- il y a 10 heures
Generative Ai The Next Productivity Frontier
Catégorie
🤖
TechnologieTranscription
00:02Sous-titrage Société Radio-Canada
00:35I must say that the expansion and acceleration of the usage of foundation models and neural networks have exceeded our
00:45estimations.
00:47We are now back to give you an accelerated estimate of AI impact due to generative AI.
00:54The ability of a small number of big techs to commit significant amount of money and researchers on the specific
01:04topic of foundation models
01:06has eventually materially changed the history of technology.
01:13So I'm excited to be with my colleague Larina to discuss this potential change to the world and create value
01:21for companies.
01:23So, if it works.
01:29Generative AI has exploded as a topic for discussion with the release of ChatGPT, as you know.
01:35One million users in five days.
01:37It has put AI back and in all of our headlines.
01:42But you can see two camps.
01:43Those who believe that AI could have a major positive impact on society.
01:50And those who believe that the massive power of AI may lead to potential dangers for our society.
01:59Actually, both groups know very well what they're talking about.
02:03And I believe we can collectively have a role to play in making the best of this technology.
02:10What is for discussion in the following?
02:13Is it finally the opportunity to end what some economists call the secular stagnation?
02:20Let's see.
02:21But what is for sure is that with great power comes great responsibility.
02:28Larina.
02:28And my name is Larina Yee.
02:30I'm a senior partner in our San Francisco office.
02:33I spend all of our time thinking about the disruption of technology and how it relates to our society and
02:38economy.
02:39And absolutely, as Eric said, this is a moment of great leadership and responsibility of all of us.
02:46And so, what we're going to talk about today is what is generative AI?
02:50Just a quick primer in case you're not 100% sure.
02:53But more importantly, how does it give us productive powers?
02:58How does it give us superpowers?
03:00And then we're going to talk about the implication for society and business.
03:04Let's get started.
03:08First, we need to understand what we can do with generative AI.
03:12Generative AI, as you understood, is part of AI.
03:15But we can do three things.
03:16We can generate new content, text, image, code.
03:19We can process and analyze unstructured data.
03:22And we can enhance human-machine interactions.
03:27So, we are now playing with even more chatbot, virtual expertise, live translation, which is unprecedented in terms of accuracy
03:37compared to the former AI models.
03:39But concretely, how does it work?
03:44We need to understand the new techniques that have made tools like ChatGPT possible.
03:49We call these techniques foundation models.
03:51They are a subset of deep learning, machine learning, and artificial intelligence.
03:55And are really the great novelty that have made tools like ChatGPT possible.
04:01These things that make foundation models differ from previous deep learning models.
04:05They are trained on massive unstructured data sets.
04:09For example, one petabyte for GPT-4.
04:13It's equivalent to 11,000 high-definition movies.
04:17It would take you around two years of non-stop binge-watching.
04:22They are way more powerful than previous deep learning generations.
04:271,000 more parameters for GPT-3 versus former deep learning techniques.
04:33And they rely on transformer architecture, which is, by the way, has been released in 2017 by Google with self
04:43-attention mechanism that can understand interdependencies in the input data sets.
04:49So, a few examples of all of this, as you know, is GPT-4, Meta, Lamar, and Tropic Clouds.
04:59In fact, speed of innovation has been much faster than we expected.
05:07And there was a clear acceleration of increase of capabilities in the past six months.
05:16The size of databases and complexity of the model led us to this tipping point.
05:21But the release of CHAT-GPT in November 2022 opened a kind of Pandora box.
05:28But just for you to know, GPT-3.5 could not pass U.S. exams in November 2022.
05:38GPT-4 is now succeeding in most of the U.S. exams to get into university.
05:44It has moved from bottom quartile to top quartile of GRE.
05:53Let's talk about money.
05:55On the left, you see the incredible acceleration of investment.
05:59And the question that comes is, who is investing and where?
06:03In fact, as you can see, private external investment have been multiplied by 40 since 2017 to reach more than
06:11$12 billion in 2023.
06:14As a comparison, AI overall only increased by four between 2017 and 2022.
06:24If you try to understand where these tech giants have invested, along with VCs, they've invested really into internal R
06:34&D to focus on this foundation model that we speak.
06:41If you look at the split between different regional geographies, it is two-thirds in the U.S., one-third
06:49in China, Europe being really minor into the amount of investment on the topic for now.
06:57So, let's try to understand what is the potential of all these two augment activities.
07:04I mean, given what Eric said, there's no doubt that this is moving fast in terms of the technology innovation.
07:10There's no doubt with a 40% increase in investment levels that there's a lot of money going into it.
07:16But, fundamentally, how does it affect how we work?
07:22How does it give us superpowers?
07:24How does it augment what we do?
07:27And so, we wanted to dig into this.
07:29So, just bear with me for one second.
07:31The way that we looked at this is we wanted to truly understand how your everyday job changes.
07:38We looked at 850 occupations globally and broke that down into 2,100 activities, exactly those.
07:46And we looked at it over 47 countries.
07:49And looking at the anatomy of work, we found that there are significant ways that generative AI specifically can affect
07:58our day-to-day.
07:59So, take, for example, being a software engineer or a software developer.
08:05There are a lot of things that you do each day.
08:08And some of the things that generative AI is incredibly good at doing is thinking about code development.
08:14If you want to actually address your backlog.
08:18If you would like to look at anomalies in data.
08:22If you want to modernize your old cobalt code and make it into Python.
08:29Generative AI does this extremely well.
08:32Now, don't worry.
08:33We still need software developers.
08:35Because there are things that the technology doesn't do well.
08:38So, it's not very good at making trade-offs.
08:41So, for example, we still need the software engineer to look at what generative AI does and make choices.
08:47And we still need our software engineers to push the boundaries of innovation.
08:52So, with each job, what we find is that up to 50% of the day-to-day could be
08:59made more efficient, could be made faster, could be a super assistant to you.
09:04And then there's a question of how does that add up.
09:07So, what it adds up to is 70% of the global workforce.
09:1270% of the global workforce could see up to 50% of their activities automated.
09:18That is a stunning number.
09:20And we've been looking at the automation of technology for many, many years.
09:25As Eric said, we looked at this in 2017, we looked at this before.
09:28And never have we seen so much of what we do being able to be done together with artificial intelligence.
09:38And so, what this graph is showing you is that when we look at all automation technologies and we compare
09:44that to what happens when we add generative AI, it makes a big difference.
09:49That idea of 50% of activities automated, that was much less when we just look at automation technologies like
09:56robotics, for example.
09:58And so, two things that you should take away is that generative AI makes a big difference in this potential
10:05and it affects a different class of work.
10:09Previously, we had things that automated physical labor.
10:12We had things that made transactions go better.
10:15And one of the things that generative AI does is it uniquely touches knowledge work, customer service agents, attorneys, financial
10:22managers, the work that Eric and I do.
10:25That's the potential.
10:26If we look ahead, then you say, gosh, the technology is moving really quickly.
10:32And on the next slide, we also take a look at how fast all of this will happen.
10:37And so, not only do we want to understand how many jobs will be affected, we want to understand under
10:42what timeline.
10:43It's certainly not happening tomorrow.
10:45And one of the big impacts that we see is that when we add generative AI to our analysis, it
10:51basically bumps up our timeline by a decade.
10:54Like, how cool is that?
10:56So, what this is saying is that if we were to think, and this is kind of how economists think
11:01about it, if we were to think about how long does it take to get 50% of the automation
11:06potential,
11:07basically the number you should remember is it's going to come faster, and that is in the tune of a
11:12decade.
11:13One more thing to share.
11:14We go to the next slide.
11:17One of the things that we talk about is that this really pushes the boundaries of the productivity frontier.
11:24You know, I was born in the 70s.
11:26It was a golden era of productivity.
11:29Of course, I didn't know that, but it was a magical time where the global economy was very productive.
11:34Eric and I have talked a lot about this.
11:36And over the years, and maybe just looking at the last 20 years when all of us have been working,
11:42we have been productive as a global economy, but not nearly as much as we would like.
11:47And so, I do think there's been a search for what will drive the next frontier.
11:52What changes the productivity curve?
11:55And we know that technology is one of those things.
11:57What's remarkable is the impact that generative AI can have to that productivity frontier question.
12:03At its maximum, it can add 3.3% productivity growth to the global economy.
12:11That may not sound like a lot, but for anyone who thinks about productivity growth, that is a huge number.
12:17And even if we're a little more modest, it's maybe 1.6%.
12:22I would take that productivity growth every day of the week and twice on Sunday if we could get it.
12:27But ultimately, what will happen is this is what the technology will do.
12:31It will be our choice as business leaders how fast we go, how quickly we implement it, how much we
12:39embrace the capabilities of the technology.
12:42As Larina said, just one point of additional productivity puts us back to what we experienced in terms of productivity
12:50growth in the 70s.
12:52So, this is the global value we assessed on industry.
12:58We studied the impact of generative AI across 16 business functions such as marketing or finance.
13:05And we found out use cases everywhere.
13:09Generative AI, as you can see, has the potential to bring between 2.6 and 4.4 trillion annual value
13:15in productivity from net new use cases.
13:19To put that into perspective, this is pretty much over bigger than the size of the United Kingdom, the UK
13:26GDP in 2021.
13:28This would increase the impact of existing, previously existing artificial intelligence and analytics by 15 to 40%.
13:38But what would drive this massive impact?
13:42It's essentially four business functions, four areas that will basically create 75% of the value.
13:54This is customer operation, marketing and sales, software engineering and R&D.
14:11So, what does it mean on the way we work, you know?
14:23Technically, people can focus on higher value activities.
14:28There's a lot of time consuming activities that are low value added.
14:32Thanks to generative AI, we can reduce time spent on this specific activity, spend more time on higher value added
14:40ones and be better at our jobs.
14:42To illustrate, as you can see here, this is an example of a marketing manager.
14:46They can use generative AI as a co-pilot in their day-to-day activities.
14:52Instead of spending time on customer data collection, cleaning, synthesis, they can leverage Gen AI for this and focus on
14:59reviewing and providing better insight, better manage content.
15:04Also, maybe automate their relationship on content management with their advertising agencies.
15:10However, this requires knowledge workers in general to master the technology and develop an expertise in the area rather than
15:21spending time on easily transferable tasks between functions and industries.
15:27So, probably pushing us all on higher added value tasks.
15:33Of course, there are risks.
15:38Many risks.
15:40Huge potential and several risks that need to be mitigated.
15:44That's where comes our collective responsibility.
15:48Opacity on foundation model is one.
15:51It can lead to non-desired biases.
15:55Model might provide also inaccurate information and, as we say, hallucinate.
16:01So, to capture solely the positive impact, companies must ensure that they understand the threats and how to mitigate them.
16:10This is what we call technology social responsibility.
16:13They should be careful of integrating generative AI without human oversight in applications where errors can surely cause harm.
16:27So, with the risks and what Eric said at the beginning is comes responsibility and opportunity.
16:35I am an optimist.
16:36I am an optimist about technology.
16:38I am an optimist about humanity.
16:41I suspect most of you in the audience are similar.
16:44Let me describe some of the potential.
16:47Generative AI can be a huge differentiator in our society.
16:51You know, we talk a lot about jobs and work.
16:54Let's take the example of a farmer.
16:56A farmer in India in a small village who doesn't speak the main language and certainly doesn't speak English.
17:05And what Microsoft did was they deployed a Gen AI assistant in this village.
17:09And why did they do that?
17:11How were they helping the farmer?
17:13One of the things that generative AI does is it's incredibly good with languages.
17:17And it basically helped give this farmer a way to communicate with the outside world and way beyond his village.
17:26That is so great.
17:28That's just an example of a societal benefit.
17:31Of being able to reach people who, you know, are hardly touched by technology in a way that connects them.
17:38So, from a society perspective, we're going to see benefits.
17:41And from a business perspective, as you can imagine, the numbers are pretty big.
17:46When we look at the industries affected, all industries will see a productivity boost.
17:52Let's just take three as examples.
17:54We look at retail.
17:56We look at life sciences.
17:58And we look at banking.
17:59And I pick these because the industries will be affected in different ways.
18:03But the bottom line is, no matter what, the numbers are big.
18:07Pick your billions.
18:08Between 400 and 600 billion of productivity growth in retail.
18:12200 to 340 billion in banking.
18:15What's happening in there?
18:17In retail, you're going to be able to see a different amount of support for how we relate to customers.
18:23And how information is put together.
18:25And how you get something that we've always dreamed about.
18:28Hyper-personalization.
18:29So that I can know you specifically, not your segment.
18:34Those are the types of things.
18:35Whereas in banking, some of the benefits are going to be in our ability to manage risks.
18:41Our ability to actually modernize legacy systems.
18:45Our ability to provide wealth management at an incredibly personalized level.
18:50So the potential across industries is unbelievable.
18:54And in life sciences, I mean certainly something close to home for everybody.
18:58Is being able to accelerate the pace of drug discovery.
19:02To solve some of our greatest challenges that we have.
19:05So from a potential perspective, we're pretty excited.
19:08Let's keep going.
19:10So companies.
19:11What are businesses, all of you doing today?
19:14Leaders are leaning forward.
19:16They're stepping up.
19:17And they're moving quickly.
19:18It's not just the investment.
19:19It's about actually moving to execution.
19:23And what's remarkable is how fast businesses have moved.
19:26These are just six examples.
19:28I could probably have 600 examples.
19:31But some of the cool things that we see are, for example, Morgan Stanley.
19:35They've armed 300 of their wealth managers with an initial pilot.
19:39Where actually generative AI helps with the drafting and curating of information.
19:43Giving more time for those wealth managers to spend time with each of us.
19:47To do the things that humans uniquely do.
19:50If you look at Salesforce.
19:52They had their big AI week.
19:53They've announced all kinds of capabilities.
19:56On the workflow software that you're already using day to day.
20:00You know, imagine.
20:01Maybe you're on Slack.
20:02Or maybe you use some kind of video conference.
20:05Imagine you're at a meeting.
20:06Or you're in a conversation.
20:07And you can push a button.
20:09And ask generative AI to summarize the notes.
20:11I mean, and give you your two action items.
20:14That would be really helpful.
20:15So when we look at what companies are doing.
20:18Think of it this way.
20:19There's a combination of very practical productivity cases.
20:22There's also the potential for more innovation.
20:25The fact that I can text in my phone to Walmart.
20:28And it will, if I ask them a question.
20:30It will give me options.
20:31And all of a sudden it will appear at my doorstep.
20:34So these are the sets of things happening.
20:36This is just a sliver of what's exciting.
20:39If we think about how to wrap this up.
20:42I think there is a quote that really captures how we think about this.
20:46And what you might think is that I would talk about technology.
20:50But really, this is about humanity.
20:52This is about what we as leaders choose to do with this incredibly powerful technology.
20:58And really, we're at the center of it.
21:00So how we think about reshaping how we work and how we live.
21:04We have so many more options now.
21:08And so what I'd love to leave you with is just a couple things to remember.
21:11One, this can be a huge benefit for the productivity of our economy.
21:17Two, it can be a massive unlock for your business.
21:20Three, it can give individuals that you work with superpowers.
21:24And as you think about all of that.
21:26Think about how to do that in a way that makes the way in which we live better.
21:32Thank you so much for your time.
21:34Thank you. Use your superpowers.
21:36Thank you. Use your superpowers.
Commentaires