- il y a 2 jours
AI Beyond Scale: Building Competitive and Sustainable Ecosystems
Catégorie
🤖
TechnologieTranscription
00:00Hello, everyone. I'm Karen Cho from CNBC.
00:03It's so great to be back with you here at Viva Tech.
00:06Let me introduce you to our panellists today.
00:09We have Dawson Tong, Senior Executive Vice President
00:12and President of Cloud and Smart Industries Group at Tencent,
00:16and also Bruno Zovem, who is the Chief Technology
00:19and Innovation Officer at Orange.
00:22Thank you both so much for joining.
00:24I'm so excited to talk about where we're at in the industry
00:27because I think we sit at a crossroads
00:30covering technology for so many years.
00:33It feels like we're at the moment where all of your advantage
00:36that you've been built up could be simply eroded
00:39if you don't get your AI strategy right.
00:42Just talk to us about this pivotal moment that we sit at
00:46and why it's so important to move at pace
00:49and get the strategy right.
00:52Okay, I'll start.
00:54Thank you, Karen.
00:55And, I mean, when we look at AI, there's so many new technologies,
01:02so many approach people are taking.
01:07So I think it's open for a lot of different teams to explore.
01:13And at Tencent, and specifically Tencent Cloud,
01:17we are investing a lot at different layers,
01:20including infrastructure, how to use the infrastructure
01:25most cost-effectively, most efficiently.
01:28We also build tools to create platforms
01:32for building different types of models,
01:36inference, how to create the most efficient services
01:40for using models, and applications.
01:44Tencent has a lot of apps,
01:45a lot of digital ecosystems
01:48that we empower users from different industries
01:55using our technology to meet different needs.
01:58Bruno, Tencent's had a great decade.
02:01Orange, I know you've been looking at many revenue models
02:03over the years, so why is now so important?
02:06So the first thing I want to say is we operate in 26 countries.
02:10We have 300 million customers.
02:13And so it's a huge amount of data,
02:15a huge amount of information to deal with.
02:18And when you think about companies like ours,
02:22we are a data-centric company.
02:24We're data-defined companies.
02:25And I think from that perspective,
02:28AI represents an incredible opportunity
02:30to get closer to our customers,
02:32to better understand their needs,
02:33to be more proactive.
02:35The other part is we also manage a huge infrastructure
02:39across all of those countries.
02:41And we want to make sure
02:43that we deliver the best service all the time,
02:46that you don't have any network outages
02:47and all those things.
02:48And again, having an AI layer,
02:50having that intelligence layer
02:51that runs on top of our infrastructure
02:53that helps us better manage it,
02:55creates a tremendous amount of opportunities
02:57from an efficiency perspective
02:58and quality of service perspective.
03:00Let's talk more about the collaboration
03:01because I've covered a lot of technology companies
03:04over the years too.
03:05And the one message is that,
03:07look, we've been building all this infrastructure.
03:09We keep on investing in the next generation,
03:125G, it was 4G before.
03:14It's costing us a lot of money.
03:15And the technology companies
03:16are simply just piggybacking off what we're creating.
03:19They're not actually bringing much value to us.
03:22So it's interesting to see the two of you now
03:23working together about 18 months,
03:25two years into the partnership.
03:27What does the collaboration look like?
03:29Well, first of all,
03:30all of our services depend on high-quality network,
03:34which what Orange can offer.
03:37We collaborate in many different fronts.
03:40We also go to market together
03:43to create new user experience,
03:46some of which we see new ways
03:51of building a digital ecosystem in China,
03:54like WeChat,
03:55and we see that a model can be applicable
03:59in other markets as well.
04:01And with the partnership with Orange,
04:03I think we're exploring some of the exciting things
04:05in our joint markets.
04:09So from an Orange perspective,
04:11we have built a super app
04:15in Africa and the Middle East called Maxit.
04:18And when we initiated that strategy,
04:21we had very lofty goals,
04:24high ambition to create,
04:25to deliver the best possible experience at scale
04:29for hundreds of millions of users.
04:31and we wanted to have things
04:33like digital wallet, e-commerce,
04:36also creating a fantastic experience
04:38so that our customers can interact with us,
04:40have access to content, mini-apps, chat.
04:43And so it became very natural for us
04:45to figure out how do we do that?
04:47How do we work?
04:48How do we do those things?
04:49And I believe we believe in open innovation,
04:52in open collaboration,
04:53and picking the best partnership in the world.
04:56And for us, it was very obvious
04:59that to have this relationship with Tencent
05:02to combine the incredible success,
05:05their incredible expertise at scale
05:07with this ambition that we have.
05:09And that's the way we've built this Maxit solution
05:11that today is based on a partnership
05:13that we announced over a year ago.
05:16And today, it's already in full production.
05:18So we're working very well
05:19and we're working very fast with our partners.
05:22I don't get why Tencent couldn't do this by itself.
05:25A billion users back home,
05:27why couldn't you simply just build out
05:28the exposure and the super-app yourself?
05:31Why did you need Orange?
05:33Well, first of all,
05:34we have many years of experience
05:37managing our WeChat pay
05:40and mini-program ecosystem in China.
05:43So a lot of those experience
05:45and understanding of the stakeholders
05:51in the ecosystem that we serve
05:53would be very applicable in other markets as well.
05:58While not only we provide technology framework
06:02to many of our partners like Orange in this case,
06:06we also work with some of the partners,
06:09content partners, service partners,
06:11who are already serving consumers in China
06:14and make them available
06:16as the early adopters of that ecosystem.
06:20and from that, we can boost user engagement.
06:24It's creating scenarios where,
06:28in the case of Maxit,
06:30more payment transactions that can be done within the app.
06:34So over time, you can grow that ecosystem
06:36by introducing more local third-party players as well,
06:41retailers as well.
06:42Do you want to comment on the super-app?
06:44Because super-apps are big in China,
06:46but they've been slow to come to Western markets.
06:49What are you doing with the super-app,
06:51and is it also something that we will see in Europe?
06:54So I think if we're extremely honest and transparent,
06:58you need to have a killer application
07:01to justify the super-app.
07:02And for us, that was what we call Orange Money,
07:05that was helping millions of people
07:09in Africa and the Middle East
07:10to have access to micropayments.
07:13Once you build this kind of relationship,
07:15you essentially help them having access
07:16to digital goods, digital needs.
07:19Then it changes the way they think about you.
07:21They see you as an enabler
07:22for having access to all of those digital services,
07:25and that's kind of the gateway
07:26for this broader offering.
07:28And we have that in MEA,
07:31and that's the reason why we managed
07:32to expand on the Orange Money to build the Maxis.
07:35The other thing that's fantastic with Tencent
07:37is not only they gave us access to great technology,
07:40but they also gave us a model of collaboration
07:43where I was talking about all those countries
07:46we have in MEA,
07:47they also have specific needs.
07:49And it gives them the autonomy
07:51to develop additional features
07:53that are custom-unique to their countries
07:56while leveraging the entire platform.
07:58Can you bring the ZOOP app to Europe?
08:01I don't think so.
08:03I don't think today...
08:06Look, even Elon Musk,
08:07when he acquired X,
08:09he wanted to build a super app.
08:12That hasn't happened.
08:13I think today...
08:14And the world changes so fast,
08:17but I don't think both in Europe
08:18and in the US right now
08:21there's an appetite for this.
08:23What might change this in the future
08:26is maybe with Identity KI,
08:27if we talk more about that in the future,
08:29is maybe that will create those new opportunities,
08:31but today I don't think it's really an opportunity.
08:34Motor shrinking,
08:35business models are changing,
08:36Agenting AI,
08:37let's get into it.
08:38Everybody's talking about Agenting AI
08:40and what it means for their fortunes.
08:42How do you perceive it?
08:43But I would also like to mention that
08:46besides super app framework,
08:49Tencent Cloud actually offers a lot of capabilities,
08:52offers a lot of platform technologies,
08:54such as our video services,
08:56very popular among the OTTs,
09:00and we also host a lot of games
09:03on our infrastructure,
09:05and especially with low latency,
09:08network latency.
09:09This is perfect for online games
09:13with multiple players,
09:14and these are important sectors
09:17that we serve in the European markets.
09:20And at the same time,
09:21we're working with a lot of European companies
09:24to go to China,
09:27help them with their go-to-market strategies.
09:29in China,
09:31in Asia-Pac,
09:32which is also a growing market
09:35for a lot of European companies.
09:37Yeah, you've got a lot of sticky users
09:39with that gaming base,
09:40don't you?
09:41Agenting AI,
09:42let's get to that
09:43because it is such a big story,
09:45and everybody's talking about the agents coming.
09:47What does it mean to you?
09:49I think it's,
09:50Agenting AI is misunderstood,
09:52and it's an evolving definition.
09:54Like, for instance,
09:55if you go on the internet and say,
09:57find me the cheapest flight
09:58to go to New York,
09:59from Paris,
10:00is that Agenting AI?
10:02Not really.
10:05That's an AI search.
10:06It's an AI search,
10:07but I don't think it qualifies.
10:09Some people might call that Agenting AI,
10:10but I don't think it's the full vision.
10:12Now, let me tell you
10:13the Agenting AI version of that.
10:15You're going to ask your agent,
10:17say, hey,
10:17I give you two weeks,
10:19and over the next two weeks,
10:20you have to find
10:22a ticket to go from Paris to New York
10:24for less than 500 euros,
10:27and you have access
10:29to my credit card,
10:30you can talk to everyone,
10:31you can actually buy
10:32the ticket on my behalf,
10:34which means the banks
10:35have to trust you
10:37that I have pre-authorized you
10:39for the next two weeks
10:40to do that.
10:40That's Agenting AI.
10:42Is the trust there from the consumer
10:45to give over their personal banking details
10:48to hand over that decision-making to a bot?
10:52So trust is going to be
10:53at the center of everything.
10:55That is the reason why we're obsessed
10:57with ethical AI,
10:59agentic AI,
10:59and all those things.
11:00If you don't have trust,
11:01this industry will not take off.
11:04But yes,
11:05you have to bring that trust.
11:06You have to understand
11:07how you're going to convince
11:09the banking industry
11:10to accept that
11:13not you as a human being
11:14that's going to authorize
11:16a transaction on your phone,
11:17but to rely on an AI
11:19that's going to represent you
11:20and is going to have
11:21some kind of rights
11:22to do things on your behalf.
11:24Those discussions
11:25are taking place right now,
11:26and they're going to require
11:27a huge amount of trust,
11:28and those are the things
11:29we have to define
11:31with tremendous amount
11:32of transparency
11:33and explain with our users
11:36what is the new kind of consent
11:37that they're going to have
11:38to give in the future,
11:39and we have to do this
11:40in a way that's very easy
11:41to understand.
11:42The reason I was so key
11:43to hear from you on that
11:44is because I feel as though
11:46some consumers are different
11:47in Europe, say, to China.
11:50But China's consumer
11:51is so advanced
11:52when it comes to using technology
11:54and seemingly more willing
11:56to allow that trust
11:57to be handed over
11:58in terms of decision-making,
11:59allowing an agent
12:00to follow through the process.
12:01Do you agree with that?
12:02Well, I agree with that,
12:04and I also agree with Bruno
12:06on the fact that
12:07there are different levels
12:09of using AI capabilities
12:11in agents
12:12to accomplish different tasks.
12:15But for me,
12:17I am more comfortable
12:18with a somewhat blurry definition
12:21of what agents are.
12:23In fact, to me,
12:25it doesn't matter as much.
12:27Most importantly,
12:28I think we look at
12:29what the users need,
12:31what are the pain points,
12:32how can we make them
12:34more efficient, right?
12:35On the consumer side,
12:37I think there are a lot of apps,
12:39tools out there
12:40that help you with,
12:43make it easier
12:44for you to buy things,
12:46for you to book
12:47your travel itinerary
12:50and whatnot.
12:51On the enterprise front,
12:53in fact,
12:54we realize that
12:55there are many needs
12:58for specialized agents
12:59in different industries,
13:02and some of which
13:03might need smarter tech support services
13:07so that they can help their users
13:09in a shorter period of time.
13:13Some others want to automate,
13:16get their coding
13:18more efficient,
13:20build the products faster.
13:21some of them
13:22want to create creatives
13:24for their advertising targeting
13:26more efficiently.
13:28So I see all kinds of scenarios
13:30where we can leverage
13:32agentic technologies
13:34in combination
13:36of different types of models
13:38to create tools
13:40that are tailored
13:41for specific use case.
13:43If you can both indulge me,
13:45I started out my week
13:46talking to some of the biggest names
13:48in advertising and marketing.
13:49In fact,
13:50I've spoken to the two big names
13:51in the industry,
13:53and the question I was asking
13:54was,
13:55will we see the marketers
13:56having to market to the boss?
13:58Because what you're spelling out
14:00is this autonomous decision-making
14:02that someone else is
14:03doing the research
14:05and deciding to book
14:06and carry out the decision.
14:08Does that mean
14:09the advertisers,
14:10the marketers in the room
14:11will actually have to advertise
14:13to the algorithm?
14:16Is that how it plays out?
14:17So the easy answer is yes.
14:22The more complex question is
14:25what language
14:26then do you need to speak?
14:27Because if you are talking
14:29to an agent,
14:31what will happen
14:32in the beginning
14:33is an agent will speak
14:34a human language
14:34to another agent
14:37so that you can understand
14:38what's going on.
14:39But over time,
14:40what we see today
14:41with semantic communication,
14:43agents will talk directly
14:45to other agents
14:46speaking a language
14:46that you don't understand.
14:48And that's the thing
14:49that's the most exciting
14:50and scary at the same time.
14:52They're going to make up
14:53their own language.
14:54And there were like
14:54a few experiments
14:55where you had agents
14:56talking to an agent
14:58and it started in English
14:59and then after like 10,
15:0020 minutes
15:01you could not understand
15:02the thing.
15:02The language morphed
15:03to something that was
15:04much more optimal
15:05and allow them to have
15:06a much broader bandwidth.
15:08Just to give you a sense,
15:10today when you're dealing
15:10with communication in AI,
15:12there's this concept
15:13of embeddings.
15:14Think about 12,000 dimensions
15:17to understand
15:18the actual meaning
15:19of a world.
15:20That's the wealth
15:21of understanding
15:21of a concept
15:22that you have in AI
15:23to make it work.
15:24If you have two systems
15:25talking to each other,
15:26they're not going to speak
15:27with 1,000 words.
15:28They're essentially going
15:29to speak using
15:30a much more complex language.
15:33I think in order
15:34to make these tools popular,
15:36being widely used,
15:38you have to give consumers
15:39the power to decide.
15:42Some users might want
15:45more control.
15:46They want the agents
15:48to do things along the way
15:50but yet give them
15:51checkpoints where they
15:53can make decisions
15:55whether to move forward
15:57or choose the path.
15:58for the next step.
16:01But there will be users
16:02who just want
16:04an easier life,
16:07authorize your agents
16:09on behalf of payment authorization
16:13and using your identity
16:16in different settings.
16:18So especially, I guess,
16:21the new generation,
16:22the younger generation
16:23born with AI
16:24will most likely
16:27more comfortable
16:27with such delegation.
16:30But I can imagine
16:31my parents
16:32will not be
16:33anywhere close to that.
16:35Let me tackle
16:36some of the big themes here.
16:38Geopolitics has come
16:39into the world
16:40of technology
16:40in a big way.
16:41We were hearing
16:43from Jensen Huang
16:44last night on stage
16:45talking about
16:46sovereign AI saying,
16:47why would you
16:48outsource it?
16:49It is your intelligence,
16:51it's your data,
16:52it's your culture,
16:54why wouldn't you seek
16:55to keep it closer
16:57to home?
16:58How are you both
16:59thinking about
16:59the implications
17:00of that
17:01if there are
17:03domestic partners
17:03on the ground
17:04building out
17:05sovereign AI?
17:06What is the implication here?
17:08How do you think
17:08about it?
17:08I think it's a complex
17:10question because
17:11what we know
17:12for sure
17:14and we're
17:15in France here
17:16that you can't
17:17build an entire
17:18solution end-to-end
17:19that will be
17:19100% French.
17:21NVIDIA is not
17:22a French company
17:22and it's going
17:24to take decades
17:24before we get
17:25an alternative
17:26to those kind
17:27of GPUs
17:28that is French.
17:30But then at the
17:30same time,
17:31can you build
17:32a sovereign AI
17:33solution
17:34where part
17:35of that,
17:35the solution
17:37can come from
17:37China or the US?
17:39I think so.
17:40I think what matters
17:41is having a very
17:42strong understanding
17:43of what we call
17:44the control points
17:45and how the data
17:47flows
17:47and what is the
17:48kind of privacy
17:49that you can
17:50commit to.
17:50And at Orange,
17:51for instance,
17:52where we essentially
17:53obsess and committed
17:54to trust
17:55and building
17:56sovereign solutions,
17:57we're not telling
17:58our customers
17:58that everything
18:00has to be French
18:01or European.
18:02We're saying that
18:03we have to make sure
18:04we understand
18:04how the data flows,
18:05we have to make sure
18:06that we have to
18:07what you're doing,
18:08you understand
18:08what are the implications.
18:10And at the end
18:11of the day,
18:11Karen,
18:11what you're going
18:12to see is
18:12you're going to have
18:13use cases
18:14where people are saying
18:15I'm using OpenAI,
18:16for instance,
18:17and it's fine
18:18if the question
18:19I just asked
18:20goes back
18:21to a cloud
18:21in California.
18:22Then I'm going
18:23to ask a more
18:24sensitive question
18:25and then I want
18:26to make sure
18:26that I understand
18:27exactly where
18:28the data goes.
18:28For instance,
18:29I want the data
18:29to stay in France,
18:30I want the question
18:31to stay in France,
18:31I want to make sure
18:32that that LLM
18:33could be an OpenAI
18:35or an open-weight version
18:36of OpenAI
18:37or DeepSeq,
18:37for instance,
18:38but it has to run,
18:39for instance,
18:40in Europe.
18:41So it's a mix
18:42of where are you
18:43going to use
18:43this technology
18:44as opposed to
18:44where it's coming from,
18:46how do you control it,
18:47how do you operate it,
18:48where the data flows.
18:48Dawson?
18:50Bruno and I
18:51were discussing
18:52backstage
18:53about the needs
18:55for transparency.
18:57We are a big
18:58open-source adopter.
19:00Many of the components
19:01on our cloud
19:02are based on open-source
19:04and we build
19:05optimized versions of it,
19:06but many of our customers
19:08feel very comfortable
19:09using that
19:10because they don't feel
19:11being locked in
19:12and many of our AI tools
19:15are the same.
19:16So in our products,
19:19consumer-facing
19:20AI chatbots,
19:22we endorse
19:23DeepSeq
19:24because of the openness
19:25and we put a lot of effort
19:28to optimize it
19:29so that it can run
19:30more cost-effectively.
19:32In fact,
19:33in some of the markets
19:34in Southeast Asia,
19:37we are already deploying
19:38the open-source
19:40DeepSeq models
19:42to enable
19:43the developers
19:45such as Indonesia
19:47to build new apps
19:48on top of that.
19:49DeepSeq
19:50was such a big moment
19:51geopolitically in a way.
19:52It really signified
19:53the dangers
19:55of cutting Chinese technology
19:56out of the equation.
19:58Do it at your peril.
20:00And now that we've had
20:00a trade war unfolding,
20:02I think many people
20:03are saying,
20:04well,
20:04we cannot simply
20:05just rely on selling
20:06into the US market
20:08and having a huge reliance
20:10on US technology.
20:11Perhaps there is a role
20:12for Chinese tech.
20:13On the back of that,
20:15what role do you think
20:16Tencent can play
20:17in the build-out
20:18of AI in Europe?
20:20Well, first of all,
20:21I think it's important
20:22to have alternatives.
20:24Introduce competitions
20:25and ultimately
20:27the users,
20:27the customers
20:28would benefit
20:29from competition.
20:31They've got more choices
20:32and also,
20:34you know,
20:35there will be
20:36a particular
20:37vendor's product
20:38that meets
20:39your very specific needs
20:41better than others.
20:41I think choices
20:42are always good.
20:44We've got about
20:4525 seconds left.
20:46Let's give the audience
20:47a takeaway message
20:48because one of the big
20:49things we've heard
20:49is, look,
20:50AI is not being deployed
20:51at scale in businesses.
20:53It is at the tech level,
20:54it is for some of the
20:56major tech players,
20:57but business is not there yet.
20:58What would you say
20:59to business
20:59about speeding up
21:01their adoption
21:02of AI
21:03and getting to scale?
21:04Well,
21:06I think
21:06there's no doubt
21:08that everybody
21:08should jump on board
21:10and adopt AI
21:12in, you know,
21:13their workflow.
21:15Today,
21:16I think
21:17some people
21:18might wonder
21:19what is the ROI
21:20for such investment.
21:22I would say
21:23it's like
21:24compound interest.
21:25I mean,
21:26the sooner you invest,
21:28the faster
21:29you ultimately
21:30get the return.
21:32and at the same time,
21:35I think
21:35it's very important
21:37to keep
21:39a very open mind
21:40with the adoption
21:42of new technology.
21:43Bruno,
21:4320 seconds.
21:44We have
21:4673,000 users
21:47of AI at Orange.
21:48We led them
21:50the first year
21:50to explore,
21:51try everything
21:52they wanted to try,
21:53but we made sure
21:54that we safeguarded
21:55the way they were
21:56using the data.
21:57Now we're getting
21:57to stage two
21:58is improve
22:00customer experience,
22:01improve network
22:01efficiency,
22:03trying to figure out
22:04how we can bring
22:05better trainings
22:05to our population.
22:07So we're getting
22:08to, I think,
22:08the next level
22:08of maturity.
22:10And, you know,
22:10we're learning,
22:11we're making mistakes,
22:12we're keeping
22:13the dialogue open,
22:14and I think
22:15everybody needs,
22:16this is the most
22:17important thing,
22:17everybody needs
22:18to understand
22:18we will not do AI
22:21at their expense.
22:23Bruno,
22:23thank you so much,
22:24and Dawson,
22:25we do appreciate
22:25your time.
22:26Thank you for joining us
22:27at VivaSAC.
22:27Thank you, Karen,
22:28thank you, Bruno.
22:31Thank you.
22:32Thank you.
22:33Thank you.
22:33Thank you.
22:34Thank you.
22:34Thank you.
22:36Thank you.
22:37Thank you.
Commentaires