- 5시간 전
- #2424
▲ "AI는 일자리를 없애기보다 일의 본질을 바꾼다"…자동화의 새로운 물결
▲ 엔비디아 다니는 두 자녀에게도 "스톡옵션 행사 가능해지면 바로 팔아라"
▲ "한국, AI 모델 경쟁보다 잘하는 산업에 AI 접목해야"
YTN 글로벌 명사 대담 6부작 가 네 번째 순서로 실리콘밸리 1세대 창업가이자 베스트셀러 『인간은 필요 없다』의 저자인 제리 카플란 미국 스탠퍼드대 겸임교수를 만났다. 수십 년간 실리콘밸리에서 기업을 경영해 온 그는 AI 시대 노동과 경제의 변화를 오래 전망해왔다. YTN 제작진은 미국 현지에서 카플란 교수와 단독 인터뷰를 진행했다. 특별 인터뷰 편은 8월 22일(토) 밤 11시 15분 방송됐다.
"AI는 자동화의 새로운 물결"…바뀌는 건 일자리 아닌 '일의 본질'
카플란 교수는 AI를 산업혁명에서 인터넷으로 이어져 온 '자동화의 새로운 물결'로 규정했다. 모든 개인이 부유해지는 것은 아니지만, 과거 자동화가 그랬듯 생산성을 높여 사회 전체의 부를 키운다는 전망이다. 실제로 미국에서는 지난 200년 동안 평균 가계 소득이 약 40년마다 두 배로 늘었고, 한국은 그 변화가 더 극적이어서 1950년 이후 75년 만에 농업 사회에서 산업 강국으로 올라섰다고 그는 설명했다.
AI가 전문직까지 대체할 것이라는 우려에는 1980년대 개인용 컴퓨터와 스프레드시트가 회계 업무를 바꾼 사례를 들었다. 계산 업무가 자동화되면서 회계사가 사라질 것이라는 전망이 나왔지만, 오히려 업무 영역이 기업 분석과 사업 전망 등으로 확장됐다는 것이다. 그는 "숫자를 채워 넣고 계산하는 일만 할 줄 안다면 일자리를 잃게 된다"면서도 "도구를 활용해 사업 전망을 예측하거나 사업계획서를 작성할 수 있다면 그것이 중요한 전문성이 된다"며, AI 역시 직업 자체를 없애기보다 기존 일자리의 핵심 업무와 필요한 능력을 바꿔놓을 것이라고 강조했다.
"AI 기술의 가능성과 투자는 별개"…엔비디아 다니는 자녀에게도 "팔아라"
AI 기술의 장기적인 가능성에는 낙관했지만 투자 시장에 대해서는 경고했다. 카플란 교수는 현재 AI 산업이 투자 관점에서는 '거품의 한복판'에 있다고 진단했다. AI가 유용하고 혁신적인 기술인지와, 지금 쏟아지는 막대한 투자가 수익으로 이어질지는 별개의 문제라는 것이다. 그는 오늘 지은 데이터센터가 3~5년이면 구식이 되고, 스무 곳이 넘는 기업이 각각 수백억 달러를 쏟아붓는 지금의 경쟁을 '신... (중략)
▶ 기사 원문 : https://www.ytn.co.kr/replay/view.php?idx=291&key=202608201013474396
▶ 제보 안내 : http://goo.gl/gEvsAL, 모바일앱, social@ytn.co.kr, #2424
▣ YTN 데일리모션 채널 구독 : http://goo.gl/oXJWJs
[ 한국 뉴스 채널 와이티엔 / Korea News Channel YTN ]
▲ 엔비디아 다니는 두 자녀에게도 "스톡옵션 행사 가능해지면 바로 팔아라"
▲ "한국, AI 모델 경쟁보다 잘하는 산업에 AI 접목해야"
YTN 글로벌 명사 대담 6부작 가 네 번째 순서로 실리콘밸리 1세대 창업가이자 베스트셀러 『인간은 필요 없다』의 저자인 제리 카플란 미국 스탠퍼드대 겸임교수를 만났다. 수십 년간 실리콘밸리에서 기업을 경영해 온 그는 AI 시대 노동과 경제의 변화를 오래 전망해왔다. YTN 제작진은 미국 현지에서 카플란 교수와 단독 인터뷰를 진행했다. 특별 인터뷰 편은 8월 22일(토) 밤 11시 15분 방송됐다.
"AI는 자동화의 새로운 물결"…바뀌는 건 일자리 아닌 '일의 본질'
카플란 교수는 AI를 산업혁명에서 인터넷으로 이어져 온 '자동화의 새로운 물결'로 규정했다. 모든 개인이 부유해지는 것은 아니지만, 과거 자동화가 그랬듯 생산성을 높여 사회 전체의 부를 키운다는 전망이다. 실제로 미국에서는 지난 200년 동안 평균 가계 소득이 약 40년마다 두 배로 늘었고, 한국은 그 변화가 더 극적이어서 1950년 이후 75년 만에 농업 사회에서 산업 강국으로 올라섰다고 그는 설명했다.
AI가 전문직까지 대체할 것이라는 우려에는 1980년대 개인용 컴퓨터와 스프레드시트가 회계 업무를 바꾼 사례를 들었다. 계산 업무가 자동화되면서 회계사가 사라질 것이라는 전망이 나왔지만, 오히려 업무 영역이 기업 분석과 사업 전망 등으로 확장됐다는 것이다. 그는 "숫자를 채워 넣고 계산하는 일만 할 줄 안다면 일자리를 잃게 된다"면서도 "도구를 활용해 사업 전망을 예측하거나 사업계획서를 작성할 수 있다면 그것이 중요한 전문성이 된다"며, AI 역시 직업 자체를 없애기보다 기존 일자리의 핵심 업무와 필요한 능력을 바꿔놓을 것이라고 강조했다.
"AI 기술의 가능성과 투자는 별개"…엔비디아 다니는 자녀에게도 "팔아라"
AI 기술의 장기적인 가능성에는 낙관했지만 투자 시장에 대해서는 경고했다. 카플란 교수는 현재 AI 산업이 투자 관점에서는 '거품의 한복판'에 있다고 진단했다. AI가 유용하고 혁신적인 기술인지와, 지금 쏟아지는 막대한 투자가 수익으로 이어질지는 별개의 문제라는 것이다. 그는 오늘 지은 데이터센터가 3~5년이면 구식이 되고, 스무 곳이 넘는 기업이 각각 수백억 달러를 쏟아붓는 지금의 경쟁을 '신... (중략)
▶ 기사 원문 : https://www.ytn.co.kr/replay/view.php?idx=291&key=202608201013474396
▶ 제보 안내 : http://goo.gl/gEvsAL, 모바일앱, social@ytn.co.kr, #2424
▣ YTN 데일리모션 채널 구독 : http://goo.gl/oXJWJs
[ 한국 뉴스 채널 와이티엔 / Korea News Channel YTN ]
카테고리
🗞
뉴스트랜스크립트
00:12미술 전체의 종류의 조합을 벌어주는
00:16다른 사회의 종류의 종류의 권한 폭력이 됩니다.
00:20.
00:21.
00:25.
00:25.
00:26.
00:26.
00:28.
00:29Q&A의 아티스트와 나의 계절을 말하는 것
00:44I think we're in the middle of a big AI bubble.
00:48it's getting pretty bad and my advice would be
00:52don't invest in artificial intelligence.
01:05먼저 시작해볼까요?
01:08저는 이즈의 기술이 제일 좋은 기술을 했죠.
01:11저는 수학과 기술이 많은 학생을 했죠.
01:16그리고 저는 지국의 사회에서 지국을 하고 있습니다.
01:20그리고 지금 이 기술을 학생을 했죠.
01:25Stanford, where I'm an adjunct lecturer.
01:58Well, the best way to understand what artificial intelligence is going to do to society in
02:05general and of course to people's jobs in particular is to understand that it is a new
02:12wave of automation.
02:14The main effect, which is important to bear in mind, is they make us much wealthier as
02:20a society in general, in total.
02:23That doesn't mean every individual necessarily gets wealthier, but society in general does.
02:31Previous waves of automation, from the industrial revolution all the way through the internet,
02:38have had exactly this effect.
02:40And the net result has been, here in the United States, that over the last two centuries, the
02:47average household income has just about doubled every 40 years.
02:52Now, I know it doesn't seem like that necessarily, but Korea, it's been even more dramatic.
02:58If you look at the average household income in 1950 in Korea, and the average household
03:05income today, it's many, many times greater today than it was then.
03:10Korea has gone over that 75-year period from being a relatively agrarian society of farms and growing food to
03:22being an amazing and powerful industrial powerhouse as a country.
03:29So, it's now, of course, among the wealthiest and most developed countries in the world.
03:36And it's a wonderful place and I love it.
03:38But, the other thing that it does is that automation changes the nature of work.
03:44And others are created how to work those machines, how to build those machines.
03:49So, I don't think it's realistic to think that jobs are going to suddenly go away and we're all going
03:56to have no work.
03:57There will be plenty of work in the future, but the nature of that work will be very different than
04:03it is today.
04:08Well, this is a very broad question.
04:10How is it going to affect the actual jobs that people actually do?
04:14Little by little, we're beginning to get in focus on what's happening and, in some sense, puts them out of
04:23work.
04:23In another sense, it gives them the ability to do more work and to do better work and in a
04:29very, very different way.
04:31Let me give you a historical example that I think points this out.
04:35When the personal computer was first brought into commercial use,
04:41and I'm sorry to say I'm old enough to remember when that actually happened, in around 1980,
04:48it became immediately obvious to people in the accounting profession
04:52that these systems could do work in minutes that would take them weeks.
04:58A spreadsheet didn't used to mean a computer program.
05:01It used to mean a large sheet with ruled columns and you would write numbers and calculate numbers onto the
05:09spreadsheet.
05:09That's where the term came from.
05:11People naturally thought that, well, accountants, they're going to be out of a job.
05:16Well, there are more accountants today than there were back then.
05:20Why is that?
05:21It's because the nature of the accounting job changed and the level of analysis that one can do using those
05:27tools
05:28is so much greater that it's done nothing but create more demand for accounts.
05:33Now, if all you knew how to do was to fill in those numbers and write those calculations, you're out
05:38of a job.
05:39But if you know how to use these tools to do business projections or to write business plans,
05:46then of course this has had a very positive impact and those are valuable skills which you can now sell
05:53as an accountant.
05:58This is a very good question.
06:00My book, Humans Need Not Apply, what I was right about was that AI is going to change a lot
06:06in the nature of work.
06:09And so it was going to put a lot of people out of work.
06:11I didn't really understand that the nature of the work changes and that the new jobs arise as consistently as
06:19the old jobs go away.
06:20And so I don't recommend that you read my book.
06:25To me, you should listen to what I have to say is the important thing.
06:33Never listen to me because I would have said your kids should learn how to program and that was exactly
06:41the wrong thing.
06:46There will be new kinds of work and I would even argue more meaningful work, work that people will take
06:53greater pride in and where they will feel like they've accomplished more.
06:57Much of the work that people do today is really routine, rote work, repetitive work.
07:05The future of work, paradoxically, because of artificial intelligence, is it's going to free us up from all of this
07:12routine work that so many people do and it's going to permit us to spend our time engaged much more
07:21directly with other people.
07:23The jobs of the future are very likely to be ones that involve person-to-person contact, your ability to
07:31express empathy, to be able to listen, to make a person feel like they've been heard, to understand their problem
07:40and to suggest ways in which they might be able to solve those problems.
07:48The truth is that, of course, a lot of people are going to find that their jobs either go away
07:56or are changed.
07:57Bernie Sanders is right in the sense that probably tens of millions of jobs are going to be eliminated over
08:05the next 10 to 20 years.
08:07But that doesn't mean that people will be out of work.
08:09What it means is that the nature of work is going to change and there will be new kinds of
08:13work and new kinds of jobs that will employ those people.
08:17And probably in much more emotionally satisfying and psychologically valuable ways.
08:53I think it's a mistake, really, to think of these machines in terms of intelligence.
09:00When people think about human intelligence, they have this myth that it's some kind of linear measure.
09:12So intelligence is not a simple linear scale of some kind and that machines are getting higher and higher on
09:18that scale.
09:19All that's happening is we're developing machines that can perform different classes of tasks.
09:25Now, we tend to think of these LLMs as being intelligent because they're so good at mimicking the way in
09:32which we talk
09:33and the way in which we interact that previously we had to perform using our native intelligence.
09:40That doesn't mean that they're necessarily intelligent.
09:43I don't know that that's an appropriate word to apply to these particular kinds of systems.
09:49This is a kind of false anthropomorphism, thinking that machines are like human beings, that they have agency, that they
09:59have desires and goals.
10:01And that is simply not the case.
10:04These machines are tools and we should use them for what they're good for and we should not use them
10:10for what they're not good for.
10:11I don't think we need to worry about them getting more and more intelligent because it can't do anything we
10:17can do.
10:19And it never will be able to do anything we can do.
10:22But there are many tasks that we can do which it can do better.
10:26Now, is that something to worry about?
10:28Absolutely not.
10:29You know, a car can drive faster than I can run.
10:33That doesn't worry me.
10:35That said, there's a lot of work going on in artificial intelligence
10:39so that these systems can be incrementally trained and continue to learn, as people do in some sense, as they
10:46go.
10:46And that will be one of the great ways in which we'll see artificial intelligence improved in the future.
10:53Yes.
10:57One of the great unintentional marketing coups in history was by a guy named John McCarthy,
11:04who invented the word artificial intelligence in 1956 at conferences.
11:09Let's work on artificial intelligence.
11:12I don't think he understood how that was going to resonate among the general public,
11:18and in particular with the media, with the entertainment media.
11:22But had he simply called it what it was at the time, which is mostly symbolic processing and reasoning,
11:30logical reasoning by a machine, we wouldn't have had this same emotional reaction
11:35or the idea that, oh, my God, we're creating something new that's potentially going to rise up and take over
11:40or harm us in some way.
11:42Some people, for example, worry about a ridiculous concept called AGI, artificial general intelligence.
11:49And this is the idea that someday machines will be so intelligent, they'll outthink us and take over and we
11:56won't be necessary anymore.
11:59This is completely ridiculous.
12:02And it's not a meaningful statement.
12:05First of all, nobody can define what AGI really is and how we would know if we got there.
12:09If we did reach some definition of AGI, whatever you want to call it,
12:15and what will happen the next day, the machine's just going to sit there until you tell it to do
12:20something.
12:21So most of the fears and dangers that people are worried about with the new technology is not really a
12:27fear of the new technology.
12:30It's a fear of what people are going to use the new technology to do.
12:35And that's something that we're going to have to be very careful about.
13:00How do you think about the concept of singularity?
13:04The idea there is that when you build the machines to a certain point, they'll be able to improve themselves.
13:10And once they can improve themselves, that will accelerate off and they'll suddenly become so intelligent we can't even converse
13:18with them or something.
13:19The idea that that's somehow going to accelerate has never been the case in the history of technology.
13:25And I'm confident it's not going to happen here.
13:27These things will increase to some point.
13:30It will accelerate our ability to make increases, improvements, but then it will begin to tailor off.
13:37So the idea of the singularity that somehow we're going, it's really a pseudo-religious concept.
13:43It's the idea of the rapture, that a god, in this case the god coming from the machine, will come
13:50and say,
13:51this is a new time and all the believers are coming with me to heaven and everybody else is going
13:57to be dead.
13:59It's a religious concept.
14:00And there's these modern mystics, religious mystics I would call them, that are promoting this nonsense.
14:08It's mostly a modern religious technical movement.
14:11And it's no more true today than it has been in the past.
14:35He's overstating the case because he knows that it will be over-interpreted, misinterpreted in terms of this.
14:44He's welcome to have the opinion that LLMs are going to be obsolete in five years.
14:49But what I can tell you is this.
14:52The technology that we have in hand today with LLMs, we haven't even begun to mine the economic and social
15:01value that we can get out of those.
15:03So even if we stopped all AI research today, we'd have easily ten years of benefits yet to be realized
15:12out of the existing technology.
15:14Are we going to have a new wave of AI sometime in the future, which will supplant LLMs?
15:20It's quite possible that that's the case.
15:22The LLMs currently are trained based on primarily words, like words and pictures.
15:29But what Jan LeCun is really saying is that you need to have some kind of common sense physics, knowledge
15:37of how the real world works,
15:39that when you pick up a glass with water in it and you tip it over, the water is going
15:44to come out.
15:45And why is that?
15:46Now, every child knows not to spill their water.
15:51If they turn it, they're going to spill their water.
15:53That kind of knowledge doesn't come naturally to today's LLMs.
15:58They have to learn that from words, what people have said about holding a glass of water and tipping a
16:06glass of water.
16:06So I do agree that an area of great potential for artificial intelligence is a new wave of work that
16:18creates what he's calling a world model.
16:20That's a major area of focus for future research.
16:23And it's a little bit of an overstatement to say, well, we're not going to have LLMs.
16:28We'll have something new.
16:37Well, I think we're in the middle of a big AI bubble.
16:42But this is an economic question, which is separate from the question of is it a good technology and will
16:47it be useful?
16:49Every time we've had a major new technology where it became obvious to the financial community that this was going
16:56to be very valuable,
16:57a great deal of investment came in and usually a lot of infrastructure got built out.
17:02And a lot of that, let's just say it didn't pay back.
17:06When you look at the amount of money that's being promised for the build out of AI data centers today,
17:12the amount of money that has to be generated or made to pay that back is so large that it's
17:19very, very hard to believe that that could ever occur
17:22in a timeframe that will matter relative to the current investment.
17:26And I say that because if you build an AI data center today, it's obsolete in three to five years.
17:34And nobody wants three year old computers.
17:36It's just as simple as that.
17:38So to me, the economics just don't add up.
17:42Already, just in the past few weeks, as the time of this filming,
17:47OpenAI went from saying, we're going to build a trillion and a half dollars worth of AI data centers to
17:54drive our business.
17:55And I think it's collapsed down to, I forget where the exact number, seven or eight hundred million.
18:01A billion.
18:02OK, well, it's already coming down quickly.
18:04That's the bubble deflating.
18:06The money was never actually there.
18:08And luckily, the sooner that people become realistic about what's needed and how much investment should get made,
18:16the better off we're all going to be so that we're not crushed by another collapse,
18:20as we were in the Internet bubble when it burst around the year 2000.
18:29Well, the sooner it bursts, the better.
18:32The real question is, is it a bubble that bursts, as happened with the Internet?
18:37Or is it a balloon that we're going to let the air out of?
18:40If it happens slowly, things will be fine.
18:44A bunch of investors are going to lose their money.
18:48And that's OK.
18:49But it's getting pretty bad.
18:54And my advice would be, don't invest in artificial intelligence.
18:58As with the automobile companies, as with the early computer companies, we don't need 10 or 20 of these.
19:05We only need two or three basic systems.
19:09These are so-called foundation models.
19:12And when you've got 20 companies building them and each of them spending tens of billions of dollars
19:17and claiming that they need all of this computing and all of these NVIDIA chips, that's what looks, it's a
19:24mirage.
19:25And that will collapse down to a few companies and the competition alone will drive down the price at which
19:33these people can sell their products.
19:34Now, I was surprised to read just recently that most of the AI companies that are providing the service to
19:42you and charging for it
19:44are losing money on the actual service that they're providing.
19:48That is to say, it's costing more to answer your questions on Google or with Gemini or with Anthropics Cloud
19:58or ChatGPT
20:00than the average person is paying for that service.
20:03That's a very dangerous situation.
20:06Does every Korean invest in AI companies?
20:10Let me say for the record, I have four children.
20:15Two of them work full time at NVIDIA.
20:18And they're doing great.
20:20But I will tell you this.
20:22My advice to them, when your stock options vest, sell it right away.
20:28Really?
20:29And they do.
20:30Every quarter.
20:31Did you sell your stock?
20:33No, dad.
20:34I don't know.
20:35I think it's going to go up.
20:37Thank you for your advice.
20:45Well, I think having some capability to develop fundamental foundation models is valuable, perhaps for Korea.
20:56But it isn't necessarily important, particularly because let the silly and dumb people in the United States and in China
21:05beat themselves up to try to beat each other at this game.
21:10And what will result is that the technology will be available to Korea at very reasonable prices without Korea having
21:17to have put that enormous investment into those platforms.
21:21However, Korea can benefit greatly from them by adding layers of value on top of them and building that technology
21:30into products in unique and interesting and different ways.
21:34That's what Korea has historically been very good at everywhere from shipbuilding all the way through memory manufacturing.
21:43And there's no reason to think that Korea has to change its business model for artificial intelligence.
21:50That's what the Korean culture has been very good at.
21:53What are the applications?
21:54How is this going to be used?
21:56Korea has a terrific history of doing that very, very well.
22:00And I look forward to buying very many Korean products using artificial intelligence over the next several decades.
22:15AI is developing its own style.
22:19I think it does feel like a new species.
22:27AI, it's not a good storyteller, but I think storytelling is a very human impulse.
22:38And that culture, there's something very resilient about culture.
22:42you're the ones we or the others at diaspora.
22:43Yeah.
22:45I think politics are really familiar with.
22:47We'll have to look around in.
22:47Here we see in Пер istойд Greta-Pbudget .
댓글