# Former Google CEO On The Importance Of Human Expertise In An Increasingly Automated World *Forbes* ยท https://www.youtube.com/watch?v=LOlejWyWJZ8 ## Summary This Forbes conversation features former Google CEO Eric Schmidt and Deloitte CEO Joe Ucuzoglu answering audience questions about artificial intelligence, business, and careers. Schmidt argues the agentic period is the best investment opportunity, estimates China is six to 12 months behind the US, and frames AI as the historic arrival of nonhuman intelligence. Ucuzoglu stresses that domain expertise and cross-functional teams are key to enterprise AI adoption, and both agree the technology will not be stopped. ## Key points - Larry Page's PhD subject was AI, and Google's first real internal use of AI was improving ad system performance. - Schmidt says the best way for an average person to make money on AI is to found an agentic AI company that builds an agent to do something. - America needs an estimated 80 gigawatt of capacity for AI data centers in the next 3 to 5 years, with 1.5 gigawatt being the size of a nuclear power plant. - Job loss from AI is showing up in customer service and entry-level software programming. - China is judged to be six to 12 months behind the US, with models like Qwen and DeepSeek near the top of rankings and a full self-sufficient hardware and software stack. - Deloitte's CEO argues AI works best when technologists are paired with business people who understand the function and industry. - Google's 20% time worked as a balance of power between creative engineers and management. - Schmidt calls this the moment in history when a nonhuman intelligence arrived as a competitor to humans, and how we shape it will determine the outcome. ## Notes ### Google's AI roots and investing advice - Larry Page's PhD subject was AI, and Google's first real AI use was improving ad system performance. - Henry Kissinger co-authored the AI book because studying Kant led him to worry AI would affect human consciousness. - For average investors, the best play is not the Magnificent 7 or highly-valued startups, but founding an agentic AI company that builds an agent to do something. ### Infrastructure and jobs - AI contributes more than 1% of US GDP, largely via data center buildout; hyperscalers are asking for 1, 5, and 10 gigawatt, and the industry may need 80 gigawatt in 3 to 5 years. - Real job losses are appearing in customer service and entry-level software programming. - Deloitte's Joe Ucuzoglu says 500,000 headcount is higher than a year ago; great professionals with domain expertise unlock the technology, and each prior wave created more jobs than it disrupted. ### Competition with China - Chinese models like Qwen and DeepSeek rank near the top, and China builds a fully self-sufficient stack including its own chips. - China is not chasing AGI but AI used everywhere in society, especially on the edge; Schmidt judges them six to 12 months behind. ### Careers and enterprise adoption - The best skill set integrates disciplines: baseline technology understanding plus knowledge of the business problem. - The 40-year-old humanities grad should learn prompt engineering; the tools are remarkably powerful for non-technical people. - Deloitte hires 'best available athletes,' and the generalist still has a big role. ### Google history and culture - Google is a misspelling of googol, 10 to the 100th; the trademark was taken. - PageRank, invented in Larry's Stanford dorm room, made Google's ranking better. - Free food came from treating the company like a family; 20% time served as a balance of power between engineers and managers. - An engineer named Omid squatted in Schmidt's office, listened to his calls, and gave the true revenue number, 140 versus the sandbagged 115 million. ### The big picture - The technology will not be stopped; it can be used for healthcare, drugs, and energy, but also oppression, war, and addicting youth. - This is the historical moment when nonhuman intelligence arrived as a competitor; alignment and how we shape it will determine the outcome. ## Transcript [00:05] So Eric, when did you when did Google first start thinking about artificial intelligence? >> Larry Page was his PhD subject was AI. So even when Larry started the company, he wanted to do AI relatively interesting things. And the first real use of AI was we used AI to improve our ad system performance. This is before LMS and so forth and that worked. That was the first time I saw it used inside the business. >> Now you've written a book with on artificial intelligence. One of your co-authors was Henry Kissinger. Did Henry Kissinger really know anything about artificial intelligence? Why was he the co-author of that book? Well, he was uh he's obviously a brilliant man and he was very interested in it not for [00:54] the technical reasons but because he had studied Kant when he was uh when he was the age of a college student well before any of us was born and when he saw AI when Demis presented to him he thought it would have an impact on human consciousness and he was very worried about it. Hey Joe. Um, usually when I'm interviewing people about artificial intelligence or something like that, there are people that have engineering backgrounds, their technology backgrounds. You're an accountant. So, what are you doing on a panel on artificial intelligence? >> Fortunately, of those 500,000 people at Deote, I have some good friends who are computer engineers and scientists and have deep expertise in the systems [01:41] themselves. So, I I've got a little bit of help. How many people are devoted to artificial intelligence out of your 500,000 people? >> I would define every one of them is in some way leveraging the technology, helping clients use it. You can define a subset that obviously are the deep experts, but it's now pervasive across the entirety of the organization. >> Eric, if I want to make money on artificial intelligence, I'm an average investor. What is the best way for me to make money today on artificial intelligence? buying a big company like uh Alphabet or finding a smaller one that's just getting off the ground. >> Well, one way to understand the value that's being created in AI is that it will there will be businesses that will [02:25] become much more successful because of AI. So, I think the best investment is to invest in the businesses where AI will make them transform. Um, it's probably pretty late to have all the gains that you have in the Magnificent 7, although it's not my area of expertise, simply because those are very highly valued. And the startups are now so highly valued when they get started that it's really hard to get into them. Um, if you really want to make money, it's actually easy. Found an agentic AI company. And I don't mean one designing agents. I mean build an agent to do something. This is the agentic period in AI and for the next year or two, everyone's going to build agents. The agents are all going to compete. If [03:11] you build a better agent than anyone else, you can probably build one incredible company and own it yourself. >> Eric, um right now, um artificial intelligence is affecting adversely which um part of the economy the most and it was affecting positively which part of the economy the most. Uh well AI is at the moment contributing more than 1% to the GDP of America largely because of the data center buildout and I just started a data center company so I've been studying this the demands the the real demands from the hyperscalers the big companies Google and so forth are immense and when I talk to them they need 1 gawatt 5 gawatt 10 gawatt each now the best study I've seen indicates that the industry in [04:00] America needs 80 gawatt in the next 3 to 5 years. Now 80 gawatt by the way let me tell you how big is that 1.5 gawatt is the size of a nuclear power plant. So this is an enormous amount of energy. So that's the the economics right now are being most felt in the buildout of the infrastructure for the next wave of AI. Joe uh if you have 500,000 employees at Deote how many of them are going to be losing their jobs because of AI so with all of the adoption of technology that half a million today is greater than the number that we had a year ago will there be some displacement and shifting in certain functions that one did manually of course as with every wave of technology adoption but we're still a big believer that great people [04:54] great professionals with domain expertise are actually the key to unlocking the technology and that as with every prior wave of technology change on a net basis you actually create more jobs than you disrupt that'll be tested but that's our base case >> okay so in your consulting business what has been some more significant the rise of the internet or the rise of artificial intelligence >> well if you look at the rise of the internet and the impact over a 30 plus year period in the acid to date that's more significant but in a confined time period in terms of this really thrusting onto the scene over the last 3 or so years it is a more accelerated impact than what we would have seen with the internet and right now I don't think [05:40] anybody doubts the power of the technology the applicability across the enterprise every sector every business function the issue is it's really hard to actually bring it to life in a large complex established bureaucratic enterprise. You've got to actually be willing to drive a fundamental endtoend change in the business processes. And history tells us that there's a lot of stickiness in the system. And that diffusion from the the lab in Silicon Valley to the actual heartland to the factory, the office that people work in, it takes time. It takes effort. So we're in the middle of a a massive transformation that's going to take many years. >> Yeah, David, let go ahead. So this has always been true of technology adoption [06:30] and the industry calls this the technology overhang. The industry we arrogantly say we keep producing things that you can't absorb fast enough. Of course, if we made it easier, maybe it would would on the jobs. There's beginning to be evidence of real job loss in two categories. One is in customer service, which is highly automated now, and the other one is in low-end or entry- level software programming, my field. Uh, and that shouldn't surprise you because the software programming is getting so good, you really can change your modality and you should expect that that will continue. The real question is when do other when do other parts of jobs get affected? >> Let's suppose I miss the internet uh [07:15] wave and I miss the AI wave as an investor, but I'm always looking for the next big wave that I can get on the ground floor. What is the next big technology wave that I could get on the ground floor of? Is it is it quantum computing or some other thing uh like uh uh nuclear fusion? What would you say is the next big wave I should invest in if I want to be on the ground floor? >> So the next big wave is AI, right? We're not done and in fact the wave is just getting started. That's not I don't agree with you David. Um we're going from language to language which is think of it as chat GBT to language to action through through agents and if you spend any time with deep research or spend some time with open AAI R3 spend some [08:01] time with DeepSeek V3 whatever the power of the reasoning models of these things is frightening right it's graduate level MIT Stanford reasoning right at your fingertips we're not done yet and indeed the investments in data centers, the new algorithms, the much larger models are entirely focused on that. So, we're not done. >> Okay. Um, Americans seem obsessed with China and China's developing AI at better pace and maybe in better ways than the United States. How do you today rate the competitive abilities of the Americans versus the Chinese on AI development? >> So, um, the US invented it and the Chinese have adopted it in a very clever way. So historically there was a view that China was well behind and China has [08:51] in the last few years caught up really really well. Uh let me give you some examples. There are three models Kimmy Quinn and Deepseek which are often at the top of the rankings. This makes the Americans annoyed because Americans claim that the Chinese did something called distillation and used the American models to train them. That may or may not be true. The important thing is the Chinese models are really good. Furthermore, China has decided to be completely self-sufficient in this technology. So, they're building the entire hardware and software stack. They have their own chips called an SN930. They have their own software layers and so forth. The Chinese model is also different. They're not going for AGI or [09:34] general intelligence. They're going for everything AI used everywhere in the society. And it's a massive massive program within the country to use AI everywhere especially on the edge. They are in my view six to 12 months behind which is not much. >> Now you've been an adviser to the defense department in a number of for a number of years. U how has AI changed warfare or is it going to change warfare? >> So the the the US government started a program called Maven. You'll you all will hear remember this about 10 years ago and I was involved with it because I was a consultant to the Pentagon during that period and Maven was can be understood as uh drones are running around looking at targets in the in in [10:26] the at time in the military and those video feeds were being watched. So the first use in the military is actually just surveillance. Why? Most soldiers in the military are just watching for things. That was the first and best use. The next one will be drones. And because of what's going on in Ukraine, the US government has decided to to learn from Ukraine. Now there's a Secretary Hexith has announced a big program around drones and drone adoption, land, sea, and air. Okay. Um, if you were advising a young person today who's graduating from college, uh, should they join a large accounting consulting firm like yours and have an accounting background as you do or should they get a technology background in AI? Which one [11:15] is going to be more exciting and make more money for that young person? I actually don't think this is an eitheror and that in fact what we're finding in the application space I agree with you Eric that we're just at the beginning of the wave of AI adoption and I'll use my baseball analogy we're in the the top of the first inning here those who have the ability to integrate across disciplines. So to really bring this to life in an enterprise, you have to both have some baseline understanding of the technology, the capability of the model, how to prompt it to yield the desired outcome, as well as the business problem you're trying to solve if you're trying to embed this into the finance function, the manufacturing function, [11:57] customer and marketing. And where we've seen things go wrong is where you take a piece of technology that's developed by someone in the tech laboratory and you try to deploy that and force it into an enterprise. Generally speaking, it fails. You need the people in the heart of the business who understand that function, who understand that industry and sector to be paired up on a crossf functional team to design something that's fit for purpose in the real world. I thought Yan's words earlier were were very accurate. The real world is far more complicated than text. And so those who have an appreciation for the context and who can work with the technologists to bring it to life, that's where the real value is being [12:42] created. >> Let's suppose somebody in the audience here is uh 40 years old. They majored in humanities in college. they have a job that's uh say a non- tech job, but they really want to learn about AI so they can talk about it intelligently with their children, if not somebody else. What's the best way for a middle-aged non-technology person to actually learn something about AI to sound reasonably intelligent on it? >> So, let me turn the question around. We have MIT and Stanford here. The most important new course that you don't have is the freshman class on prompt engineering. Because if you think about it, the student that is entering MIT or Stanford, their entire life is going to be prompt [13:30] engineering as it evolves because everything they do will be with a computer assisting them. So the same applies to the 40-year-old, right? So the 40-year-old needs to get good, too. These tools are remarkably powerful for a non-technical person. Um, if you wanted to write code, you wanted to do something, have fun, learn how to do it. They will produce working systems that can make your life better. Okay. So today um a lot of the big technology companies that are uh have trillion dollar market caps, they seem to be moving away from everybody else in terms of their ability to um develop AI models and the ability be a leader in it. What about the smaller companies? Is there a chance for a smaller company today to [14:16] become uh a gigantic uh alphabet or is that day gone already too late to become gigantic again in the technology world? So, um, the big companies are getting bigger because their revenue is growing fast. And it's growing fast because they've been around for a while and they have a lot of capital and they have a lot of data centers, right? That's what's really going on. They're not just being nice and the valuations are going up. So, for smaller companies, the real question is how do they get the capital to play in that game? It's going to get harder and harder and harder as the industry matures. And let me ask you about Nvidia. And Nvidia is making a lot of money on artificial intelligence. Is [14:58] anybody else really making money or are they just spending money on artificial intelligence? Look, when you when you look at where the rents have accred to date, obviously the sort of infrastructure layer at the front end as people are building all of this out has captured a a disproportionate share. The real issue from a societal standpoint is how to take those gains and spread them broadly and actually drive the productivity benefit into the heart of the of the enterprise system. And frankly, it's a societal imperative if we're going to earn the confidence of people that in an environment where so much fear is being attached to the technology and some of the risks, the sort of misinformation, [15:44] the potential job displacement, and that just spurs up backlash. It spurs up political challenge. We have to be able to demonstrate that for the average citizen, benefits are going to acrue, that there's going to be incremental economic opportunities, that there's going to be improvements to the quality of human life, that there's going to be some incremental value in terms of educating their children. And so that process of taking the technology and making it work in the lives of real people, we're in the really early stages. All >> right. Let me ask you, for your company, you have 500,000 employees roughly. All right. Suppose somebody wants to be an employee in your company now. They're graduating from college and they are in [16:24] a literature major or they majored in social science subjects or humanities subjects. They have any chance of getting hired. And if you get hired by your firm, do you have to be a STEM major today? And how do these non- STEM majors fair? Do they actually wind up staying there after a year or two or they get washed out? I want best available athletes. Now they have to want to acquire the skill sets that Eric shared in terms of the prompt engineering in terms of understanding the core of the technology and how that sort of applies in terms of a modern business enterprise. But those who were deep thinkers and in fact those who can best understand the societal implications of this and how you actually adapt the use of the technology [17:07] in an ethical fashion. I'm still a believer that the generalist has a big role here. >> Okay. Eric, you were the CEO of Noville, right? And then uh two young guys were looking for a CEO for a new company they were starting that came the name Google. Uh first of all, why did they spell Google improperly? So u there's a number called Google G O G O L 10 to the 100th. And the trademark was already taken. So Sergey changed it to LE and the rest is history. We ultimately settled with the family of the mathematician who invented the Google um because they were slightly annoyed at our reframing of that number. >> When you joined the company um people said there are a lot of search engines out here. Why do we need one more search [17:54] engine? What was Google's ability to do something in search that nobody else had done? Was it a better system or better mousetrap? Um, so if you look at technology these days, every wave has five or 10 companies that are similar. In Google's case, it was the last of a series of companies that were doing ranking. But Larry Paige had invented in his dorm room an algorithm called page rank which is mathematically a fastforier transform of the importance of something and he used it for ranking and Google's ranking was just better. He started it in the dor dorm room in Stanford on the campus started to use it. >> Now one of the things you had at Google at the very beginning was something when I toured it you gave me a tour early on. [18:43] You had free food. Um, and that was a I guess a way to keep people on the campus and not waste time going out for restaurants. Was that a good idea? Whose idea was that? >> So, so Sergey and Larry, and I inherited this, believe that a company was like a family and that you should feed them. Um, and you should feed them together because that's how they would all communicate. They had a number of strange beliefs. for example, they loved hiring Olympians because they figured that Olympians were so disciplined to to become an Olympian that they would be great employees even if they knew nothing about the job that they were working on. That turned out to be true, too. Um, so so in any case, the the free [19:24] food has continued and I think it's now standard in >> they also had a thing called 20% time. You could spend 20% of your time doing something else. Would that work out? >> Uh, it did. Um, and so Google had a rule that every employee, at least the technical ones, could spend 20% of their time on what they wanted and not what their manager was. Now, you sit there and you go, how is that make any sense at all? But the way it worked was if the boss came in and said you have to spend all of your time on this and was very manipulative. It was a control on the power of the management because you could always say yes I'll give you four fifths of my day of my year my week but I'm not going to give you all of it. It [20:10] was a balance of power between the creative engineers and their management. And a final question about the early days of this company. You were the CEO when you went on a trip. Somebody moved into your office. You're the CEO and somebody took your office and you couldn't get him to leave. Why was that a good thing to do? >> Well, I I had left I had left town for the for a sales call and I came back and I found a fellow named Emit in my office. He was at 8 by12 office sitting next to my little table. And I said, "Who are you?" And he identified himself and I said, "Um, who gave you permission?" And I said, "I asked Wayne who worked for me." and Wayne said, "Sure, move into Eric's office." Now, [20:49] this is a test as an executive because when this happens, you can say, "Get the hell out of my office." Which is culturally not correct. So, I figured, okay, I'll just sit there and see what happens. So, he wore he was a programmer and he wore headphones. And I used to talk on the phone and I was talking to the sales executive at the time. We didn't know what our revenue was. And he said, "I think the revenue is going to be 115 million for the quarter. Imagine these numbers." And I said, "I think you're sandbagging," which is a term in sales for not telling the truth. Um, so I have the conversation. I don't make very much progress. And so I put the phone down and a meet takes his headphones off and says, "I can tell you [21:32] the revenue number." And I said, "I knew you were listening on my phone." He'd been listening to all of my phone calls. And so what happens is I said, 'What's the revenue number? And he said 140. And so I made Omid present all of his sales forecastes standing on a sand sandbags as punishment for lying to me. >> So has anybody ever squatted in your office? >> Uh not yet. I'm old. >> My advice to those of you as CEO is get the person doing the revenue in the company in your office and watching it. And the simplest rule for those of you who are CEOs is the only thing that matters in the CEO's life is revenue. Solves all problems. Now you're running I'm sorry. We became best friends. You're running Deote. How many people [22:19] come into you and tell you bad news as opposed to people coming in to tell you good news? The ratio is undoubtedly higher good to bad, but not [clears throat] not as detrimental as you'd think. And in fact, I I actually think the number one dimension to effective leadership in an environment like we have today where there's so much disruption and so much uncertainty is creating a culture where there is a willingness to surface problems where you see a top- down command and control and there's a fear and all the dashboards are green or or that is a recipe for an organization that will be disrupted where you create a culture where people are actually willing to keep an open mind even if the strategy said X, the market is shifting, [23:09] the environment is shifting, the technology is shifting, we're going to challenge our assumption, we're going to be sort of embodying a learning mindset. Those are the organizations that are actually going to thrive today. So you you need a lot more willingness to hear things that are unpleasant. And let me ask you about artificial intelligence this respect today in colleges when people get exams uh that say they're take-home exams uh and they write them up. How do they professors know that they're really not written by AI? And as I understand it, there is now some software that can determine whether it's been written by AI, but there's software that can determine that the how to fool the professor's software. How does that [23:48] work? I've got two kids in college right now and this dominated the conversation over the winter break. The worst thing that an educational institution could possibly do is ban the usage of AI, ban the usage of the tool that's actually going to help make the individual more effective. And yet you have this huge dilemma because if all the student is doing is putting the assignment into the AI engine and turning in something that the LLM spit out, that is a useless earning learning exercise. And so how you actually allow for the usage but still demand the human overlay and the critical thinking and how did I take the output and use it to push myself. And frankly, these software programs that are detecting what percentage was [24:36] actually derived by AI, they have some directional sense, but there's a lot of false positives. >> How has your job changed by AI? Do you use AI yourself? You go and chat GBT all the time to get answers to things. >> Look, I I use it constantly in my personal life. I find it actually makes me much more effective at any variety of things that we come across in our day-to-day life. And I am embracing and trying to lead by example in terms of the tools, the demos, the proofs of concept, and then ultimately those things that look like they're going to scale that we're putting into production of using them myself, of providing feedback in terms of trying to help train the model so that where there are [25:18] things that are valuable, feed that back and let's capitalize it. And also calling out the meter that things that look great on a screen and to an average uh recipient of the information appear to be very impressive, but to somebody who's an expert in the area is a bunch of gibberish. And that's the real dilemma right now. This stuff looks really impressive on the surface. And some of it is in substance and some of it's garbage. And it takes the human who has the domain knowledge to actually pick between the two. Eric, uh, bad news usually sells better than good news. People like to read bad news, it seems, and that's why CNN or the equivalent shows they when they have a crisis, it ratings go up and so forth. Um can you [26:02] to lead finish this uh conversation by making people feel that AI is going to be a good thing for civilization not a bad thing and let people leave here thinking that if they know more about AI they'll be better off and living in a world with AI being more and more important is a good thing for everybody. >> So we have some choices as humans and those choices are really important to get right. So first place, this technology is going to happen. It's not going to get prevented. It's not going to get stopped. There's too many countries, too many people, too many incentives. It's going to happen. So what does this mean? It means great great solutions for healthcare, new drugs, better energy solutions, better [26:49] power distribution, better physics, better chemistry, better sociology, better research. It also means that it can be used for bad. It can be used for example for oppression. It can be used to limit freedom in governments hopefully not in the west. Uh it can be used in war. Um the technology itself is so addictive that it can affect our young people and we should make sure that our young people are protected from some of the worst parts of the technology. I could go on. But we face choices. I've written two books on this. We face choices now about how we want to deal with this incredibly powerful technology. I will tell you and it's really important to understand that we are living through a moment that will be [27:32] in history for thousands of years. It is the moment in history when a nonhuman intelligence arrived and it was a competitor to us. Right? How we use that, how we shape it, right? How we guarantee alignment will determine the outcome of your >> final question. If you had the chance to rename artificial intelligence and give it a name that's more appropriate, would you change the name or would you just say it's better? The best name is artificial intelligence. >> The the best name is artificial intelligence because it is intelligence and it's artificial. Remember that its intelligence is not the same as human intelligence. It doesn't have the same values. It doesn't think the same way. It doesn't think at all. It's a it's a [28:12] different kind of intelligence. When we work with it, we will be more powerful. When it runs us, we will be very unhappy. >> Is it harder to learn artificial intelligence or pass the CPA exam? Well, those two are arguably becoming one and the same. That's one of the tests that's been used to define the efficacy of the technology. But we do have a real issue of education around the societal benefit. And there's some recent polling where those who are actually using the technology themselves are far more positive on the potential impact than those who aren't using it. And I I do worry, Eric, the technology is coming. it is going to happen. But we're in this powder keg of a moment society where you see the sort of the [28:58] populist concerns and the fears over the future and then you layer this onto it and all the worries over job disruption. Our number one job as leaders is to actually go out and tell the story and make this real in terms of how this is going to improve the quality of human life. >> Thank you both for a very interesting conversation. How many people here feel they know a lot more about AI than they did a half hour ago? >> Okay, thank you very much. >> Thank you, Dave. Thank you.