Panos MADAMOPOULUS-MORARIS, Founding Managing Director of Industry Programs & Partnerships, Human-Centered AI (HAI) institute, Stanford University, Anant MAHESHWARI, President and CEO, Global High Growth Regions, Honeywell Moderator: Clay CHANDLER, FORTUNE
Category
🤖
TechTranscript
00:00So, there has been tremendous rhetoric about the advent of AI and what it means to firms,
00:06what it means to the global economy.
00:08And we have heard a huge range of, you know, from techno-optimism to techno-doomism about
00:16AI's impact.
00:17You know, people comparing it to the Industrial Revolution, electricity, I've heard people
00:25compare it to the discovery of fire, which feels a little over the top.
00:29And we've seen a number of really fascinating studies come out.
00:33I mean, the, you know, there was one that MIT's economist Darren Asamoglu came out
00:41that was quite pessimistic.
00:43McKinsey's had one that was a little more in the middle.
00:45Goldman Sachs has had one that has estimated that it could add as much as 15 percent to
00:51the global economy over the next 10 years, which seems extraordinary.
00:56So I'd love to, you know, think about those a little bit with you.
01:00And I'm just reminded of the 70s and the 80s when we had all these massive investments
01:05in computer technologies and robotics.
01:09And there was this great conundrum among economists because as, you know, Nobel Prize winner Robert
01:16Solow observed in 1987 that you could see evidence of the computer age everywhere except
01:22in the productivity statistics.
01:23It wasn't really showing up.
01:25Well, eventually it did show up.
01:27But I wonder whether, you know, they called that moment the productivity paradox, whether
01:31we're seeing a similar productivity paradox with AI.
01:35And both of you are kind of perfect people to address this topic.
01:39You know, Anant, you're right on the front lines of this at Honeywell.
01:43Tell us a little bit about what you think about these kind of wildly disparate macro
01:48productivity estimates.
01:50Yeah, and maybe just to start, optimism and pessimism can be seen both ways depending
01:56on which side you're looking at it from.
01:58But just a quick context of where I come from, we operate in three megatrends on the planet.
02:04That is future of aviation, automation, and the energy transition.
02:08So my comments will be centered around that.
02:11But these are real world industries.
02:12These are really people who operate aircraft, people who are operating buildings and plants,
02:19manufacturing plants, and people who are in the energy business.
02:23Now, we recently did a study across 1,600 people, executives across different companies
02:31in 12 countries, with each of them having 1,000 people at least.
02:35And we realized that nearly 17% of companies have actually implemented full scale sort
02:42of projects, which are related to productivity primarily, because that's the number one play
02:49that people are seeing around productivity and efficiency.
02:51There is, of course, another play, which is around cybersecurity, where people are deploying
02:56a lot of AI.
02:57And productivity, I would say, is really around saving cost of energy, saving materials, and
03:05getting more with less.
03:07Parno, say a little bit about your role at the Stanford Center, and love to hear your
03:14take on this kind of paradox.
03:18Yeah.
03:19Thanks for having me, Clay.
03:20I'm excited to be here.
03:21Always excited to be in Southeast Asia.
03:23So yeah, at Stanford Institute of Human-Centered AI, we are sort of the umbrella organization
03:30for all things AI on campus, and we are very committed in advancing AI policy, research,
03:37education, and practice, underscoring practice to accelerate human condition.
03:43As part of this, we work with a number of Fortune 50 companies, help them accelerate
03:49the AI journey.
03:50And this means different things to different people.
03:51So there has been, indeed, a lot of skepticism around productivity or ROI.
03:58A number of reports came out, both from analysts in Wall Street, but also in our world in Silicon
04:05Valley.
04:06But also economists are following this very closely.
04:09You mentioned productivity paradox, Eric Brioso, my colleague at Stanford, has been
04:15doing work on this topic for the last 30 years or so.
04:19Let me unpack.
04:20There was a couple of things here we should be unpacking.
04:23One in terms of the concerns.
04:24Yes, there are some really credible concerns.
04:28People are looking at the CapEx, this Goldman report mentioned a trillion dollars in CapEx
04:34built out infra.
04:35The hyperscalers alone, they are committing $60 to $80 billion this year.
04:40In excess of the already committed CapEx for research.
04:45And then in startup land, you have unlimited amount of capital getting into AI startups.
04:53And we don't have, at the same time, that many startups in the 100 million ARR territory.
04:59So people are skeptical.
05:00To what extent do those startups can live to the valuations they raise money?
05:06And then for capital allocators, there's a lot of pressure because companies now are
05:11staying private for longer.
05:14And with scaling lows advancing the hyperscalers and the magnificence of M&A, M&A will become
05:22harder and then it will be more challenging for them to build sustainable modes.
05:28So two things on our end, having worked with a number of those Fortune 50 companies over
05:32the course of the last three years.
05:34One, I think people are more of the count that we are underestimating the long-term
05:41impact of AI at scale and overestimating the short term.
05:46So that has created unrealistic expectations when it comes to ROI, both in terms of revenue
05:52or productivity or whatever way you want to measure it.
05:56It's one thing to see public interest spiking and yes, open AI went to 100 million users
06:01in two months.
06:03I was reading the other day, 20% of US retailers, it took them 20 years to get online.
06:11And there are multiple use cases, Apple revenue curve, Salesforce revenue curve, many large
06:18organizations still, they are not in cloud properly.
06:22So expectations versus reality is one, those things take time.
06:26And second is what lies beyond tech.
06:28So most of the work we do with companies, it's not really on, let's buy this tech or
06:33let's build this model or let's dump some AI to just replace the existing workflow.
06:39We feel that most of the value will come by thinking AI native, which means how you really
06:45reinvent your business model, how you're revisiting your processes, how you upscale and reskill
06:51your people.
06:52This is where most of the value will come from.
06:55That will take more time.
06:56And I think even Goldman in this report recognized the railroad analogy, the first wave and the
07:01second wave.
07:02So we are more of this kind.
07:03Well, you said something this morning that I thought was quite insightful and a couple
07:07of other panelists on the main stage this morning have mentioned it, which is the difference
07:12between using AI to achieve incremental savings on the production process that you've always
07:22used versus saying, okay, no, we have to completely reinvent the production process
07:28around the new technology.
07:30And we definitely saw that in the computer investment waves, where originally what everyone
07:35was doing was fiddling at the margins, but not rethinking how they do everything.
07:40And that this long-term curve that you're talking about, Panos, where the real gains
07:46kicked in was when people finally said, oh, you know what, we got to rip up the whole
07:49legacy system and rethink it from the beginning.
07:53Are you seeing that with sort of the firms that you work with, and if so, what does that
07:59imply about how long the kind of payoff horizon might be for this technology?
08:04So maybe, again, putting in context of very three different kinds of industries and use
08:10cases, as you say, I'd really like to do a poll in this room if you would fly an aircraft
08:16if I said I'm not going to have the pilot in the aircraft and that's the aircraft that
08:22you're on for 16 hours or eight hours, that'll take you to your destination.
08:27There would be a few adventure seekers here, but I would imagine a lot would say, you know,
08:33I'll wait on it, let a few more people do it, and let the regulators approve it before
08:38I get to do this.
08:39Now, this is an industry that is short of aircrafts and short of pilots in the air right
08:45now.
08:46Today, it's regulation, two pilots in the aircraft to fly you.
08:50Will it get down to zero?
08:52Maybe not, but can it get down to one with the power of AI and therefore have a lot more
08:56aviation going?
08:58That's possible.
09:00And that's where I think productivity and the workforce really comes in because you
09:04can get a lot of input and truly AI becoming a co-pilot to that pilot.
09:11Now, the second industry, if I go, and this is again a thing that when people are not
09:17talking about AI, they're talking about sustainability and the energy transition.
09:22Now, if you look pre-pandemic to today, actually the number of jobs have gone up in the energy
09:28industry because AI and the need for energy is accelerating the energy transition.
09:34There's a lot more renewables, there's a lot more different kind of green electron coming
09:39along, and it's short of people.
09:42So when you're short of people, whether you're in the United States or Germany or Japan,
09:47then how do you augment that workforce?
09:50And you again bring AI to help you augment that workforce.
09:55And if you go down to the developing part of the world, they need to train up a lot
09:59of people for new industry and automation therefore is coming in.
10:03And that's again where you need AI to help that labor come up the learning curve very
10:08quickly.
10:09So a lot of training, a lot of simulation, a lot of testing where the system might fail
10:14and therefore training somebody to do that.
10:17So overall, I would say there is definite use cases of productivity, of workforce augmentation,
10:23of training and enhancement, of savings in energy that are satisfying the CFOs in every
10:31company to say, you know, there is something in it.
10:35You know, part of the question about the productivity gains is in a way two questions.
10:41I mean, it's a question about how much do we think AI is going to be able to reduce
10:47labor costs or the cost of total production?
10:51But then I guess the other part of the question is what percentage of tasks or jobs do we
10:57think that AI can really be able to automate?
11:00So if you think about your airplane example, I'm not sure I would feel comfortable getting
11:04into, I would get into a robo-taxi, but would I get into an airplane that didn't have two
11:09pilots?
11:10I might not, right?
11:11Just because the stakes are so much higher.
11:13But if you think about what already happens in the cockpit at most airplanes, already
11:18maybe 90%, I think people don't realize that when you get on a plane, pilots aren't touching
11:22very much until the actual, you know, except the takeoff and the landing, right?
11:25They're focusing on those really critical tasks that still haven't been automated.
11:29So I think there's probably a lot of equivalent to that in the production processes of different
11:34companies.
11:35You know, Darren argues at MIT that, in fact, the number of tasks that can be automated
11:41is probably not as great as people think, and so that's where he gets, you know, he
11:46says probably if we get a productivity gain, it will only be 1% at one percentage point
11:51a year rather than these crazy, you know, 2% and 3% kind of estimates.
11:58Can we talk a little bit about the job impact of this transformation?
12:06You know, the famous William Gibson quote is that the future is already here, it's just
12:14unevenly distributed.
12:15And I think that there is a little bit of that associated with AI, is that the fruits,
12:22the benefits, the productivity gains of this are going to be different in different parts
12:26of the economy and for different types of jobs.
12:30How do you see that, Panos?
12:31I know you're thinking about this with your industry, you know, contacts.
12:36We are thinking about it a lot, indeed, and in multiple dimensions.
12:42You know, there are a lot of parallels in AI, and so organizations are trying to figure
12:46out how do they manage in a world where alpha is so asymmetrically distributed.
12:56So there's a big delta between how quickly research is moving and then the time that
13:03organizations have to effectively respond and how they respond to this, in our view,
13:08is going to define pretty much winners and losers for the next cycle.
13:14So there are a few things.
13:16I think this year we've seen a lot of intentionality from executive suite and boards.
13:23Most of the transformation journeys are led by either the CEO or the CIO.
13:30Boards are very, very actively involved.
13:31As a matter of fact, I was in New York before this, and I ran a poll in a room like this
13:35with CIOs, and I asked them, how many of you are you expecting to present in every board?
13:41And the vast majority of them told me every board we expect to go in and report progress
13:48on our journey on the AI side.
13:51Second is, I think alliances and ecosystems as a mode, because things are moving so quickly,
13:59there's so much internal and external pressure to executive leaders, costs are still very,
14:06very high, and human capital is secure.
14:10I think that has pushed the envelope a little bit and expanding the surface area of creativity
14:17in terms of business models.
14:18So even organizations that were traditionally very conservative, they wanted to do everything
14:23themselves, now you see those global alliances, ecosystems, and both of us spent a lot of
14:29time around the world actually managing those ecosystems, which I think, bottom line, is
14:34great for the organizations, it's great for the discipline, like AI discipline as a whole.
14:39You see the rise of this co-opitation.
14:43Because of this, many companies at the same time, they might be competing, but they might
14:47be partnering.
14:48And one last thing I would say is that, yeah, alpha will be asymmetrically distributed,
14:53at least for the foreseeable future, but leaders in their own industries can identify and curate
15:00their own pillars of alpha, because alpha has many different colors.
15:05And something that we've seen again and again, particularly the last six months, is the human
15:11centricity of AI, like how increasing awareness from executive leadership in different organizations
15:18on how they build, design, deploy at scale in a human centric way can deliver alpha both
15:25internally and externally on the safety side, on the design side, on the implementation
15:31side.
15:33And I think, as a matter of fact, we're going to find ourselves in a few months that those
15:39transformation journeys, they will be measured.
15:42Their success will be measured by how human centric they are.
15:45And that connects to your argument about jobs and displacement.
15:49We've got a few minutes.
15:51We'd love to take questions from the floor, if anyone has them.
15:56Yeah, here?
15:58Do we have floor mics?
16:02Where are our floor?
16:06There he comes.
16:10Thanks.
16:13Thanks.
16:15Seth Dobrin.
16:17You know, I think there's three things I really didn't hear addressed that are important to
16:20this conversation.
16:22One is a lack of diversity.
16:24And so much so, you know, for instance, Panos, like, you know, Fei-Fei, who founded, you
16:28know, the Human Centered AI is really the grandmother of AI, and no one ever talks about
16:34that.
16:36They just talk about the fathers of AI.
16:38You know, lack of global representation, which is really important, because people
16:42talk about, you know, fine tuning models, but fine tuning actually breaks the model,
16:46so you can't do that when you look at other cultures, which can cause actual type of colonialism,
16:52a new type of colonialism, because you're inflicting Western and Chinese cultures on
16:56the rest of the world.
16:58And then the short term impacts on jobs.
17:00Yes, we're going to create new jobs, but you're going to have a transition period where you're
17:04going to have a lack of job creation.
17:06You're going to have people leaving the workforce.
17:08What are things that can be done?
17:10And what is, you know, high found in the near term, you know, what's going on through the
17:14research that you guys have done?
17:18Is the question addressed to me exclusively or to the panel?
17:20No, no, to the whole panel, because there's some that's the whole panel.
17:22You want to start?
17:24Three topics.
17:26One and a half minutes.
17:28So let me touch quickly on two of them, and I'm happy to talk offline about everything
17:32else.
17:34Yeah, indeed, FEFE is extremely committed in advancing diversity.
17:38As a matter of fact, we had our five years anniversary a few weeks ago, and in our dinner
17:44session, we had the godmothers of AI, and I encourage everyone to go on YouTube and watch
17:50the conversation.
17:52It was illuminating.
17:54We are very actively on the research side trying to empower and enable this on campus
18:00for our grant programs.
18:02But then there are bigger issues that are affecting diversity in AI, which is the availability
18:08of resources overall.
18:10It has become extremely expensive to run major AI experiments on campus.
18:14So, you know, academia is losing talent to the industry, and we are committed both in
18:20terms of our public policy efforts, but also our work with industry to reverse that as
18:26much as possible.
18:28Really quick on the impact of jobs.
18:32Short-term versus long-term.
18:34Yes.
18:36Right now, we see certain use cases that they prove the thesis that the human and the machine
18:44are better.
18:46Human and machines collaborate and are better for organizations and better for society.
18:52We see that in customer support.
18:54We see that in diagnosis.
18:56We see that in a number of use cases.
18:58A few papers coming out of Stanford, MIT, other places that they prove that thesis.
19:04There's this economic concept of complementarities where there are certain tasks that humans
19:10are better.
19:12Emotional intelligence, creative problems, contextual understanding.
19:16And there are other tasks, like the ones that are predominantly repetitive in the nature
19:22of computing data that ai's are more effective.
19:26So bringing those two together, we believe will amplify human creativity and augment
19:34like we're very committed in augmentation versus replacement.
19:38Second, talking about pilots is another economic principle.
19:44So when the jet engines were invented and technology dropped, increased productivity,
19:52dropped the costs, the demand for pilots skyrocketed.
19:56It didn't went down.
19:58There was a lot of talk about, you know, in the health care
20:02Community about radiologists and how radiologists are going to be in space.
20:06Well, guess what?
20:08U.S. Cannot have enough.
20:10There are 2x number of postings of the people who are actually available to work as radiologists.
20:16And that boils down to this thesis where for some tasks, we're going to still need humans in the
20:22Loop, and we have agency, business leaders and policy makers that have agency in making
20:28Those decisions.
20:30Those are not decisions that will come embedded with products, but humans have agency in
20:34Directing the strategic direction of the organization.
20:37We are unfortunately out of time, but a lot of challenges, a lot to think about.
20:43I would encourage everyone to find both anant and panos, you know, in the hallway, in the
20:49Lobby to follow up and explore more about these topics.
20:53Gentlemen, thank you very much.
20:55Thank you so much.