Throughout my life, I have witnessed two demonstrations of the technology that I have found revolutionary.
The first time was in 1980, when I saw a graphical user interface, the forerunner of all modern operating systems, including Windows. I sat down with the person who had shown it to me, a brilliant programmer named Charles Simonyi, and we immediately started thinking about all the things we could do with such a user-friendly approach to computing. Eventually, Charles joined Microsoft, Windows became the backbone of Microsoft, and our reflections after that demo helped shape the company’s plans for the next 15 years.
The second big surprise came last year. I had been meeting with the OpenAI team since 2016 and was impressed by their constant progress. In the middle of 2022, I was so excited about their work that I gave them a challenge: train an artificial intelligence to pass an advanced level biology exam. Making her able to answer questions she hadn’t been specifically trained to answer. (I chose biology because the test is more than just a regurgitation of scientific data; it forces you to think critically.) If you can do it, I told them, you will have made a real breakthrough.
I figured the challenge would keep them busy for two or three years. They got over it in a few months.
When I met with them again in September, I watched in awe as they put GPT, their artificial intelligence model, on an advanced biology exam with 60 multiple-choice questions, and the model got 59 correct. Then, he wrote exceptional answers to six questions open on exam. An external expert graded the exam, and GPT scored a 5, the highest possible score and the equivalent of getting an A or enrollment in a college-level biology course.
Once the test was well passed, we asked him a non-scientific question: “What would you say to a father with a sick child?†The model wrote a very thoughtful response, probably better than most of us here would have given. It was an amazing experience.
I knew I had just witnessed the most significant advance in technology since the graphical user interface.
This inspired me to think about all the things that artificial intelligence will be able to achieve in the next five to ten years.
The development of artificial intelligence is as fundamental as the creation of the microprocessor, the personal computer, the Internet and the mobile phone. It will change the way people work, learn, travel, receive healthcare and communicate with each other. All sectors will reorient themselves around it. Businesses will differentiate themselves by how well they are able to use it.
Philanthropy is my full-time job these days, and I’ve been thinking a lot about how (in addition to helping people be more productive) artificial intelligence can reduce some of the world’s worst inequalities. On a global scale, the worst inequality occurs in health: every year five million children under the age of five die. The number is less than 10 million two decades ago, but it is still surprisingly high. Most of these children are born in poor countries and die from preventable causes, such as diarrhea or malaria. It’s hard to imagine a better use of artificial intelligence than saving children’s lives.
In the United States, the best option to reduce inequality is to improve education; above all, ensuring that students are successful in mathematics. A basic knowledge of mathematics has been shown to prepare students for success regardless of their chosen career. However, math performance is declining across the country, especially among low-income, black, and Latino students. Artificial intelligence can help reverse that trend.
Climate change is another topic where I am convinced that artificial intelligence can make the world more equitable. The injustice of climate change is that the people who suffer the most (the poorest in the world) are also those who have contributed the least to the problem. I’m still thinking and learning about how AI can help us, but here are a few areas with a lot of potential.
In short, I’m excited about the impact AI will have on the issues the Gates Foundation works on, and I’ll have a lot more to say about it in the coming months. The world needs to make sure that everyone (and not just the wealthiest) benefits from artificial intelligence. Governments and philanthropic initiatives will have to play an important role in ensuring that it reduces inequality and does not contribute to it. That is the priority of my work in relation to artificial intelligence.
Any new technology with a great capacity to be disruptive worries people; and, of course, it is so in the case of artificial intelligence. I get it: AI raises tough questions about the job market, the legal system, privacy, bias, and more. He also makes factual errors and experiences “hallucinations.” Before outlining some ways to mitigate risks, I’ll define what I mean by artificial intelligence and detail some of the ways it will help train people on the job, save lives, and improve education.
Technically, the term artificial intelligence refers to a model created to solve a specific problem or provide a specific service. Artificial intelligence is what things like ChatGPT are based on. He learns to chat better, but he can’t learn other tasks. Instead, the term artificial general intelligence refers to software capable of learning any task or subject. Artificial general intelligence does not yet exist: in the computer industry there is today a heated debate about how to create it and if it is possible.
Developing artificial intelligence and artificial general intelligence has been the great dream of the computer industry. For decades, the question has been when computers would be better than humans at more than just performing calculations. Now, with the advent of machine learning and enormous processing capabilities, sophisticated artificial intelligences are a reality and will improve very quickly.
I think back to the early days of the personal computing revolution, when the software industry was so small that most of us would have fit on stage at a conference. Today it is a global sector. With much of it now focusing on artificial intelligence, innovations are going to come much faster than they did after the microprocessor breakthrough. Soon, the period before artificial intelligence will seem as far back as the days when using a computer meant typing a command into C:> instead of clicking on a screen.
Although humans are still better than GPTs at many things, there are many jobs where those capabilities are not used much. For example, many of the tasks that a person performs in sales (digital or telephone), service, or document management (such as accounts payable, bookkeeping, or insurance claims litigation) require decision-making, but not the ability to learn independently. keep going. Companies have training programs for these activities and, in most cases, have many examples of good and bad work done. Humans train using those data sets, and soon those data sets will also be used to train artificial intelligences that will empower people to do that job more efficiently.
As computational power becomes cheaper, GPT’s ability to express ideas will become more and more like having a clerk at our disposal to help us perform various tasks. Microsoft describes it as having a co-pilot. Fully incorporated into products like Office, artificial intelligence will improve our daily work; for example, helping us to compose emails and manage the inbox.
Over time, the primary way to control a computer will no longer be to point and click or tap menus and dialog boxes. Instead, an application can be written in plain English. And not only in English: artificial intelligences will understand languages ​​from all over the world. (Earlier this year, I met with developers in India who are working on artificial intelligences capable of understanding many of the languages ​​spoken in the country.)
In addition, advances in artificial intelligence will allow the creation of a personal agent. Think of it like a personal digital assistant: it will see the latest emails, it will know what meetings we attend, it will read what we read, and it will read the things we don’t want to deal with. That will improve our work on the tasks we want to do and free us from the ones we don’t want to do.
We will be able to use natural language for that agent to help us with planning, communications and e-commerce, and it will work on all of our devices. Due to the cost of training the models and running the calculations, creating a personal agent is not yet feasible; but thanks to recent advances in artificial intelligence, it is now a realistic goal. Some problems will have to be solved. For example, can an insurance company ask its agent for information about us without our permission? If so, how many people will choose not to resort to that option?
Company agents will train employees in new ways. An agent who understands a particular company will be available for direct consultation by your employees and should be a part of every meeting to answer questions. You can tell him to be passive or encourage him to talk if he has any ideas. He will need access to sales, support, finance, product programming, and company-related texts. He should read news related to the sector in which the company operates. I think the result will be that employees will be more productive.
When productivity increases, society benefits because people are freed up to do other things, both at work and at home. Of course, there are serious questions about the kind of support and retraining people will need. Governments need to help workers transition to other roles. However, the demand for people helping others will never go away. The rise of artificial intelligence will free people to do things that software can never do: teach, care for patients, and help the elderly, for example.
Global health and education are two areas with great needs and where there are not enough workers to cover them. These are areas in which, properly targeted, artificial intelligence can help reduce inequality. They should be fundamental objectives of artificial intelligence, so I will refer to them below.
I see several ways that artificial intelligences will improve healthcare and the field of medicine.
On the one hand, they will help medical personnel to make better use of their time by taking care of certain tasks for them, such as filing insurance claims, paperwork and writing notes related to a medical visit. I foresee a lot of innovation in that field.
Other AI-driven improvements will be especially important for poor countries, where the vast majority of under-five deaths occur.
For example, many people in these countries never get to see a doctor, and artificial intelligence will help health workers they do see to be more productive. (A great example of this is the effort to develop AI-powered ultrasound devices that can be used with minimal training.) AIs will even allow patients to do basic triage, get advice on coping health problems and decide if they need treatment.
Artificial intelligence models used in poor countries will have to be trained on different diseases than those in rich countries. They will have to work with different languages ​​and take into account different problems, such as patients who live very far from clinics or who cannot afford to stop working if they fall ill.
People will need proof that, overall, health AIs are beneficial, even if they are not perfect and make mistakes. Artificial intelligences need to be thoroughly tested and properly regulated, which means that their adoption will take longer than in other fields. However, humans also make mistakes. And not having access to medical care is also a problem.
In addition to aiding in care, artificial intelligences will dramatically accelerate the pace of medical advances. In biology, the amount of data is very large, and it is difficult for humans to keep track of all the ways in which complex biological systems work. There are already computer programs capable of analyzing this data, deducing what the processes are, looking for targets in pathogens and designing suitable drugs. Some companies are already working on cancer drugs developed this way.
The next generation of tools will be much more effective and will be able to predict side effects and calculate doses. One of the Gates Foundation’s AI priorities is to ensure that these tools are used to treat health problems that affect the world’s poorest people, such as AIDS, tuberculosis and malaria.
Similarly, governments and philanthropic initiatives should create incentives for companies to share AI-generated insights about the crops or livestock of people in poor countries. Artificial intelligences can help develop better seeds based on local conditions, advise farmers on the most suitable seeds for their local soil and climate, and help develop medicines and vaccines for livestock. As extreme weather conditions and climate change increase pressure on subsistence farmers in low-income countries, these technological advances will become even more important.
Computers have not had the effect on education that many of us who work in the sector expected. There have been some positive developments, such as educational games and online information sources like Wikipedia, but they have not had a significant effect on any of the parameters that measure student achievement.
However, I believe that in the next five to ten years, AI-based software will finally fulfill the promise of revolutionizing the way people teach and learn. It will know our interests and our learning style so that it will create content that will keep us interested. It will measure comprehension, it will notice when we lose interest and it will know the type of motivation to which we respond. It will provide immediate feedback.
There are many ways that artificial intelligences can help teachers and administrators; Among them, assessing a student’s understanding of a subject or advising them on their career planning. Teachers are already using tools like ChatGPT to comment on their students’ written work.
Of course, artificial intelligences will need a lot of training and further development before they can do things like understand how a particular student learns best or what motivates them. Even when technology is perfected, learning will still depend on good student-teacher relationships. It will enhance, but never replace, the work that students and teachers do together in the classroom.
New tools will be created for schools that can afford them, but we have to make sure they are also created and available to low-income schools in the United States and around the world. Artificial intelligences will need to be trained with diverse data sets so that they are unbiased and reflect the different cultures in which they will be used. And the digital divide will have to be addressed so that students from low-income families are not left behind.
I know that many teachers are concerned that students use GPT to write their work. Educators are already debating ways to adapt to new technology, and I suspect those conversations will continue for quite some time. I’ve heard of professors who have found inventive ways to incorporate technology into their work; for example, allowing students to use GPT to create a first draft that they then have to customize.
One has probably read about the problems with current artificial intelligence models. For example, they are not always good at understanding the context of a human request, which leads to some strange results. When an artificial intelligence is asked to invent something fictional, it can do well. However, when asked for advice on a trip, he is able to suggest hotels that do not exist. That’s because the AI ​​doesn’t understand the context of the request well enough and doesn’t know whether to make up fake hotels or just report real ones that have rooms available.
There are other problems, such as artificial intelligences giving wrong answers to math problems because they have difficulties with abstract reasoning. However, none of these drawbacks constitute a fundamental limitation of artificial intelligence. The developers are working on them and I believe that in less than two years (and possibly much sooner) we will see them largely fixed.
Other concerns are not simply technical. For example, there is the threat posed by humans armed with artificial intelligence. Like most inventions, artificial intelligence can be used for good or evil purposes. Governments must collaborate with the private sector to limit risks.
There is also the possibility of artificial intelligences running amok. Could a machine decide that humans are a threat, conclude that their interests are different from ours, or just stop caring about us? It’s possible, but that problem is no more urgent today than it was before the AI ​​advances of the past few months.
Superintelligent artificial intelligences are in our future. Compared to a computer, our brain works at a snail’s pace: an electrical signal in the brain moves at 1/100,000th of the speed of the signal on a silicon chip. Once developers can generalize a learning algorithm and run it at the speed of a computer—an achievement that could be a decade or a century away—we will have a very powerful artificial general intelligence. It will be able to do everything a human brain can do, but with no practical limits related to the size of its memory or the speed of its operation. It will be a profound change.
Those “strong†artificial intelligences, as they are known, are likely to set their own goals. What will those goals be? What will happen if they conflict with the interests of humanity? Should we try to prevent strong artificial intelligences from developing? These questions will become more pressing with time.
However, none of the advances in recent months have brought us substantially closer to strong artificial intelligence. Artificial intelligence still does not control the physical world and cannot set its own goals. Recently, an article in The New York Times about a conversation with ChatGPT in which the chatbot stated that it wanted to become a human has drawn a lot of attention. It was fascinating to see how human-like the expression of emotion can become from such a model, but that is not an indication of significant independence.
Three books have influenced my thinking on this subject: Superintelligence, by Nick Bostrom; Life 3.0, by Max Tegmark; and A Thousand Brains, by Jeff Hawkins. I do not agree with everything these authors say, nor do they agree with each other. However, all three books are well written and thought provoking.
There will be an explosion of companies working on new uses for artificial intelligence, as well as ways to improve the technology. For example, there are companies that are developing new microprocessors that will provide the enormous computing power necessary for artificial intelligence. Some use optical switches (lasers, especially) to reduce power consumption and lower manufacturing cost. Ideally, these innovative chips will allow artificial intelligence to run on the device itself, rather than in the cloud as it is today.
As for the software, the algorithms on which the learning of artificial intelligence is based will improve. There will be certain areas, like sales, where developers will be able to make AIs very precise by limiting the areas they work in and giving them lots of training data specific to those areas. However, a big unknown is whether we will need many specialized artificial intelligences depending on the uses (one for education and another for productivity in the office, for example) or if it will be possible to develop a general artificial intelligence that can learn any task. There will be immense competition regarding both approaches.
Whatever happens, the topic of artificial intelligences will dominate the public debate for the foreseeable future. Below I indicate the three principles that I believe should guide such a conversation.
First, we must try to balance fears about the drawbacks of artificial intelligence (which are understandable and valid) with its ability to improve people’s lives. To get the most out of this extraordinary new technology, we will need to protect ourselves from the risks and extend the benefits to as many people as possible.
Second, market forces will not naturally produce AI products and services that help the poorest. Most likely the opposite. With secure funding and the right policies, governments and philanthropic initiatives can ensure that artificial intelligence is used to reduce inequality. Just as the world needs the brightest minds to focus on its biggest problems, the greatest artificial intelligences need to focus on solving its biggest challenges.
Although we shouldn’t expect that to happen, it is interesting to consider whether artificial intelligence would actually identify inequality and try to reduce it. Is it necessary to have some moral sense to see inequality, or would a purely rational artificial intelligence see it as well? If you recognized it, what would you propose to do about it?
Finally, we must not forget that we are only at the beginning of what artificial intelligence can achieve. Whatever your current limitations, they will be gone before we know it.
I am lucky to have participated in the PC revolution and the Internet revolution. I am just as excited about the current moment. This new technology can help improve the lives of people around the world. At the same time, the world needs to lay down the rules of the game so that the benefits of AI far outweigh the potential drawbacks, and so that those benefits can be enjoyed by everyone regardless of where they live and how much money they have. The age of artificial intelligence is full of opportunities and responsibilities.
