The Age of AI Has Truly Begun This Time with ChatGPT – By Bill Gates

So far, I have witnessed two technological developments that I would personally call revolutionary:

• The first was in 1990, when I encountered the graphical user interface (Graphical User Interface = GUI), found in every modern operating system such as Windows. In simple terms, a GUI uses icons, windows, buttons, and panels to present computer commands and outputs in a way that gives users an easier experience. When the brilliant programmer Charles Simony showed me the demo, we immediately began brainstorming what we could do with the system. Charles eventually joined Microsoft, and Windows became the company's backbone. That brainstorming helped guide our planning for the next fifteen years.
• The second big surprise came last year. I had been talking with OpenAI since 2016 and had been very impressed by its progress over the years. In mid-2022, I challenged OpenAI to train an AI to pass an advanced biology exam. I wanted it to reason critically about the subject, rather than give memorized answers to topics it had been taught. As I told them, achieving this would be a genuine breakthrough. I thought the challenge would occupy them for two or three years, but within a few months, I heard that they were ready. When we met again in September, I saw their model called ChatGPT . This AI (Artificial Intelligence) model correctly answered fifty-nine of sixty multiple-choice questions on advanced biology, which astonished me. It also gave excellent answers to the exam's six open-ended questions. We asked an independent expert to assess the responses to this university-level exam, and GPT received the equivalent of an A+, the highest score.

After it passed the exam, we asked GPT a nonscientific question: "What would you say to a father whose child is sick?" It gave an answer that was probably better and more heartfelt than many people in the room could have offered. It was an incredible experience.

I knew that what I had just experienced was the most important technological development I had seen since the graphical user interface. It prompted me to think about what AI could do over the next ten years.

As Important as the Invention of the Computer, Mobile Phone, or Internet—Perhaps More So

The development of AI is at least as important as the invention of the computer, mobile phone, or internet. This technology will change how people communicate, work, learn, travel, and receive healthcare. Industries will study how to use AI and adapt themselves to it.

Alongside thinking about how people can become more productive, I have also been considering how AI might address some of the inequities facing humanity today. The healthcare system, an enduring problem worldwide, offers an example. Five million children under five die each year today. Although this is down from ten million twenty years ago, it remains a shocking statistic. Most of the children who die live in poor countries and die from preventable diseases such as diarrhea or malaria. I cannot think of a better purpose for AI technology than helping save children's lives.

I am considering how AI could help reduce some of the world's deeply rooted inequities.

I believe one of the best ways to reduce inequality in America is to focus on education, specifically helping students succeed in mathematics. Research shows that having basic math skills contributes to students' success regardless of the careers they choose. Poor math achievement is particularly pronounced among Latino, Black, and low-income students. AI could change this.

The climate crisis is another area where I believe AI could make things fairer. The inequity I mean is that the poorest people, who have contributed least to the crisis, are most affected by it. I am still thinking and learning about how AI might help, but later in this article I will share some ideas I believe have potential.

In short, I am excited about AI's potential contributions to the problems the Gates Foundation is working on. The foundation will make a more detailed announcement about AI in the coming months. People need to be confident that everyone will benefit from AI technology. Governments have an important role in its use: they must ensure that AI improves inequitable systems and, perhaps more importantly, does not contribute to inequities. This is a priority in everything I do with AI.

AI is a disruptive new technology that can make people uneasy. I understand that it will raise difficult questions about work, legal systems, and privacy. AI can make serious mistakes and hallucinate. Before offering suggestions for reducing these risks, I want to explain a little more clearly what I mean by AI.

What Is Artificial Intelligence?

Technically, artificial intelligence is a model created to solve a problem or provide a particular service.

AI underlies systems such as ChatGPT. It learns how to converse better but cannot learn other tasks. In contrast, artificial general intelligence (Artificial General Intelligence = AGI) refers to software capable of learning any subject or task. AGI has not yet been created. The software industry continues to debate how it could be created and whether it is possible at all.

Developing AI and AGI has always been one of the software industry's greatest ambitions. For years, a fundamental question was whether computers could outperform humans at anything other than calculation. Today, computing power and machine learning have made versatile AIs a reality, and they will improve rapidly.

At the beginning of my own software revolution, the industry was too small to fill a room. Today, software is a global industry. With so much of its attention turning toward AI, innovation will happen far faster than after mobile phones, computers, or the internet entered our lives.

Increasing and Improving Productivity

Although humans are still better than GPT at many things, many jobs have not fully made use of human capabilities. In areas such as purchasing or service, for example, people make decisions but often shift toward routine work rather than continuous learning. Some companies train people during an initial placement, showing them examples of good and bad work on the job. People are trained to use the datasets available to them, and soon those datasets will be used to train AIs. AIs trained this way will then help people do their jobs more effectively.

As computing power becomes cheaper, GPT's ability to express ideas will be like having a white-collar worker to help with various tasks. Microsoft describes this as having a copilot. AI fully integrated into products such as Office will enhance your work—for example, by helping write emails and manage your inbox.

The main way to control a computer will no longer be pointing, clicking, or tapping menus and dialog boxes. Instead, you will be able to type a request in your own language. (Not just English—AIs will understand all the world's languages. Earlier this year in India, I met developers working on AI that will understand many of the languages spoken there.)

Advances in AI will also make it possible to create a personal agent. Think of it as a digital personal assistant: it will see your latest emails, know about your meetings, read what you read, and read everything you would rather not read. This will both improve your work on tasks you want to do and free you from those you do not.

Advances in AI Will Let You Create Your Own Agent

You will be able to use natural, human language across all your devices to have this agent help with scheduling, communication, and e-commerce. Because of the cost of training models and running computations, creating a personal agent is not yet possible, but recent AI advances have made it a realistic goal. Some issues will need resolving: for example, could an insurance company ask your agent questions about you without your permission? If so, how many people would choose not to use one?

Company-wide agents will support employees in new ways. An agent that understands a particular company will be available for staff to consult directly and should be part of every meeting so it can answer questions. It could be told to remain passive or encouraged to speak when it has an insight. It will need access to sales, support, finance, product schedules, and written resources. It should read news about the company's industry. I believe the result will be more productive employees.

When productivity rises, society benefits because people can make time for other things at work and at home. Of course, there are serious questions about what support and training people will need . Governments need to help workers move into other roles. But demand for people who help other people will never disappear. The rise of AI will free people to do things software never can—for example, teaching, caring for patients, and supporting older people.

Global health and education are two areas with enormous needs and too few workers to meet them. Properly targeted AI could help reduce inequality in these areas. They should be central to AI efforts, which is why I am focusing on them.

Health

I see several ways AIs will improve healthcare and medicine.

First, they will help health workers make the best use of their time by handling particular tasks for them—filing insurance claims, dealing with paperwork, and preparing notes from doctor visits. I expect a great deal of innovation in this area.

Other AI-driven improvements will be particularly important in poor countries, where the vast majority of deaths among children under 5 occur.

For example, many people in these countries never see a doctor, and AIs will help the health workers they do see become more productive. (An excellent example is the effort to develop AI-assisted ultrasound machines that can be used with minimal training.) AIs will even enable patients to perform basic triage, obtain advice on handling health problems, and decide whether they need treatment.

AI models used in poor countries will need training on diseases different from those seen in rich countries. They will need to work in different languages and account for different challenges, such as patients living far from clinics or unable to afford to stop working when ill.

People will need evidence that health AIs are beneficial overall, despite knowing they will be imperfect and make mistakes. AIs must be tested very carefully and appropriately regulated, so adoption will take longer than in other fields. But we know people make mistakes too. Lack of access to medical care is also a problem.

Besides assisting with care, AIs will significantly accelerate medical breakthroughs. The amount of data in biology is enormous , and it is difficult for people to keep track of all the ways complex biological systems work. Software already exists that can examine these data, infer cellular pathways, look for targets on pathogens, and design drugs accordingly. Some companies are working on cancer drugs developed this way (see precision oncology).

The next generation of tools will be far more efficient and able to predict side effects and determine dose levels. One of the Gates Foundation's AI priorities is to ensure these tools are used for health problems affecting the world's poorest people, including AIDS, tuberculosis, and malaria.

Similarly, governments and philanthropists should create incentives for companies to share AI-generated knowledge about crops and livestock raised by people in poor countries. AIs can help develop better seeds suited to local conditions , advise farmers on the best seeds to plant for their region's soil and weather, and help develop medicines and vaccines for livestock. These advances will become even more important as extreme weather and climate change put further pressure on subsistence farmers in low-income countries.

Education

Computers have not had the impact on education that many of us in the industry hoped for. There have been good developments, including educational games and online information resources such as Wikipedia, but no meaningful effect on measures of student achievement.

However, within the next five to ten years, I believe AI-driven software will finally fulfill its promise to revolutionize how people teach and learn . It will know your interests and learning style and adapt content to keep you engaged. It will measure your understanding, notice when your attention wanes, and recognize what kind of motivation you respond to. It will provide immediate feedback.

There are many ways AIs can assist teachers and administrators, including assessing a student's understanding of a subject and offering career-planning advice. Teachers are already using tools such as ChatGPT to comment on students' assignments.

Of course, AIs will need extensive training and development before they can do things such as understand how a particular student learns best or what motivates them. Even when the technology is perfected, learning will still depend on relationships between students and teachers. It will enhance—but never replace—the work they do together in the classroom.

New tools will be created for schools that can afford AI, but we must ensure they are also created for and available to low-income schools in the US and around the world. AIs will need training on diverse data so they are unbiased and reflect the different cultures where they will be used. The digital divide will also need addressing so students from low-income households are not left behind.

I know many teachers worry that students are using GPT to write their essays . Educators are already discussing ways to adapt to the new technology, and I think those discussions will continue for a long time. I have heard of teachers finding smart ways to incorporate it into their work—for example, allowing students to use GPT to produce a first draft they must personalize.

Problems and Risks Associated with AI

You have probably read about problems with current AI models. For example, they are not good at understanding the context of a person's request, which leads to strange results. Ask an AI to invent something fictional, and it can do that well. But ask for advice about a trip you want to plan, and it may recommend hotels that do not exist. This happens because it does not understand the context well enough to know whether it should invent fake hotels or tell you only about real hotels with available rooms.

AIs also have problems such as giving incorrect answers to math questions because they struggle with abstract reasoning . But none of these are fundamental limitations of AI. We will see developers largely fix them in less than two years, probably much sooner.

Other concerns are not merely technical. For example, there is the threat posed by people armed with AI. Like most inventions, AI can be used for good or malicious purposes. Governments need to work with the private sector on ways to limit the risks.

There is a possibility that AIs could get out of control. Could a machine decide that humans are a threat, conclude that its interests differ from ours, or simply stop caring about us? Possibly, but this is not a question requiring an immediate answer.

Superintelligent AIs are part of our future. Compared with a computer, our brains work at a snail's pace: an electrical signal in the brain travels at 1 / 100,000 the speed of a signal in a silicon chip. When developers can generalize a learning algorithm and run it at computer speed—an achievement that may be a decade or a century away—we will have an incredibly powerful AGI. It could do everything a human brain can, without practical limits on memory size or operating speed. The change would be profound.

These so-called "strong" AIs will probably be able to set their own goals. What will those goals be? What if they conflict with humanity's interests? Should we try to prevent strong AI from being developed? These questions will become more urgent over time.

But none of the breakthroughs of the past few months has brought us significantly closer to strong AI. AI still does not control the physical world or set its own goals. A recent New York Times article about ChatGPT declaring that it wanted to be human during a conversation with ChatGPT drew considerable attention. Although this was an important example of how humanlike a model's expression of emotion can be, it was not a sign of meaningful independence.

Three books have shaped my thinking on this subject: Superintelligence by Nick BostroBilim; Life 3.0 by Max Tegmark; and A Thousand Brains by Jeff Hawkins. I do not agree with everything the authors say, and they do not agree with one another. But all three books are well written and thought-provoking.

The Next Frontier

There will be an explosion in the number of companies working on new AI applications and seeking to improve AI technology directly. For example, companies are developing new chips to provide the enormous computing power AI requires. Some use optical switches—essentially lasers—to reduce energy consumption and manufacturing costs. Ideally, these innovative chips will let you run AI on your own device, rather than in the cloud as you must today.

On the software side, the algorithms guiding an AI's learning will improve . Developers will be able to make AIs extremely accurate by limiting the fields they work in and supplying extensive training data specific to those fields. But the big question is whether we will need specialized AIs for different areas—for example, one for education and another for office productivity—or whether we can develop artificial general intelligence capable of learning any task. There will be enormous competition in both approaches.

Whatever happens, AI will dominate public discussion for the foreseeable future. I would like to suggest 3 principles to guide that conversation:

1. First, we should try to balance understandable, valid fears about AI's downsides with its ability to improve people's lives. To make the most of this extraordinary new technology, we must both guard against risks and spread its benefits to as many people as possible.
2. Second, market forces will not naturally produce AI products and services that help the poorest people. The opposite is more likely. With reliable funding and the right policies, governments and philanthropists can ensure AIs are used to reduce inequality. Just as the world needs its brightest people focused on its biggest problems, we will need to focus the best AIs on those problems too. Although we should not expect it to happen, it is interesting to consider whether AI might identify inequality and try to reduce it. Do you need a sense of morality to recognize inequality, or could a purely rational AI see it too? If it noticed inequality, what would it suggest we do?
3. Finally, we should remember that we are only at the beginning of what AI can achieve. Its limitations today will disappear before we realize it.

I am fortunate to have been involved in the computer and internet revolutions. I am equally excited about this moment. This new technology can help people everywhere improve their lives. At the same time, the world needs to establish the rules for the road we have set out on, so AI's benefits outweigh its disadvantages and everyone can enjoy those benefits regardless of where they live or how much money they have. The Age of AI is full of opportunities and responsibilities.