Microsoft CEO Nadella on AI: I want the world to enjoy the benefits that the industrial revolution brought to the West

In 2016, I mentioned 3 things that excited me: mixed reality, quantum, and artificial intelligence. Now I'm just as excited about these 3 things.

Wired took a close-up photo of Microsoft CEO Satya Nadella during the interview.

Microsoft is in the limelight on the wave of generative artificial intelligence (AI). As the helm of Microsoft's AI strategy, Microsoft CEO Satya Nadella (Satya Nadella) recently accepted an exclusive interview with Steven Levy, a senior reporter and columnist of the US technology media "Wired", he said: "My dream is that every 8 billion people on the planet can have an AI teacher, an AI doctor, an AI programmer, or an AI consultant." ”

Nadella was born in India and joined Microsoft in 1992. Since becoming CEO of Microsoft in 2014, he has led the company through a series of changes and innovations, including open source software, acquisition of well-known companies, and enhancement of cloud computing services, which has brought Microsoft's market capitalization to $2 trillion. His thoughtfulness and humility contrasted with his predecessors, Bill Gates and Steve Ballmer.

Still, Microsoft never seems to have fully regained its 90s glory until recently. When startup OpenAI began developing generative AI products, Nadella quickly realized that a partnership with the company and its CEO, Sam Altman, would put Microsoft at the center of a new AI boom. He led Microsoft in forging a close partnership with OpenAI, although that collaboration has recently been revealed to be contradictory. But regardless, Microsoft has introduced some impressive generative AI products. Nadella's rapid and widespread adoption of AI technology shows a bold attitude that harks back to Microsoft's early dynamism. Now, everyone wants to know what he thinks about the most important tech topic of the century: artificial intelligence.

The following is the transcript of Steven Levy's interview with Nadella published in Wired on June 13, translated by The Paper (www.thepaper.cn), with deletions:

Why partner with OpenAI

Steven Levy (senior reporter and columnist at Wired): When did you realize that this phase of AI could be so transformative?

Satya Nadella (CEO of Microsoft): When we transitioned from GPT-2.5 to GPT-3. The researchers discovered some new features, and GPT-3 began to exhibit scaling effects. The researchers didn't specifically train it to program, but it was very good at programming. I've been blown away ever since. I thought, "Wow, that's really awesome. ”

Levy: Was there a moment when you wanted to bet on AI all out?

Nadella: It was this ability to write code that led us to create Copilot. But in the summer of 2022, I saw for the first time what is now called GPT-4, and that was an eye-opener. I can give you an example: machine translation is now mature and has many excellent results, but it does not yet grasp the nuances of deep meaning in poetry. When I grew up in Hyderabad, India, I always wanted to read Persian poetry—especially Rumi's work, which was always translated first into Urdu and then into English. GPT-4 does not need to be so complicated, it can be translated directly into English. It's not just machine translation, it's the ability to cross the boundaries of two languages and preserve the essence of poetry. It's really cool.

Levy: Microsoft has been investing in AI for decades – don't you have your own big language model? Why do you need OpenAI?

Nadella: We also have our own set of efforts, including a language model called Turing, which is embedded in Bing and available on Azure and other platforms. But I think OpenAI shares our vision. So I don't want to try to train 5 different base models, but I want a foundation that makes this foundation the support of the platform effect. So we cooperated, they chose us, we chose them. They take care of the underlying models, and we do scaling work on top of them, including responsible AI and AI security tools. At the end of the day, we're two deeply cooperating independent companies, and we choose to do it methodically rather than multiple teams messing around. We had the same goal and said, "Let's go and achieve this goal and create something that really grabs the world's attention." ”

Levy: Are you trying to acquire OpenAI?

Nadella: When I was growing up at Microsoft, I often dealt with partners in a variety of interesting ways. Previously, we worked closely with SAP (the German market leader in enterprise applications) to develop SQL Server (Microsoft's relational database and data management and analytics platform). So this kind of thing doesn't surprise me. OpenAI has a special structure, though—it's nonprofit.

Levy: It seemed like an unsolvable problem, but you and OpenAI came up with a complex solution.

Nadella: They created a for-profit entity and we said, "We can take it. "We have a good business relationship. We have a long-term stable agreement.

Levy: Obviously, with this arrangement, not only OpenAI will be able to profit from your deals, but Microsoft itself will also be able to profit. But there's a profit cap on your collaboration, and when you reach that limit, like Cinderella's carriage turns into a pumpkin — OpenAI becomes a pure nonprofit. How will the partnership evolve by then? Can OpenAI say, "We're completely non-profit, and we don't want to be involved in commercial operations." ”

Nadella: I think their blog has explained that. But essentially, their long-term goal is that we can achieve superintelligence. If it did, I guess everything would become unpredictable, right?

Dream of 8 billion Earthlings having their own AI teachers, doctors and consultants

Levy: Do you agree with OpenAI CEO Sam Altman that the level of AGI superintelligence will become a reality?

Nadella: I'm more concerned about what benefits AGI (General Artificial Intelligence) can bring us. I've always regretted that the Industrial Revolution didn't change where I grew up. So I wanted to find something that might transcend the Industrial Revolution and allow people all over the world to enjoy the benefits that the Industrial Revolution had brought to the West. So I'm not afraid of AGI (General Artificial Intelligence) coming out, or coming out too quickly. Isn't that great? This means that 8 billion people around the world can live prosperous lives. It will be a wonderful world.

Levy: What is the roadmap to achieve this vision? You're now embedding AI into your search engines, databases, developer tools. But those who need help don't use these things.

Nadella: That's a good question. We can start at the forefront of developers, and one of the things I'm really looking forward to is that development will become interesting. Microsoft started out as a tool-based company, specifically developer tools. But over time, the complexity of software development has made developers lose their original focus and fluency. Now, we use Copilot, an AI programmer, which can help write mundane code and allow programmers to focus on more difficult problems. Now, 100 million developers on GitHub can enjoy programming. As AI transforms the way programming is done, it can scale 10-fold — 100 million can become 1 billion. When you give a hint to a large language model, you're programming it.

Levy: Can anyone who has a smartphone and can talk become a developer?

Nadella: Absolutely. You don't have to write equations or master grammar or algebra. If you think that typing hints equals development, it's easier to learn. You can even ask, "What is development?" "It will become widespread.

About how to make this technology work 8 billion people around the world. For example, I went to India in January this year and saw a very powerful demonstration. The Indian government has a program called Digital Public Goods, one of which is text-to-speech systems. In the show, a rural farmer uses the system to query for a subsidy scheme he saw on the news. The system introduced him to the plan and the application forms he could fill out. In general, he will be told where to pick up the form. But there was an Indian developer who trained GPT with all the Indian government documents, so the system automatically completed the form for him, in another language. What was born a few months ago on the west coast of the United States has now spread to a developer in India, who went on to write a module that would allow a rural Indian farmer to use the technology via WhatsApp on his phone. My dream is that every of the 8 billion people on the planet can have an AI teacher, an AI doctor, an AI programmer, or an AI consultant.

Levy: It's a beautiful dream. But generative AI is an emerging technology, and it's also a bit mysterious. We don't know exactly how it works internally. There is still some bias, and some believe that it is too early to use it widely. Google, for example, has had generative AI technology for years, but out of caution, it has been slow to make progress. You've applied it to Bing and challenged Google to do the same, regardless of its concerns. You once said, "I want people to know that we let Google dance to our rhythm." And Google did dance to Microsoft's rhythm, changing its strategy and entering the market with Bard, a generative AI search product. I don't want to say it's a risk, but it's understandable that because you boldly embed AI into Bing ahead of time, you start a cycle of panic in which competitors large and small join regardless of whether their technology is mature or not.

Nadella: In a way, the beauty of our industry is that it doesn't care who has the ability, but who can put that capability into practice and turn it into a valuable product. If you want to debate this, you can go back to Xerox Parc Research or Microsoft Research and say that everything they developed should be stranded. The question is, who can make something beneficial that really moves the world forward? That's what I think we should do. Who would have thought that search engines would become interesting again? Google has done a very good job in this regard, and it dominates the industry with excellent products and distribution. Google Search is the default search engine for Android, the default search engine for iOS, the default search engine for many popular browsers, and so on. So I said, "Let's innovate and change the search paradigm to make Google's 10 blue links look like AltaVista!" ”

Levy: You're talking about the '90s search engine (AltaVista), which immediately fell behind when Google's innovation left it behind. This metaphor is cruel.

(Editor's note: AltaVista, one of the world's most well-known online search engine companies, also provides search engine back-end technical support and other related products, was acquired by Yahoo in 2003 and closed by Yahoo on July 8, 2013.) ）

Nadella: Actually, after I used Bing Chat, I didn't want to use anything else, not even the previous Bing. This makes absolutely no sense. So I'm happy to have Bard and Bing now. Let them compete fiercely, so that people can experience the charm of innovation.

Levy: I think you're excited to finally launch an innovation in search that has brought attention to Bing. I remember how frustrated you were when you ran Bing in 2009, when it seemed like you were chasing a competitor you couldn't beat. Now, with AI, are we at an inflection point, where the position of the industry is reshuffled and the winners who were originally dominant become vulnerable?

Nadella: That's right. In a way, each change brings us closer to the vision described in Vannevar Bush's article ("How We Might Think," published in The Atlantic in 1945, which first presented a picture of a computer-dominated information paradise. The question is, how to really achieve this from Bush to Joseph Ricklyde. C. R. Licklider (1960 conceived of "the symbiosis of humans and computers"), to Doug Engelbart (who invented the mouse and windows) and Alto (Xerox Parker Research Center's graphical interface personal computer), to the success of a series of turning points in personal computers and the Internet. It's all about "Could there be a more natural interface that allows us as humans to improve our cognitive abilities and do more?" "So that's an example. Copilot is a metaphor because it's a human-centered design choice. So don't think of our product as autonomous driving – it's a co-pilot. A lot of people are saying, "Oh my God, AI is coming!" Or "You know what?" AI is already pervasive. "In fact, all behavior targeting uses a lot of generative AI. It is a black box with an unknown internal structure, and you and I are just being targeted.

Still bullish on mixed reality

Levy: In my opinion, the future will be a tug-of-war between "co-pilot" and "autopilot".

Nadella: The question is, how do humans harness these powerful abilities? One way to do this is to align the model itself with the core human values we value. These are not technical issues, but more socio-cultural considerations. On the other hand, design decisions and product manufacturing are made according to specific situations. This requires ensuring that the context in which these models are applied is coordinated with security.

Levy: Are you patient with the voices that say, "We should stop developing AI for 6 months?"

Nadella: I have a lot of respect and would like to take the time to listen to anyone say, "Let's really think about all the difficult challenges of alignment, and make sure we don't have AI that is out of control." "If AI technology advances by leaps and bounds, we'd better be in control. Thinking back to when steam engines were first deployed and factories were created, can we avoid centuries of horrific history if we consider both child labor and factory pollution? So whenever we're excited about a new technology, it's rewarding to think about the unintended consequences it can bring. At this stage, though, I'm not just saying stop, I'm going to say we should step up and do a good job of creating these alignments. The first day I saw the GPT-4, we didn't immediately launch Sydney (the internal codename for the Bing chatbot) because we needed a lot of work to build a security fixture. But we also understand that we can't do all the alignment in the lab. To align an AI model with the world, you have to align it in the real world, not in some virtual world.

Levy: So do you know about Sydney's claim to be in love with New York Times reporter Kevin Roose?

Nadella: We didn't expect someone to do an eight-dimensional (personality test) analysis of Sydney within 100 hours of launch.

Levy: Do you think AI has the potential to destroy humanity?

Nadella: If there is a situation that completely gets out of control, that's a problem and we shouldn't allow it to happen. We have the ability to deal with powerful technology. In fact, electricity has also brought some unexpected effects, we have ensured the safety of the power grid, established regulations, and taken protective measures. Similarly, with regard to nuclear energy, we have addressed the dangers of nuclear proliferation. In both examples, we can learn how to handle powerful technologies.

Levy: Right now, one of the big problems with big language models is that they create illusions, like Sydney and other language models, they just create things haphazardly. Is there a way to solve this problem effectively?

Nadella: There are actually some very practical ways to reduce the occurrence of hallucinations. And the technology is constantly improving. There will definitely be a way. But sometimes illusions are also a sign of "creativity". Humans should be able to choose which mode they want to use.

Levy: That would be an improvement because now we don't have a choice. But I would like to ask about another technology. Before, you were full of praise for the metaverse. In 2021, you say mixed reality is a huge breakthrough. But now we're only talking about AI. Has this craze made the metaverse gradually disappear?

Nadella: I still believe in the value of [virtual] existence. In 2016, I mentioned 3 things that excited me: mixed reality, quantum, and AI. Now I'm just as excited about these 3 things. Today we're talking about artificial intelligence, but I think existence is the ultimate killer app. Then, of course, quantum accelerated everything.

Levy: Right now, AI isn't just a topic of discussion. You've put Microsoft at the center of this transformative technology. How do you manage it?

Nadella: I often use an analogy to illustrate that when we switch from steam engines to electricity, you need to rewire the factory. You can't just put the motor where the steam engine was and nothing else moves. That's the difference between Stanley Motor Company and Ford Motor Company, and Ford is able to change the entire workflow. So within Microsoft, the way software is made is changing. It's a huge change in basic workflows within Microsoft, and it's how we communicate our results — and how it changes every school, every organization, every family.

Levy: How has AI Copilot changed your job?

Nadella: A lot of intellectual work is drudgery, like email sorting. Now, I don't know how I would live without AI Copilot in my mailbox. Responding to an email is not just writing in English, it can also be a customer support ticket. Copilot will consult my customer support system and bring the relevant information back. It's like the feeling of when personal computers first appeared in the workplace and permeates every aspect of our products.

Levy: Microsoft did well during your tenure, but do you think you'll be remembered for your AI transformation?

Nadella: It's up to someone like you or someone else to say what I'll be remembered for. But I'm excited about it. Microsoft has been around for 48 years. I don't know how many companies that are so old can still remain relevant to the world, not because of what they did in the '80s, '90s or 2000s, but because of what they've done in the last few years. As long as we do that, we have the right to live. And when we don't, we shouldn't be seen as a great company.