AI and the crucial decisions that lie ahead, according to Bill Gates

I have only ever had two jobs in my life. In the first, I helped develop software with the aim of empowering people to achieve more, through my work at Microsoft. In the second, which I have been doing full-time since 2008, I am giving back the wealth I accumulated at Microsoft with the aim of making the world a healthier, more educated and fairer place. This is the work I will be doing for the rest of my life. Both of these experiences influence the way I view artificial intelligence

From the perspective of fairness, AI will either be the greatest tool for equality ever invented, or the worst source of injustice. The challenge is enormous. Even under the best circumstances, the transition to this new era of artificial intelligence will be one of the most turbulent periods in human history. How will we use this technology to make the world a fairer place, whilst preventing it from widening the gap between rich and poor even further? How will we protect the most vulnerable people from the harm caused by artificial intelligence, including those who will lose their livelihoods and their sense of control over their own futures?

I believe that answering these questions and acting accordingly should be the world’s top priority. If we make the right choices, AI will be a positive force and will make everyone’s lives better. Unfortunately, we are not currently preparing for this. I see no evidence that leaders, experts and communities are addressing these challenges adequately. There is no plan to guide us into the age of AI. (…)

As far as I can remember, I have always wanted innovation to move faster. With AI, however, my feelings are more complex. I wish the world could reap the benefits quickly and delay as long as possible the problems this technology will cause. 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. Those who need time the most are those with the fewest resources: the administrative clerk replaced by a bot, or the worker earning $20 an hour who loses their job to a robot that costs $10 an hour.

Many observers argue that this technological transition will be similar to previous ones. They cite, for example, the shift in the United States from an economy in which many people worked in agriculture to one centred on office work. But that process took place over several generations and created new jobs that required human cognitive skills. In this case, however, technology may replace precisely those skills.

As it can see, hear, speak and reason, and will eventually be able to carry out physical tasks just as naturally as a human being, it will not be of interest to just one sector. To ensure we maximise the positive effects of this unprecedented technology and minimise the negative ones, so that we come out of it better off overall, we must understand both the benefits and the risks. I’ll start with the risks. Because the transition to AI entails three major risks.

In 1933, during the Great Depression, unemployment in the United States stood at around 25 per cent. It remained in double figures for much of the following decade. It eventually fell as demand, investment and growth returned. AI may not lead to a similar level of unemployment, but its impact will not simply disappear with an economic cycle. The jobs most at risk are those at entry-level and intermediate levels, whilst the new jobs that will be created will mostly require skills that take many years to acquire.

White-collar jobs are already feeling the effects of this change, albeit to a limited extent. Following the large-scale adoption of generative artificial intelligence, employment has fallen significantly amongst young workers in sectors particularly vulnerable to automation, but not amongst their older colleagues. This trend is expected to continue, and it will not be confined to a handful of industries or professions. Jobs in sales and customer service—both online and over the phone—as well as in software engineering and paralegal work, could be among the first to be affected. However, the transformation will extend much further as AI takes on tasks that currently still require skilled workers.

Manual labour will also be affected. Although robots are not as advanced as AI, their cost will eventually fall dramatically too. Many Americans I speak to do not realise how quickly robots capable of handling objects with precision are advancing, because much of the more advanced work is carried out in other countries, particularly China. Alternatively, they may be misled by recent viral videos of robots dancing clumsily. I believe that by the end of the decade, ‘intelligent’ robots will begin to compete with people in certain physical jobs, for example in construction and the hospitality sector.

Robots and AI together can create a vicious circle. Once a company has adopted them and used the savings to lower prices, competitors will feel enormous pressure to do the same. If existing businesses do not adopt these technologies, start-ups will. Many people will move into other jobs, but the upheaval caused by job losses, the need to retrain and the search for new employment will be significant. Market forces will drive adoption ever faster and, if we do not intervene, there will be fewer good jobs available and the benefits will end up concentrated in the hands of a small group.

The biggest change for workers will come when AI is able to perform a job with virtually no errors. At that point, it will be able to operate autonomously, without the need for human supervision, and companies will have every financial incentive to let it do so. This will lead to a fundamental shift in the way we think about work, income and economic security. How can an economy built around employment function if fewer people are working, or if many are working fewer hours? In a capitalist society, work is how most people earn the money needed to pay for life’s basic necessities, as well as being an essential source of dignity and social connections.

When a community experiences high unemployment, the knock-on effects can be far-reaching. Research suggests that, in some parts of the United States, factory closures contribute to a rise in opioid overdose deaths. Now imagine similar pressures being exerted simultaneously on white-collar and blue-collar workers across the country.

Long before artificial intelligence became commonplace, information on how to create weapons such as bombs, biological weapons and even computer viruses was already available on the internet. AI will make it much easier not only to obtain this information, but also to put it into practice. Even criminals with very limited skills will be able to target victims at every level: individuals, businesses and governments.

AI-enabled fraud, disinformation, deepfakes and surveillance are the threats that many people will feel most directly in their daily lives. AI capabilities are beginning to be used for cyberattacks. The brightest cybersecurity experts I know are alarmed by what might happen in the coming years, because attackers are acquiring powerful new capabilities faster than defenders can patch all the vulnerabilities.

The risks all relate to the way in which AI will enable malicious actors – who are currently relatively powerless – to cause greater harm. But these same tools will also concentrate power where it already exists. Autonomous weapons, for example, will make governments even more capable of using lethal force without any human involvement in the decision-making process. Monitoring and manipulating public opinion will become easier, less costly and even more effective.

We are only just beginning to understand the dangers that the internet, and social media in particular, can pose to young people’s development. We are seeing compulsive use, sleep disorders, cyberbullying and exposure to harmful content. AI could amplify many of these risks, making them more persuasive and harder to avoid, and we should not wait another generation before we start taking them seriously. Countries including Australia, the United Kingdom and Norway are introducing measures to protect children online. China has gone a step further. Its regulations generally restrict AI companion apps, ban projects designed to foster emotional dependence, and prohibit minors from having virtual relatives or artificial romantic partners.

I am also concerned about the impact of AI on education. Paradoxically, the very tool that will enable people to learn more than ever before could also lead many to learn less. A preliminary study has suggested that greater use of AI is associated with a reduced capacity for critical thinking. The effect was most pronounced among young people. This would be the worst possible time for humans to lose their ability to think critically. In an age of deepfakes and disinformation tailored to each of us, the ability to distinguish between what is true and what is not is becoming an essential life skill.

It is often said that we tend to overestimate how much will change in the short term and underestimate how much will change in the long term. With AI, I see things differently. Some people see only the positive aspects and do not focus enough on the negative ones. Others make the opposite mistake: they focus exclusively on the dangers – which are real – whilst losing sight of the potential benefits.

We need both: a deep concern about the harms caused by AI, which we must minimise, and a realistic optimism regarding its positive effects, if we are to maximise them for everyone. Maximising the benefits is just as important as minimising the harms.

Thanks to its ability to synthesise knowledge from every scientific field, AI can accelerate innovation in the world’s most difficult technical challenges: ensuring clean and reliable energy for all, combating climate change, producing enough food, eradicating diseases and much more. Researchers working on cancer treatments or nuclear energy can use AI to analyse vast amounts of scientific literature. It can help them identify patterns that a human might miss and decide which experiments offer the most promising prospects. When intelligence is no longer the limiting factor that it is today, smaller companies will be able to compete with organisations with much larger research budgets. Research and development, and innovation, will receive a powerful boost.

Healthcare is an area where AI can help solve real-world problems. Many small American hospitals do not have on-site specialists capable of quickly diagnosing a patient during a life-threatening emergency. In such settings, AI could ensure that a heart attack is detected in time and that a family avoids the devastating costs of a medical emergency. Viz.ai is one example. It analyses scans to detect strokes and other emergencies and helps medical teams coordinate patient care. It is used in nearly 2,000 US hospitals. AI will also help GPs make better diagnoses and stay in touch with patients when they are not in the surgery. It will help patients understand test results and complex medication regimens.

Many people are surprised when I tell them that a second sector – agriculture – is where I see the most rapid impact of AI in low-income countries. In most of these countries, farmers do not receive reliable weather forecasts or advice on which seeds to use, how to protect crops and livestock from disease, or how to improve the soil. With growing populations in these countries and the challenges posed by climate change, these farmers need help more than ever. Thanks to AI, farmers in low-income countries will soon be able to receive better advice on all these issues than even the wealthiest farmers have access to today, significantly increasing their output.

Public services are a third sector in which AI can make people’s lives easier. In the United States, I met families who, understandably, felt overwhelmed by the process of applying for health insurance, student grants or food aid. Faced with a huge pile of complicated bureaucratic forms, many were tempted to give up. AI can drastically simplify these processes, enabling them to receive the help they need more quickly and allowing the government to operate more efficiently. Governments can vastly improve the experience for citizens, starting precisely with those who are most in need of social safety nets.

AI can also be a great asset in education. It can free up teachers to spend more time with pupils individually or in small groups and provide them with a clearer picture of the difficulties faced by the whole class. For pupils, an AI tool that preserves what researchers call ‘productive struggle’ – the cognitive effort that builds understanding – can enhance learning. When a pupil first encounters a new idea, AI provides substantive explanations and suggests both questions and answers. Later, when checking how much the pupil has understood, it withholds the answer and helps them arrive at it on their own. Taken together, advances in all these areas could make everyday life simpler, less expensive and less dependent on a person’s income or knowledge.

In all these cases, the key word is ‘can’: AI can improve people’s lives across all income brackets. However, this will not happen automatically. As with any new technology, we must act consciously to ensure that it benefits everyone, not just a small, affluent elite. This will require a strong role for governments and philanthropic organisations, so that even less affluent citizens and low-income countries can benefit fully. (…)

It is encouraging that some AI companies are proposing solutions to the challenges posed by their own technology, but we should not expect them to take the lead on this. Some of these issues fall outside their remit and, in a democratic society, it is not up to them to make these decisions.

Solutions should instead be developed through a democratic and public process involving elected representatives, public policy makers, educators, healthcare professionals, local administrators and community leaders. Millions of people will see their lives turned upside down, and we will need a more robust and flexible social safety net to help them cope with the transition. Local communities are already raising concerns about the energy and water required by data centres. Without solutions, some groups may push to halt the development and roll-out of AI altogether.

Solutions should be shaped by the answers we provide to the profound questions raised by artificial intelligence, including that of how to preserve our humanity in an age when machines can think better than we can. In this endeavour, religious leaders can play a vital role, as they devote their lives to reflecting on what it means to be human. Pope Leo XIV’s encyclical on AI, ‘On the Protection of the Human Person in the Age of Artificial Intelligence’, provides a solid foundation for the work that lies ahead. Over the coming months, further ideas will be shared to ensure that the benefits of AI outweigh any potential harm. For now, here are three, beginning with the most important.

The top priority is a monumental task: to establish a national and international framework for dealing with AI. None of our current institutions were designed to manage a technology that is spreading so rapidly and affecting so many aspects of our lives. We will therefore have to create new ones.

It is difficult to overstate the scale of such an undertaking. Following the 9/11 attacks, the US government underwent its most extensive reorganisation since the Second World War with the aim of improving a single function: national security. AI will demand much, much more. It will have an impact on national security, but also on employment, education, taxation, energy, elections, air and water, public health, the financial system, law enforcement, transport, public land and IT systems.

These sectors overlap in ways that our current bureaucracy is not designed to handle. A Ministry of Labour may understand changes in the labour market, but not the safety risks. A business regulator may understand market concentration, but not the effects of AI on children and teenagers. Left to their own devices, institutions will see only part of the system, whilst the consequences of AI will ripple through the entire system.

At the national level, countries will need bodies capable of setting priorities that cut across different government departments. But even a country that manages to put its own house in order will remain exposed to risks that cross borders. That is why, in parallel, an international organisation will need to be established. It will be unlike any institution we have ever created, although it may draw inspiration from certain existing systems. There is a nuclear weapons inspection regime, international aviation regulations and agreements to protect the ozone layer. A new global organisation for AI will need to incorporate elements of all three of these models, and more besides.

It is legitimate to ask whether global institutions are up to the task of designing and implementing this new architecture. Governments move slowly, if at all, and polarisation within and between countries makes it harder than ever to achieve results. A degree of cooperation between the United States and China will be necessary.

We do not have the luxury of taking our time. The starting point is to set in motion a process to build the right institutions before the transformation forces governments to operate in crisis mode. Building the framework I am referring to will take years. That is why we must start immediately.

My father died of Alzheimer’s in 2020. In the most advanced stages of the disease, he was cared for day and night by carers who understood him even when he struggled to express himself. He couldn’t always tell them when he was hungry, but they knew anyway. My family and I will always be grateful to that extraordinary team of professionals. The care they provided was irreplaceably human: no robot could have – nor should have – carried out that task.

I think of that group whenever the question arises as to which jobs will disappear and which will remain. I believe that, as AI and robots improve, we will decide to reserve certain activities exclusively for humans. I have started calling this area ‘Human Reserved’, and it is an example of the sort of ideas we will need to consider.

The concept of ‘Human Reserved’ raises a number of questions to which I do not yet have answers. Who decides what should be reserved for humans? What criteria should we use? How can we prevent companies from circumventing the rules and using robots anyway? What happens to international trade when one country allows robots to produce something and another does not? All of this will need to be defined publicly, as part of the plan to manage the transition.

As workers are pushed into new jobs, they will need training and other forms of support guaranteed by the social safety net. But they will work fewer hours, which means they will pay less income tax, and government revenue will fall just at the very moment when demand for these services is set to rise most sharply. The resources will have to come from somewhere, at a time when budgets will already be under pressure.

I believe we should tax AI tokens and robots. Today, if you are an employer and hire a person, you pay social security contributions and employment taxes based on their salary. If, on the other hand, you purchase a robot, you can usually deduct the cost immediately as a business expense. The tax system, therefore, incentivises you to replace people with machines. A tax would at least partially slow down the rush to replace human labour and would raise funds for retraining and a stronger social safety net. It should be calibrated so as not to hinder the purely beneficial uses of AI, such as those that make healthcare and education more affordable. (…)

I will use my voice and my time to push AI and equity higher up the public agenda. I will raise this issue with lawmakers every time I go to Washington and when I meet leaders around the world. It will be at the heart of my conversations with the people who develop AI models. I will support the creation of the national and international framework I described earlier. The Gates Foundation will help to promote the beneficial uses of AI, including in Africa. My message to leaders is this:

You have the opportunity to act now, before unemployment rises dramatically, communities begin to suffer and public confidence erodes. You can ensure that your government tackles the problem as a whole, rather than dividing it up amongst numerous small bureaucratic fiefdoms. You can ensure that AI benefits everyone. And you can work with other governments to tackle this national and global challenge.

If we do things the right way, the benefits for humanity will be extraordinary and the world will become a fairer place. I rarely stop thinking about AI, not because it has all the answers, but because the questions it raises are too important to be left to a small group of technologists. Leaders from academia, business, government and civil society all have a role to play in shaping what comes next.

We need both a deep concern for the harms caused by AI – which we must minimise – and a realistic optimism regarding its positive effects, if we are to maximise them for everyone. Maximising the benefits is just as important as minimising the harms. The world needs a plan