 50 minutes.

A traditional builder needs over 15,000 hours of labor to go from foundation to move-in ready for a standard 2,000-square-foot family home. Cuby crosses the same finish line in just 1,659 hours. It’s building the same house with one-ninth the human effort.

This allows Cuby to pump out more affordable homes while not compromising on quality. Its houses come with steel framing and triple-pane windows, typically luxuries in the US.

On the desert outskirts of Las Vegas…

Cuby’s first US Mobile Micro-Factory is going up. It’s scheduled to pump out homes this fall for a local developer building 3,300 units. I can’t wait to visit.

But one factory won’t solve the housing crisis.

That’s why Cuby stood up a “papa factory” in China. This is the machine that builds the machines. Its job is to mass-produce the 122-container Mobile Micro-Factories.

Next year the papa factory will pump out four MMFs. The year after, 8 to 12. The exponential curve is starting now.

Talk to Aleh for five minutes and you realize he’s a serial inventor. He walked me through a dozen patented technologies, from the “magnetic skin” that lets you swap a home’s exterior like a phone case, to the pressurized factory dome that inflates like a tennis bubble.

And Cuby’s ultimate invention is, to quote Aleks, “a universal manufacturing engine that scales to whatever the world needs next. We’re already working on military barracks, data centers, and contractor garages.”

Housing is arguably the most broken industry in the world, with tough competition from healthcare and education. It’s a gigantic market that affects us all.

High housing costs mean fewer kids. It also warps politics as people feel locked out. Just look at NYC voting in a communist mayor!

If Cuby wins, the payoff is civilization-scale. I asked Aleks and Aleh for their vision:

“A 25-year-old schoolteacher in North Carolina no longer spends her weekends touring open houses she can’t afford. She opens an app to design her house like she’s playing The Sims.

You can drag and drop rooms, see the exact cost update in real-time, push a button to see available plots and finally click order. The MMF gets to work, and she moves in one month later.”

Aleks ended with: “We want to build more homes than anyone else on earth.”

If Cuby succeeds it has a shot at rebuilding the American Dream.

What future would you build if you could have a cheap, custom house by next month? Let me know in the comments below. And remember to click “like” and “restack” to help us spread rational optimism.

—Stephen McBride

Mountains of Evidence

Mike's Notes

Another excellent article from After Babel. I agree 100% that kids shouldn't have social media. It's causing a mental health epidemic.

Update 01/09/2026

From Peter H. Diamandis. The growing threat of AI sex companions to the emotional development of teenagers. China has just banned them. Good.

"If you’re the parent of an adolescent or young adult, stop what you are doing and read this. The most intimate relationship your kid may form this year might not be with a person. It might be with a machine. And the government of China just decided that is dangerous enough to ban." - Peter H Diamandis

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > After Babel
  • Home > Ajabbi Research > Library > Subscriptions > Megatrends
  • Home > Handbook > 

Last Updated

01/09/2026

Mountains of Evidence

By: Jon Haidt and Zach Rausch
After Babel: 15/01/2026

.

Two new projects catalogue research on social media’s many harms to adolescents. Some of the strongest evidence comes from Meta.

Much of the confusion in the debate over whether social media1 is harming young people can be cleared away by distinguishing two different questions, only one of which needs an urgent answer:

The historical trends question: Was the spread of social media in the early 2010s (as smartphones were widely adopted) a major contributing cause of the big increases in adolescent depression, anxiety, and self-harm that began in the U.S. and many other Western countries soon afterward?

The product safety question: Is social media safe today for children and adolescents? When used in the ordinary way (which is now five hours a day), does this consumer product expose young people to unreasonable levels of risk and harm?

Social scientists are actively debating the historical trends question — we raised it in Chapter 1 of The Anxious Generation — but that’s not the one that matters to parents and legislators. They face decisions today and they need an answer to the product safety question. They want to know if social media is a reasonably safe consumer product, or if they should keep their kids (or all kids) away from it until they reach a certain age (as Australia is doing).

Social scientists have been debating this question intensively since 2017. That’s when Jean Twenge suggested an answer to both questions in her provocative article in The Atlantic: “Have Smartphones Destroyed a Generation?” In it, she showed a historical correlation: adolescent behavior changed and their mental health collapsed just at the point in time when they traded in their flip phones for smartphones with always-available social media. She also showed a correlation relevant to the product safety question: The kids who spend the most time on screens (especially for social media) are the ones with the worst mental health. She concluded that “it’s not an exaggeration to describe iGen [Gen Z] as being on the brink of the worst mental-health crisis in decades. Much of this deterioration can be traced to their phones.”

Twenge’s work was met with strong criticism from some social scientists whose main objection was that correlation does not prove causation (for both the historical correlation, and the product safety correlation). The fact that heavy users of social media are more depressed than light users doesn’t prove that social media caused the depression. Perhaps depressed people are more lonely, so they rely on Instagram more for social contact? Or perhaps there’s some third variable (such as neglectful parenting) that causes both?

Since 2017, that argument has been made by nearly all researchers who are dismissive about the harms of social media. Mark Zuckerberg used the argument himself in his 2024 testimony before the U.S. Senate. Under questioning by Senator Jon Osoff, he granted that the use of social media correlates with poor mental health but asserted that “there’s a difference between correlation and causation.”

In the last few years, however, a flood of new research has altered the landscape of the debate, in two ways. First, there is now a lot more work revealing a wide range of direct harms caused by social media that extends beyond mental health (e.g., cyberbullying, sextortion, and exposure to algorithmically amplified content promoting suicide, eating-disorders, and self-harm). These direct harms are not correlations; they are harms reported by millions of young people each year. Second, recent research — including experiments conducted by Meta itself — provides increasingly strong causal evidence linking heavy social media use to depression, anxiety, and other internalizing disorders. (We refer to these as indirect harms because they appear over time rather than right away).

[IMG]

Source: Shutterstock

Together, these findings allow us to answer the product safety question clearly: No, social media is not safe for children and adolescents. The evidence is abundant, varied, and damning. We have gathered it and organized it in two related projects which we invite you to read:

  • A review paper, in press as part of the World Happiness Report 2026, in which we treat the product safety question as a mock civil-court case and organize the available research into seven lines of evidence. The first three lines reveal widespread direct harm to adolescents around the world. Lines four through seven reveal compelling evidence that social media substantially increases the risk of anxiety and depression, and that reducing social media use leads to improvements in mental health. Taken together, these lines of evidence provide a firm answer to the product safety question.
  • MetasInternalResearch.org, a new website that catalogues 31 internal studies carried out by Meta Inc. The studies were leaked by whistleblowers or made public through litigation — despite Meta’s intentions to keep them hidden. The most incriminating among them: an experiment designed to establish causality, where Meta’s researchers concluded that social media causes harm to mental health.

In the rest of this post we present the Tables of Contents from these two projects, so that you can jump into the projects wherever you like and see for yourself the many kinds of research demonstrating harm to adolescents. After that, we return to the historical trends question to suggest an answer. We show that the scale of harm we found while answering the product safety question is so vast, affecting tens of millions of adolescents across many Western nations, that it suggests (though does not prove) that the global spread of social media in the early 2010s probably was a major contributor to the international decline of youth mental health in the following years. We suggested this in Chapter 1 of The Anxious Generation. The two mountains of evidence we present here make that suggestion even more plausible today.

The Review Paper: Seven Lines of Evidence

The World Happiness Report (WHR) is a UN-backed annual ranking that has become the global reference point for national well-being research. It draws on Gallup World Poll data from more than 150 countries. We were invited to write a chapter for the upcoming WHR on the 2026 theme: the association between social media and well-being. Following their 2024 report, which documented a widespread decline of well being among young people, this year they ask whether social media’s global spread in the 2010s was a major contributor to that decline. Our chapter, “Social Media is Harming Young People at a Scale Large Enough to Cause Changes at the Population Level,” offers an answer to the product safety question — no — and to the historical trends question — yes.

The editors graciously allowed us to post our peer-reviewed chapter online before the March 19 publication date so that discussion and debate on this topic can begin immediately.

We structured the chapter as if we were filing a legal brief offering 15 exhibits organized into seven separate lines of evidence. The first three lines are the equivalent of testimony from witnesses in a trial. If the people who had the clearest view of an event say that Person A punched Person B, that would count as evidence of Person A’s guilt. The evidence is not definitive — the witnesses could be mistaken or lying — but it is legitimate and relevant evidence. Here’s the structure of that part of the chapter:

After establishing that the most knowledgeable witnesses perceive harm from social media, we move on to the four major lines of academic research. While most researchers agree that correlational studies find statistically significant associations between social media use and measures of anxiety and depression, and that social media reduction experiments find some benefits for mental health, the debate centers on whether the effects are large enough to matter.2 We show that the experimental effects and risk elevations are larger than is often implied — in fact, they are as large as many public health effects that our society takes very seriously (such as the impact of child maltreatment on the prospective risk of depression.)3

Furthermore, we take a magnifying glass to some widely cited studies that claim to show only trivial associations or effects between social media use and harm to adolescents (e.g., Hancock et al. (2022) and Ferguson (2024). We show that these studies actually reveal much larger associations when the most theoretically central relationships are examined — for example, when you focus the analysis on heavy social media use (rather than blending together all digital tech) linked specifically to depression or anxiety (rather than blending together all well-being outcomes) for adolescent girls (rather than blending in boys and adults).

Meta’s Internal Research: Seven More Lines of Evidence

Throughout 2025, a variety of lawsuits against social media companies were progressing through the courts. In the briefs posted online by various state Attorneys General, we found references to dozens of studies that Meta had conducted. Some of this information had been available to the general public since 2021, when whistleblower Frances Haugen brought out thousands of screenshots of presentations and emails from her time working at Meta. Others were newly found by litigators in the process of discovery.4

The descriptions of these studies are scattered across multiple legal briefs, most of which are hundreds of pages long, so it has been difficult to keep track of them — until now. We have collected all publicly available information about the studies in one central repository, MetasInternalResearch.org. Indexed in this way, the scattered reports form a mountain of evidence that social media is not safe for children. The evidence was collected and hidden by Meta itself.

We found information on 31 studies related to the product safety question that Meta conducted between 2018 and 2024. Meta has long hired PhD researchers, particularly psychologists, to conduct internal research projects. (In January 2020, Jon met with members of this team and shared his concerns about what Instagram was doing to girls.) Meta’s researchers have access to vast troves of data on billions of users, including what exactly users saw and what emotions or behaviors they showed afterward. (This is known as “user-behavioral log data.”) Academic researchers never get access to rich data like this; they must devise their own surveys, which obtain a few crude proxy variables (such as “how many hours a day do you spend on social media?” and “How anxious were you yesterday?”). So we should pay attention to what Meta’s researchers found and how they interpreted their findings.

In one example, recently unsealed court documents from lawsuits brought by U.S. school districts against Meta and other platforms reveal that Meta conducted its own randomized control trial (considered to be the best way to study causal impact) in 2019 with the marketing research firm Nielsen. The project — code-named Project Mercury — asked a group of users to deactivate their Facebook and Instagram accounts for one month. According to the filings, Meta described the design of their study as being “of much higher quality” than the existing literature and that this study was “one of our first causal approaches to understand the impact that Facebook has on people’s lives… Everyone involved in the project has a PhD.” In pilot tests of the study, researchers found that “people who stopped using Facebook for a week reported lower feelings of depression, anxiety, loneliness, and social comparison.” One Meta researcher also stated that “the Nielsen study does show causal impact on social comparison.”

In other words, Meta’s own research on the effects of social media reduction confirms those from academic researchers that we report in Line 6 of our review paper. Both sets of researchers find evidence of causation, not mere correlation.

We were impressed by the great variety of methods that Meta’s researchers used. In fact, the 31 studies we located fit neatly into seven lines that are similar to the seven lines we used in our review paper. The findings from Meta researchers are highly consistent with the findings from academic researchers, which gives us even more confidence in our conclusions about the product safety question.

Here’s the Table of Contents. Once again, after the introductory material, we present three lines of testimony:

We then move on to lines 4, 5, and 6, which correspond exactly to lines 4, 5, and 6 in the review paper: correlational, longitudinal, and experimental studies, although line 7 is unique. (It involves reviews of academic literature conducted by Meta’s researchers.)

Returning to the Historical Trends Question

The product safety question is distinct from the historical trends question. A consumer product (e.g., a toy or food) can be unsafe for children without it producing an immediate or easily detectable increase in national rates of a particular illness.5

But social media is an unusual consumer product because of its vast user base and the enormous amount of time it takes from most users. It’s as if a new candy bar, intentionally designed to be addictive, was introduced in 2012 and, within a few years, 90% of the world’s children were consuming ten of these candy bars each day, which reduced their consumption of all other foods. Might there be increases in national rates of adolescent obesity and diabetes?

In our WHR review paper, we estimate the scale of direct harms (e.g., cyberbullying, sextortion, and exposure to disturbing content) and indirect harms (e.g., elevated risks of depression, anxiety, and eating disorders). We then show that these estimates are likely underestimates because they don’t account for network effects inherent to social media, nor the heightened impact of heavy use during the sensitive developmental period of puberty. All told, the number of affected children and adolescents likely reaches into the hundreds of millions, globally.

Once we consider the vast scale at which social media operates — used by the large majority of young people, for many hours each day, over many years, and across nearly all Western nations — it becomes clear that social media companies are harming young people on an industrial scale. It becomes far more plausible that this consumer product caused national levels of adolescent depression and anxiety to rise, especially for girls.

Conclusion: What Now?

Academic debates over media effects often take decades to resolve. We expect that this one will continue for many years. But parents and policymakers cannot wait for resolution; they must make decisions now, based on the available evidence. The evidence we have collected shows clearly that social media is not safe for adolescents.

We believe that the evidence of direct and indirect harm that we have collected in these two complementary projects is now sufficient to justify the sort of action that the Australian government took in 2025 when it raised the age for opening or maintaining a social media account to 16. Just as the recent international trend of removing smartphones from schools is beginning to produce educational benefits, the research we reviewed suggests that removing social media from childhood and early adolescence is likely to produce a great variety of benefits, including lower rates of depression and many fewer victims of direct harms such as sexual harassment and sextortion.

Countries around the world ran a giant uncontrolled experiment on their own children in the 2010s by giving them smartphones and social media accounts at young ages. The evidence is in: the experiment has harmed them. It is time to call it off.

  1. By “social media” we mean platforms that include user profiles, user-generated content, networking, interactivity, and (in most cases) algorithmically curated content. Platforms such as Instagram, Snapchat, TikTok, Facebook, YouTube, Reddit, and X all share these features. This means that ordinary use includes interacting with adult strangers.
  2. For examples of studies showing substantial risk elevations, see Kelly et al. (2019), Riehm (2019), Twenge et al. (2022), and Grund (2025). For examples of meaningful experimental effects, see Burnell et al. (2025).
  3. Burnell et al. (2025) report an average effect of roughly g = 0.22 (about one-fifth of a standard deviation) for “well-being” outcomes in sustained social-media-reduction studies. Grummitt et al. (2024) estimate that the increased risk of depression and anxiety attributable to childhood maltreatment corresponds to effects of d = 0.22 and d = 0.25, respectively. See section “Indirect Harms to Millions” for more details.
  4. We note that this is our only source of this information because Meta lobbies against legislation that requires them to share data with researchers, such as the Platform Accountability and Transparency Act.
  5. The trend of any particular harm may of course have several major influences, some of which may counteract each other. This can add considerable complexity to the historical trends question.