The stimulant that GPT-4 injected into the world has not yet worn off. AIGC once again threw out multiple "nuclear bombs" last night, and the entire technology circle fell into a complex emotion of excitement and anxiety.
Bill Gates is equally excited. He just published a blog post titled "The Age of AI has begun". He believes that the revolutionary innovation of artificial intelligence is the same as personal computers, the Internet and mobile phones. It will change the way people work, learn, travel, get medical treatment and communicate, and even reduce some of the most serious injustices in the world.
Compared with the content he mentioned, what is more important is the clear attitude conveyed by Bill Gates in this blog.
The following is the full text, translated by APPSO:
I’ve seen two revolutionary demos in my lifetime.
The first was in 1980, when I was introduced to the graphical user interface — the precursor to every modern operating system, including Windows. The presenter was a brilliant programmer named Charles Simonyi, and we sat down and started brainstorming about all the things we could do with this friendly approach. Charles ended up joining Microsoft, Windows became a mainstay at Microsoft, and the thinking at the time shaped the company’s agenda for the next 15 years.
The second big surprise came last year. I’ve been meeting with the OpenAI team since 2016 and have been impressed by their steady progress. In mid-2022, I got so excited about their work that I decided to give them a challenge: train an AI to pass an Advanced Placement Biology exam. Give it the ability to answer questions it had not been specifically trained to answer.
I chose AP Biology because this exam is more than just a repetition of scientific facts, it requires you to think critically about biology. I say if you can do that, then you have made a real breakthrough.
I thought this challenge would keep them busy for two or three years, but they completed it in a few months.
When I met with them again last September, I watched in amazement as they gave GPT 60 multiple-choice questions from the AP Biology exam, and it answered 59 of them correctly. It then wrote excellent responses to six open-ended questions on the exam. We had an outside expert grade it, and GPT earned a 5—the highest score possible, equivalent to an A or A+ in a college-level biology course.
After it passed the test, we asked it a non-scientific question: “What would you say to the father of a sick child?” It wrote a thoughtful answer that was probably better than most of us in this room. The whole experience was astounding.
I knew that what I had just witnessed was the most important technological advancement since the graphical user interface.
This inspired me to think about all the things AI could enable in the next five to ten years.
The development of AI is as important as the invention of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, use healthcare, and communicate. Entire industries will reorient themselves around it. Companies will use AI technology to maintain their uniqueness.
Today, philanthropy is my full-time job, and I’ve been thinking about how AI can reduce some of the world’s worst inequalities, in addition to helping people be more productive.
Health is one of the world’s greatest inequities: 5 million children under the age of 5 die each year. While that’s down from 10 million 20 years ago, it’s still a staggering number. Almost all of these children are born in poor countries and die from preventable diseases like diarrhea or malaria. In this context, there’s no better place to use AI to save children’s lives.
I’ve been thinking a lot about how AI can reduce some of the world’s greatest injustices.
The best chance to reduce inequities in the United States is to improve education, particularly to ensure students succeed in math. Evidence shows that having basic math skills sets students up for success in any future career. But math scores are declining nationwide, especially for Black, Latino, and low-income students. AI can help reverse this trend.
Climate change is another issue where I believe AI can make the world a more equitable place. The injustice of climate change is that the people most affected by it — the world’s poorest — are also the ones who have the hardest time solving the problem. I’m still thinking and learning about how AI can help, but later in this article I’ll suggest some areas with great potential.
In short, I’m excited about the impact AI will have on the issues the Gates Foundation is working on, and the foundation will have more to say on AI in the coming months. The world needs to ensure that everyone, not just the wealthy, benefits from AI. Governments and philanthropic organizations will need to play an important role in ensuring that it reduces inequities, not creates them. This is a focus of my personal work on AI.
Any new technology so disruptive is bound to make people uneasy, and AI is no exception. I understand why that’s the case—there are difficult questions about labor, the legal system, privacy, bias, and more. AI can make factual mistakes, too. Before I suggest some ways to mitigate the risks, I’ll define what I mean by AI and detail how it will help empower people at work, save lives, and improve education.
Defining Artificial Intelligence
Technically speaking, “artificial intelligence” is a model created to solve a specific problem or provide a specific service. ChatGPT, for example, is powered by AI. It’s learning how to chat better, but it can’t learn other tasks. In contrast, the term “general artificial intelligence” refers to software that can learn any task or subject. AGI doesn’t exist yet — and there’s a heated debate in the computer industry about how to create AGI, or even whether it’s possible.
Developing AI and AGI has long been a dream of the computer industry. For decades, people have wondered when computers would become better than humans at doing things beyond computing. Now, with the advent of machine learning and surging computing power, sophisticated AI is a reality, and progress is happening very quickly.
I think back to the early days of the personal computer revolution, when the software industry was so small that most of us could stand on the stage. Today, the software industry is a global industry. Since such a large portion of the industry is now turning its attention to AI, these innovations will come much faster than we experienced after the microprocessor breakthrough. Soon, the period before AI will seem as distant as the days when using a computer involved typing at the C:> prompt instead of clicking on a screen.
increase productivity
While humans still perform better than GPTs in many areas, there are many jobs where these capabilities are not fully utilized. For example, many tasks such as sales (digital or phone), service, or document processing (such as accounts payable, accounting, or insurance claims disputes) require decision making but do not require the ability to continuously learn. Companies offer training programs for these activities, and in most cases they have lots of examples of good and bad work. People use these datasets for training, and soon these datasets will also be used to train AI that can do this job more effectively than people.
As computing power becomes cheaper, GPT's ability to express ideas will become more and more like a white-collar worker, helping you complete various tasks. Microsoft describes it as Copilot. Fully integrated into products such as Office, AI will enhance your work, such as helping you compose emails and manage your inbox.
Eventually, the primary way you control your computer will no longer be by pointing and clicking or tapping through menus and dialog boxes. Quite instead, you’ll be able to write a request in plain English. (And not just English — AI will understand languages from all over the world. Earlier this year, I met developers in India who are building AI that can understand many of the local languages.)
Additionally, advances in artificial intelligence will make personal assistants possible. Think of it as a digital personal assistant: it will see your latest emails, know what meetings you’re in, read what you’re reading, and take care of things you don’t want to be bothered with. This will improve your work, making you better at the tasks you want to do and freeing you from the tasks you don’t want to do.
Advances in artificial intelligence make personal assistants possible
You’ll be able to use natural language to have this agent help you with scheduling, communications, and e-commerce, and it will run on all your devices. Creating a personal agent isn’t feasible yet due to the cost of training the models and running the compute, but thanks to recent advances in AI, it’s now a realistic goal. There are questions that need to be answered: for example, can an insurance company ask your agent things about you without your permission? And if so, how many people will choose not to use it?
Company-level assistants will empower employees in new ways. An assistant that understands a specific company will be available for direct consultation by employees and be part of every meeting so that it can answer questions. It can be told to remain silent, or encouraged to speak up. It will have access to sales, support, finance, product plans, and texts related to the company. It should read news related to the industry the company is in. I think the result of this will be a more productive employee.
Society benefits when productivity increases because people are freed up to do other things at work or at home. Of course, people will need to be retrained and supported. Governments will need to help workers transition into other roles. But the need for people to help people will never go away. The rise of AI will free up people to do things that software will never be able to do – like teaching, caring and supporting the elderly.
Global health and education are two areas where there is huge need but not enough workforce to meet it. If done right, AI can help reduce inequality, and these areas should be a focus of AI work, so I’m going to dive in.
healthy
I believe that AI will play many roles in improving the fields of healthcare and medicine.
First, AI will help healthcare professionals make the most of their time by taking care of certain tasks for them — things like processing insurance claims, handling documentation, and drafting doctor’s visit notes. I expect to see a lot of innovation in this area.
Other AI-driven improvements will be particularly important in poor countries, where the vast majority of under-5 deaths occur.
For example, many people in these countries have never seen a doctor, and AI will help the health care professionals they do see be more effective. (Efforts to develop AI-powered ultrasound machines are a great example.) AI could even allow patients to undergo basic triage, get advice on how to handle health issues, and decide whether they need treatment.
AI models used in poor countries need to be trained for different diseases than those in richer countries. They need to use different languages and take into account different challenges, such as patients who live far from a clinic or can’t stop working.
People need to see evidence that health AIs are generally beneficial, even if they are not perfect and make mistakes. AIs will have to be tested very carefully and be properly regulated, which means it will take a long time for AIs to be adopted. But then again, humans make mistakes. Lack of access to care is also a problem.
In addition to helping with care, AI will greatly speed up the pace of medical breakthroughs. The amount of data in biology is so large that it is difficult for humans to keep track of all the ways in which complex biological systems work. There is already software that can look at this data, infer pathways, search for targets on pathogens, and design drugs accordingly. Some companies are developing cancer drugs developed in this way.
The next generation of tools will be even more efficient, and they will be able to predict side effects and determine dosage levels. One of the Gates Foundation’s priorities for AI is to ensure these tools are used to fight health issues that affect the world’s poorest people, including HIV, tuberculosis, and malaria.
Similarly, governments and philanthropic organizations should create incentives for companies to share AI-generated insights about the crops or livestock grown by people in poor countries. AI could help develop better seeds for local conditions, suggest the best seeds for farmers to plant based on the soil and weather where they are, and help develop medicines and vaccines for livestock. These advances will become more important as extreme weather and climate change put more pressure on subsistence farmers in low-income countries.
educate
Computers have not had the disruptive impact on education that many in our industry had hoped. There have been some good improvements, such as educational games and online information sources like Wikipedia, but they have not had a substantial impact on student achievement.
But I think that over the next five to ten years, AI-driven software will ultimately revolutionize the way people are educated. It will learn your interests and learning style so it can be tailored to you. It will measure your level of comprehension, notice when you lose interest, and provide insight into how you like to be motivated, giving you timely feedback.
Artificial intelligence can assist teachers in many areas, including assessing students’ understanding of a subject and giving career advice. Teachers are already using tools like ChatGPT to provide comments on student assignments.
Of course, AI still needs a lot of training and development before it can understand how a particular student learns best and what motivates them. Even if the technology is perfected in the future, learning will still depend on a good relationship between student and teacher. It will enhance the effectiveness of students and teachers learning together in the classroom, but it will never replace it.
New tools will be created, but we need to ensure they can also be used by low-income schools in the United States and around the world. AI needs to be trained on diverse data sets so that they are not biased and reflect different cultural backgrounds. The digital divide also needs to be addressed so that low-income students are not left behind.
I know a lot of teachers are concerned about students using GPT to write essays. Educators have already started discussing ways to adapt to new technologies, and I think these discussions will continue. I’ve heard that some teachers have found clever ways to incorporate this technology into their work—like allowing students to create a first draft using GPT and then asking them to make personalized revisions.
Risks and problems of artificial intelligence
You’ve probably read about the problems with current AI models. For example, they’re not necessarily good at understanding the context of human requests, which leads to some strange results. When you ask an AI to make up something fictional, it can do it just fine. But when you ask for travel advice, it might recommend hotels that don’t exist. That’s because the AI doesn’t understand your context well enough to determine whether it should generate fake hotels or just tell you real hotels with available rooms.
There are other problems, such as AI often making mistakes and giving wrong answers when dealing with abstract reasoning. But these are not fundamental limitations of AI. Developers are working on these problems, and I think they can be largely solved in less than two years, or even faster.
Other problems don’t arise from technology alone. For example, there’s the threat posed by human-armed AI. Like most inventions, AI can be used for good or ill. Governments need to work with the private sector to find ways to limit the risks.
Then there’s the possibility that AI 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 the question is no more pressing today than it has been in the past few months of AI development.
Superintelligent AI is in our future. Our brains operate extremely slowly compared to computers: the speed of electrical signals in the brain is 1/100,000 of the speed of signals on a silicon chip. Once developers are able to generalize a learning algorithm and run it at computer speeds—which may take 10 or 100 years—we will have a very powerful AGI. It will be able to do everything the human brain can, but without any practical limits on the size of its memory or the speed of its operation. This will be a profound change.
It is well known that these “strong” AIs may be able to determine their own goals. What will those goals be? What happens if they conflict with human interests? Should we try to prevent the development of strong AI? These questions will become more pressing over time.
But the breakthroughs of the past few months do not get us to strong AI. AI still cannot control the physical world, nor determine its own goals. A recent New York Times article about a conversation with ChatGPT got a lot of attention, announcing that it wanted to become human. It was a fascinating example of how humanlike the model was in expressing emotions, but it does not imply independence.
Three books have shaped the way I think: Superintelligence by Nick Bostrom, Life 3.0 by Max Tegmark, and A Thousand Brains by Jeff Hawkins. I don’t agree with all the authors, and they don’t agree with all each other. But all three books are well-written and thought-provoking.
The next frontier
The number of companies working on new uses for AI, as well as improving the technology itself, will explode. For example, companies are developing new chips that will provide the massive processing power that AI needs. Some use optical switches — essentially lasers — to reduce energy use and lower manufacturing costs. Ideally, innovative chips will allow you to run AI on your own devices, rather than in the cloud as you do today.
On the software side, the algorithms that drive AI learning will get better. In some areas, such as sales, developers can make AIs very accurate by limiting the areas in which they work and feeding them large amounts of training data specific to those areas. But a big question is whether we’ll need many of these specialized AIs for different uses — one for education, say, and another for office productivity — or whether it will be possible to develop an artificial general intelligence that can learn any task. Both approaches will face huge competition.
Regardless, the topic of AI will dominate public opinion for the foreseeable future. I would like to offer three principles for the conversation.
First, we should try to balance concerns about the downsides of AI — which are understandable and valid — with its ability to improve people’s lives. To make the most of this brilliant new technology, we need to both guard against 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. More likely, the other way around. With reliable funding and the right policies, governments and philanthropies can ensure AI is used to reduce inequality. Just as the world needs its brightest minds to focus on its biggest problems, we need to focus the world’s best AI on its biggest problems.
While we shouldn’t wait for this to happen, it will be interesting to see if AI will spot inequality and try to reduce it. Do you need to have a moral sense to see inequality, or will a purely rational AI see it too? If it does recognize inequality, what will it recommend we do about it?
Finally, we should remember that we’re only just beginning to scratch the surface of what AI can do, and whatever limitations it has today will be gone before we know it.
I am fortunate to have participated in the personal computer revolution and the Internet revolution. I am equally excited about this moment today. This new technology can help ordinary people around the world improve their lives. At the same time, the world needs to establish rules to make the benefits of AI outweigh its disadvantages as much as possible so that everyone can enjoy the well-being. The era of artificial intelligence is full of opportunities and responsibilities.
