- il y a 1 semaine
Can AI Help Us Speed Up the Energy Transition
Catégorie
🤖
TechnologieTranscription
00:00Hello, thank you all for being here.
00:03My name is Jeremy Kahn, and I'm the AI editor at Fortune magazine,
00:07and I'm also the author of the forthcoming book Mastering AI,
00:11A Survival Guide to Our Superpowered Future, which will be out in July.
00:15We've got a great panel here to talk about whether AI can help us with the green energy transition.
00:21We have Philippe from Schneider Electric, Claire from Inria,
00:25Hiroshi from Mitsubishi Heavy Industries,
00:28and Tomas from the International Energy Agency.
00:32I just want to get started quickly, and I want to allow some time for questions.
00:37We will have some time for questions at the end, so please be thinking of your questions,
00:40and when I call for questions, if you put your hand up, we'll try to get a mic to you.
00:46I want to just sort of set the scene a little bit.
00:51Just last week, some of you may have seen that Microsoft,
00:54which had made a very big pledge that it was going to try to get not just to net zero,
00:58but net negative by 2030,
01:01said that their carbon footprint had actually increased by 30% since 2020,
01:08and they said that most of that was because of their big push to use AI more,
01:14and the way that they have installed specialized graphics processing units,
01:19which are the chips that are used to run most AI applications in their data centers.
01:24Now, most of their data centers run on green power,
01:28but these were all carbon emissions that are sort of secondary emissions
01:31that come from the manufacture of those chips
01:33and from the actual construction materials that are involved in building the data centers,
01:38so steel and concrete,
01:41which was sort of a dismaying, you know, finding.
01:44So I want to start out by asking our panelists,
01:47you know, do they think that the sort of AI revolution that we're seeing on balance
01:52is going to help us with this green energy transition,
01:55or is it actually going to make the problem much worse?
01:58And I'll just quickly go down the line and see what you all think.
02:01Why don't we start with Philippe?
02:03Yeah, it would be pretty ironic that we put so much effort in using AI
02:07to reduce energy need, optimal energy consumption,
02:11and at the end, negatively impact the planet.
02:14Probably what I would say is always good to keep the order of magnitude in mind.
02:19Data centers is 1% of the electricity used in the world,
02:240.2% of the energy used in the world,
02:27if with AI, and I think we'll dip into that,
02:30if we are able to reduce the energy needs of the rest,
02:34by let's say only 5% or 10%,
02:36if you save 10% or 99%,
02:39you can still increase a bit the 0.2%.
02:43So, my belief is that AI can really actually help the climate transition,
02:47claim the energy savings,
02:49despite the energy consumption associated with it.
02:52Great, and we're going to get into specifically how it may help in a minute,
02:55but I just want to, again, keep going down the line here
02:58and see if people think on balance,
03:00you know the technology is going to be a net positive
03:02or potentially quite problematic.
03:05Claire, what's your view?
03:06I'm glad Philippe talked about the question on balance
03:10because I really just wanted to answer for climate science
03:13and weather prediction and modeling.
03:15This is one area where AI can have a win
03:18in greening compute and reducing energy consumptions.
03:23Currently, when we want to do a physics-driven standard weather prediction
03:28or also our climate modeling efforts that inform the IPCC,
03:34we need hours to compute weather predictions
03:38and weeks to produce these climate predictions.
03:42And this is on high-performance computing,
03:44simulating the actual physics.
03:46So, machine learning can be used to essentially emulate
03:50these physics-driven models.
03:53You train them once,
03:54and then inference is on the order of seconds.
03:57And in terms of weather, this is already happening,
04:00for example, at the European Centre for Medium-Range Weather Forecasting,
04:04which has shown that some of the newest algorithms
04:06coming out of DeepMind, coming out of Huawei, etc.,
04:09actually perform better at forecasting weather
04:12than the physics-driven models.
04:13So, that's good news.
04:15Great.
04:15Hiroshi, how are you viewing this at Mitsubishi?
04:19I think that, you know,
04:22yes, they are trying to use the clean energy
04:25to run the data centre,
04:28but, you know, one-third of the energy
04:30is used for the cooling the chips.
04:34And I'm optimistic because, you know,
04:37now innovative solution
04:39to enhance the efficiency of the cooling
04:42is being developed.
04:44At Mitsubishi,
04:46we are developing innovative solution
04:49for the cooling system.
04:51That is emerging cooling.
04:54That is used liquid to cool the chips
04:57instead of air
04:58and enhance the cooling efficiency dramatically.
05:01So, that will reduce the consumption of electricity
05:04from the data centre.
05:06In addition to that,
05:08as you said,
05:09construction of the data centre
05:12involves lots of concrete and steel.
05:14We are working very hard
05:17to decarbonise these sectors,
05:21boosting efficiency
05:22of the energy efficiency
05:24and also deploying hydrogen
05:27or carbon capture technology
05:30in these sectors.
05:31So, we are optimistic.
05:33Fantastic.
05:33Thomas, are you optimistic or pessimistic
05:36about what the overall effect here is going to be?
05:39I think at the International Energy Agency,
05:41we're optimistic.
05:42I very much agree with the perspective
05:44given by Philippe.
05:45Let me just give you
05:47a little thought experiment.
05:48There are 5.4 billion internet users
05:51in the world today.
05:52Let's imagine that every one of them
05:55queried an AI model
05:57three times every day of the year.
06:00In terms of electricity demand,
06:02this would result in about
06:0330 terawatt hours of electricity demand.
06:07Global electricity demand
06:08is 26,000 terawatt hours.
06:11So, we need to keep things in perspective.
06:13If we look at the history of the data centre sector,
06:19data centre electricity demand
06:20has grown about 40% since 2015.
06:25But data centre workloads have grown 350%.
06:29The amount of internet traffic has grown 600%.
06:34So, there is a strong track record
06:36of energy efficiency also in the data centre sector.
06:40So, we need to keep the risk of the AI footprint in mind,
06:45but keep it in perspective
06:47relative to the size of the energy system
06:49and all of the rest of the things
06:50that we need to do to decarbonise energy.
06:52Fantastic.
06:54Since we have two representatives
06:56of industry here
06:58and manufacturing processes,
07:00I want to ask about
07:01the potential for AI
07:03to help optimise those processes
07:05and to really help decarbonise
07:07some of these sectors
07:08that I think,
07:10at least in people's minds,
07:11they think are very carbon intensive.
07:14Philippe, maybe we'll go to you
07:15and then to Hiroshi.
07:16But, at Schneider Electric,
07:18how are you using AI
07:19to try to optimise processes
07:21and decarbonise?
07:23Yeah, thank you for the question.
07:24And I will come back, actually,
07:26or start from what Claire was saying.
07:28If you want to optimise any process,
07:31industrial process
07:32or any kind of process,
07:35technically, you need a model,
07:36you need a simulation,
07:37you need something to be able
07:38to test your hypothesis
07:40and optimise the parameters
07:41of your process.
07:42One way is to do physical models.
07:45It's great,
07:46but it's usually very expensive,
07:48takes a lot of knowledge,
07:50very complicated,
07:51and it's actually reserved
07:53to high-end, complicated processes.
07:55If you want to optimise
07:57much more simple processes
07:58like basic industrial processes
08:00or even the process
08:01of heating and cooling
08:03a building like this one,
08:04it's actually much cheaper
08:06and much faster to use AI.
08:08Leveraging on the data
08:09accumulated since many years
08:11because digitisation is not new.
08:13companies have been now
08:15accumulating data
08:16on their processes
08:17since years.
08:18So using this data
08:19to build a model
08:20using AI
08:21is usually cheaper,
08:23faster,
08:24easier to use
08:25to optimise your process.
08:26So the first thing
08:28we use in Schneider
08:29and we help our customers use
08:31is use AI
08:32to build models
08:33easier than physical models
08:35and use these models
08:36to optimise their processes
08:38from industrial
08:39to building
08:40to whatever processes.
08:41What kind of savings
08:42are you seeing
08:44by doing that?
08:45It really depends.
08:47If you start from
08:47a non-optimised industry
08:49at all
08:50like heating
08:50and cooling a building,
08:52you can save 30%.
08:54If you go to an industry
08:55which are already optimised
08:57to test their processes
08:58with traditional ways,
09:00saving 2-3%
09:01is already a good achievement
09:02but still a lot.
09:04People would,
09:05I would say it,
09:06sell the soul
09:07of their mother
09:07in these industries
09:09to find 2-3% energy.
09:11So,
09:11depending really.
09:12But it can start
09:13with 30%
09:14in things
09:14not optimised at all
09:16again like heating
09:17or cooling
09:17something like that.
09:18Great.
09:18And Hiroshi,
09:19what do you find
09:20at Mitsubishi?
09:21and I hope
09:21you haven't sold
09:22your mother's soul
09:22to achieve.
09:24I'll give you
09:25some examples
09:25where we are
09:26using the AI
09:27in our
09:27actual businesses.
09:30Process optimisation.
09:31AI can,
09:33I guess,
09:33analyse the
09:34production process
09:35and workflow
09:36and identify
09:38the inefficiencies.
09:39So,
09:40for example,
09:41we are
09:41group companies
09:42called Mitsubishi
09:43Logisnext
09:44handling
09:45manufacturing
09:46automated
09:48forklift.
09:49and the control
09:50system
09:51can,
09:52I guess,
09:53using AI
09:53and plan
09:54and optimise
09:55the best route
09:56in the warehouse.
09:57Another is
09:58quality control.
10:00AI can,
10:01I guess,
10:01monitor production
10:02quality
10:03in real time
10:04with a lot
10:05of data
10:06and reduce
10:07waste
10:08and rework,
10:09et cetera.
10:10Our group
10:11company,
10:12Prime Metal
10:12Technologies
10:13manufacturing
10:14facilities
10:15for the
10:15steel making
10:16has an AI
10:18control system
10:20called TPQC
10:21through process
10:22quality control
10:23using AI
10:25and the
10:26feature
10:26correct data
10:28from each
10:28steps
10:29and in
10:30the steel
10:31production
10:31and identify
10:33eliminate
10:34quality issues.
10:35Also,
10:36supply chain
10:37management
10:37is area
10:38that you
10:39can predict
10:40demand,
10:44routing,
10:45et cetera,
10:45and manage
10:46the production
10:48and inventory,
10:49et cetera.
10:50So,
10:50lots of potential
10:51areas where
10:52AI can help
10:53optimise processes
10:54and save energy
10:56and help us
10:56decarbonise.
10:58I want to talk
10:58about another
10:59aspect where
11:00AI can help,
11:00which is to
11:02better manage
11:03our power grids
11:03and to better
11:04manage renewable
11:05resources on
11:06those power grids.
11:07Claire,
11:07I know you've been
11:08doing a lot of
11:08work on this.
11:09Can you tell us
11:10a little bit
11:10about what some
11:11of these examples
11:11are of where
11:12AI can really
11:13make a difference
11:14in how we manage
11:15our power systems
11:16in order to
11:18really get the
11:18benefit of green
11:19energy?
11:20Sure.
11:21I'll give one
11:22example in the
11:23US where we've
11:24worked with the
11:24National Renewable
11:25Energy Lab
11:27to get better
11:29forecasts of when
11:30it's actually
11:31going to be sunny
11:32because the
11:33uncertainty around
11:34solar and energy
11:35and similar can be
11:36said of wind
11:37is when is it
11:39going to be sunny?
11:39and it turns
11:41out at least
11:42in the US
11:42where it's
11:43mostly market
11:44forces driving
11:45a grid operator's
11:46decision on
11:47how much to
11:48rely on
11:50renewables to
11:51balance its
11:51energy load
11:52is that if you
11:53have uncertainty
11:55in when it's
11:56going to be
11:56sunny, that
11:57translates to
11:58cost.
11:58Why?
11:59Because you
12:00have to have
12:01other energy
12:02sources available.
12:03It could also
12:04translate to
12:06sort of
12:06dirtier energy
12:08in that you
12:08might have
12:10a carbon
12:11heavy energy
12:12source as
12:13your backstop.
12:15So even a
12:16little lift in
12:17predicting when
12:18it's going to be
12:19sunny can
12:21reduce this
12:22cost and
12:22help incentivize
12:24grid operators
12:24to transition to
12:25renewables.
12:26And, you know,
12:28AI has now been
12:29shown to be
12:31good at
12:31prediction.
12:31You know, the
12:32idea of chat
12:33GPT, it's
12:34predicting sort
12:35of text that
12:36would come
12:36after previous
12:37text in a
12:38conversation.
12:39Well, applied
12:40to weather data,
12:41meteorological data,
12:43we've showed that
12:43deep sequence
12:44learning models can
12:45outperform numerical
12:47weather prediction
12:48at predicting solar
12:49irradiance, which is
12:50going to translate
12:51to the power that
12:52you can generate
12:53from your solar
12:53farm, at week
12:55ahead lead times.
12:56So this really
12:57helps with energy
12:58planning.
12:59Fun fact, if you
13:01go at the minutes
13:02up to, say,
13:04half an hour,
13:04it's really hard
13:06with your big
13:07fancy AI to do
13:08better than
13:09what's called
13:09persistence, which
13:11is saying, you
13:12know, the solar
13:14power in 30
13:15minutes from now
13:16is pretty well
13:17estimated with
13:18the current
13:19amount.
13:20So very simple,
13:22cheap algorithms
13:23like that work
13:24better than big
13:24fancy models at
13:25very short
13:26timescales, but
13:27we see AI
13:28helping at these
13:30week ahead to
13:31longer lead
13:31times.
13:32And then also
13:33when we talk
13:34about mitigation
13:35in France, we
13:36talk about
13:36attenuation of
13:38the worst risks
13:39of climate change.
13:40That's what
13:41actions can we
13:42take now to
13:44reduce carbon
13:45emissions?
13:46And EDF, the
13:47electricity utility
13:49here in France,
13:50is wanting to
13:52design the new
13:53wind turbine
13:54because winds
13:55are changing,
13:56place them in
13:58different places
13:59to be future
14:00proof because if
14:01the wind patterns
14:01are changing, we
14:02might have the
14:03wind turbines in
14:04the non-optimal
14:05places and
14:06similar for the
14:07PV panels.
14:08So there what
14:09we're trying to
14:09do is look at
14:10the only
14:12predictions we
14:13have, you know,
14:1410 to 100
14:15years into the
14:16future, of course,
14:17are from these
14:18big physics-driven
14:19climate models that
14:20inform IPCC.
14:21But they're at
14:22coarse grid scales
14:23because they're
14:24trying to predict
14:24very far in advance
14:26and because they're
14:27compute-heavy.
14:28So much bigger
14:29than a single
14:30solar farm or a
14:31single wind farm.
14:32So we're using
14:33AI, a generative
14:35probabilistic AI
14:36applied to the
14:39data of interest,
14:40solar, wind,
14:42temperature, to
14:44learn models that
14:45can map
14:47probabilistically
14:50domain adaptation
14:51and also
14:52downscale, get
14:53to a finer
14:54grid from
14:55climate models
14:56to observation
14:58data sets that
14:59we have in the
15:00present day.
15:01Then that model
15:02can be used to
15:03basically translate
15:04a climate model's
15:05prediction to
15:07local scales.
15:08And this is
15:09helping ODAF
15:10for designing
15:11the future PV,
15:13the future
15:13eolien or wind
15:14turbine and
15:15where to place
15:16them.
15:16That's great.
15:17I think Philippe
15:17wants to get in
15:18with a comment on
15:18this and then I
15:19want to go to
15:19Thomas in a
15:20minute.
15:20But Philippe,
15:21what were you
15:22going to say?
15:22Yeah, I love
15:22what you say and
15:24working for Schneider
15:24Electric, of
15:26course, everything
15:27moving towards
15:27more electric,
15:28more renewable
15:29is dear to my
15:30heart.
15:31And what you
15:32mentioned is
15:32super important
15:33and super clear.
15:34I wanted to
15:35balance it with
15:36the importance of
15:36the demand side
15:37because when we
15:38think renewable,
15:39we always think
15:40production side,
15:41forecasting of
15:42the production
15:42is extremely
15:43important.
15:44But I also
15:45believe, and
15:45especially a
15:46company like ours
15:47and all of us
15:48by the way,
15:49in our homes,
15:49in our buildings,
15:50have a role to
15:51play in the
15:51demand.
15:52Because if we
15:52want to maximize
15:53the green
15:54percentage of
15:55electricity produced
15:57in the electricity
15:58that we used,
15:59we need to
15:59avoid what we
16:00call the peak
16:01of consumption
16:02because when
16:03everybody wants
16:04electricity at the
16:05same time,
16:05and I hope
16:06Thomas will agree
16:06with me,
16:08almost the only
16:08way for the
16:09utilities is to
16:10start coal or
16:12gas-fired power
16:13plants which are
16:14flexible, easy to
16:15use by generating
16:16carbon.
16:16So we can also
16:17use AI at
16:18individual level,
16:19at home level,
16:20at building
16:21levels, to
16:22optimize what
16:23we call the
16:24prosumer to
16:25reduce the
16:25demand during
16:26the peak.
16:26Let's say in
16:27a building like
16:28that you have
16:28solar panels,
16:30some batteries.
16:31With AI, you
16:32can forecast your
16:32consumption, you
16:34can forecast your
16:36needs, you can
16:36forecast your
16:37production, leveraging
16:38the forecast coming
16:39from the grid from
16:40my neighbor, you
16:41can then optimize
16:42and avoid to buy
16:43from the grid at
16:44time of peak.
16:44So the combination
16:45of optimizing
16:46production with
16:47AI with
16:49optimizing local
16:50consumption, local
16:51demand with AI, is
16:52for me the two
16:53pillars that can
16:54really make us go
16:55into a green
16:55transition.
16:56Sorry for taking
16:57the word, but I
16:57had to say that.
16:58No, no, absolutely.
16:59Thomas, I want to
17:00ask you a little
17:01bit about this.
17:02And if you sort
17:04of look at policy
17:05interventions, where
17:07are you going to
17:07get the most
17:08impact?
17:08I mean, there's
17:08lots and lots of
17:09ways in which, you
17:10know, you can save
17:11a little bit here
17:12and a little bit
17:12there and it all
17:13adds up.
17:13But there may
17:14also be some
17:14really big
17:15wins, some
17:16things where, you
17:16know, where if
17:17we get the right
17:17policies, you
17:18know, we can
17:19make a really
17:19big dent in
17:20these things.
17:21And how do
17:21you kind of view
17:22the balance
17:23between these
17:23sort of big
17:24things versus
17:25lots of little
17:25things?
17:26Yeah, no, this
17:27is a really
17:27great question.
17:28I'm very glad
17:29that you asked
17:29it.
17:30So let me
17:30take the case
17:31of the power
17:32system.
17:34We, you know,
17:35very much agree
17:36with the
17:36perspective of
17:37Philippe that
17:38demand-side
17:39management is a
17:40critical tool
17:41for managing
17:43high-renewables
17:44power systems
17:45and integration
17:46with other
17:47flexible resources
17:48like batteries,
17:49for example.
17:51But if we look
17:54at, you know,
17:54what is making
17:56the renewables
17:58transition proceed
17:59more slowly than we
18:00need, I think there
18:01are a couple of
18:01additional challenges.
18:03So even if we
18:04could perfectly
18:06forecast renewables
18:08output and
18:09perfectly manage
18:10demand, there
18:12would still be
18:13periods of the
18:13year where
18:14renewables output
18:15is low and we
18:16would need backup
18:17resources for those
18:19periods even if we
18:21could perfectly
18:21forecast them.
18:23Right?
18:23And those backup
18:24resources in a
18:25high-renewable system
18:26that we simulate
18:27tend to be very
18:28costly because you
18:29don't use them
18:30very much.
18:31So for us, you
18:32know, one of the
18:33big wins would be
18:34looking at
18:35innovation to lower
18:36the costs of what
18:38we call seasonal
18:39flexibility.
18:40So flexible
18:41resources that you
18:42use for maybe
18:43three, four weeks
18:45of the year
18:45intensely.
18:46And that's where
18:47the application of
18:48innovation to AI,
18:50of AI to
18:51innovation could be
18:53very interesting.
18:54So for example,
18:55innovation in new
18:57long-term battery
18:59storage chemistries
19:01in chemical
19:03storage options,
19:04in thermal storage,
19:05for example.
19:07So for us, we
19:08would see the
19:09contribution of
19:10sort of the
19:10integration of AI
19:11into the energy
19:12sector as really a
19:13story of lots of
19:14these small wins
19:16that will allow us
19:17to keep the energy
19:17transition moving
19:18and then thinking
19:20very carefully and
19:21ensuring that there's
19:22a dialogue between
19:23the energy sector and
19:25the tech sector so
19:26that we can identify
19:27these problems and
19:28then apply AI to
19:30the problems that
19:31are most amenable
19:32to solutions from
19:34AI innovation
19:35because it's not all
19:36of them, you know.
19:37And I'll finish here.
19:38The renewables
19:39transition is also
19:40being slowed down
19:41by, you know,
19:43long permitting
19:44processes for
19:45transmission
19:46infrastructure.
19:47Okay, maybe AI
19:48can help us to
19:49speed up the
19:51submission and
19:52processing of
19:52permit applications
19:53but there's also
19:54a fundamental
19:54problem of public
19:56engagement and
19:57public acceptance
19:58which we can't get
19:59around with
19:59technology, okay.
20:01So we need to
20:02think very carefully
20:03in dialogue between
20:04the tech sector and
20:05the energy sector.
20:07What are the small
20:08wins that we need
20:09but also what are
20:10the big problems
20:11that we need to
20:11identify and work
20:12on together and
20:13if I have one
20:14message that I want
20:15to bring here it's
20:16this need for dialogue
20:17between the two
20:17sectors so that they
20:19can understand each
20:19other's problems but
20:20also each other's
20:21potential solutions.
20:23That's great and I
20:24want to ask the
20:25other panels about
20:26that dialogue in a
20:27minute but Thomas
20:28if there were, I
20:29mean you mentioned
20:29batteries, what would
20:31you say were the
20:31one or two big
20:34breakthroughs that you
20:35think would be kind
20:36of game changing
20:38if we could achieve
20:39them and where AI
20:40might have a role
20:41to play?
20:43So I would mention
20:45three and just
20:46caveat what I'm
20:47saying by announcing
20:49that the IEA is
20:50going to be launching
20:50a program of work
20:53on this topic so
20:54we will be looking
20:55in depth at the
20:56intersection between
20:57AI and energy and
20:58we look forward to
20:59partnering with a lot
21:00of different
21:01institutions in that
21:03but I would identify
21:04three.
21:05The first one is
21:06energy storage.
21:08It's not just
21:09batteries of course
21:10although batteries are
21:11one of the most
21:11interesting technologies
21:12but we tend to say
21:14batteries and think
21:15of it as a single
21:16thing but even
21:17within lithium-ion
21:17there's a huge
21:18diversity of battery
21:19chemistry and then
21:20we have other
21:22chemistries that are
21:23emerging that are
21:24interesting for
21:24different applications
21:25so I think of
21:26batteries as a suite
21:27of technologies that
21:28could potentially play
21:29different roles within
21:31the energy system.
21:33The other is high
21:35temperature heat.
21:36This is a really
21:38significant problem in
21:40the industrial sector.
21:42There's interesting
21:43emerging technologies
21:45around using surplus
21:46renewables to generate
21:47high temperature heat
21:49that can then be stored
21:50and applied in
21:51industrial processes
21:52and so you have this
21:53combination of energy
21:54storage and application
21:56in industrial processes.
21:57That's another
21:58technology area that I
21:59think is very
22:00interesting.
22:02And the third is
22:03carbon management.
22:04So we will need to be
22:06able to manage carbon
22:07across the energy sector
22:09including for example
22:10to make synthetic fuels
22:12for aviation.
22:13We'll need to be able
22:14if to do that in a
22:14climate neutral way
22:15you need to take carbon
22:17out of the atmosphere
22:18because you can't take
22:18it from the ground.
22:20And so technologies
22:22that could help us to
22:23manage carbon across
22:25the energy system
22:26you know chemical
22:27innovation in more
22:28effective absorbance
22:30that take carbon out
22:31of the atmosphere
22:32with less energy input
22:33for example
22:34that could be a big
22:35game changer to unlock
22:36you know additional
22:37emissions reductions
22:38in sectors like
22:39aviation or shipping
22:41which are probably
22:42going to continue to
22:43need to run on
22:44carbon based fuels.
22:46Great.
22:46I want to ask Hiroshi
22:48and Philippe about
22:48this issue of dialogue
22:49between sort of tech
22:50and industry.
22:52How would you sort of
22:53characterize the status
22:54of that dialogue right
22:54now?
22:55Are you guys sort of
22:56speaking the same
22:56language or do you
22:57feel like when you
22:58talk to technologists
22:59or people in the
23:00tech industry
23:00they're not necessarily
23:02focused on the issues
23:03that you as somebody
23:04in industry feel
23:05you know really need
23:06to be addressed?
23:07How would you sort of
23:08say that dialogue is
23:09at the moment?
23:10Maybe we'll go to
23:10Hiroshi and then
23:12Philippe.
23:13Hiroshi why don't you
23:14go first.
23:14Oh no Philippe go
23:15first.
23:15They're not fine.
23:16Sorry for that.
23:17Yes there is a gap
23:18and I will give an
23:19example that maybe
23:20speaks to some of
23:21you.
23:22In Schneider we
23:23manufacture millions
23:24of very small
23:25products in which
23:26the microcontroller
23:27costs let's say
23:28five six dollars.
23:29Every day I've got
23:30a startup calling
23:31me saying Philippe
23:32we now have a very
23:33cheap microcontroller
23:34that can run AI.
23:35It costs only
23:36150 dollars.
23:38So yes there is a
23:39gap definitively
23:40we don't always speak
23:41the same language
23:42but I believe that
23:43company like
23:44Orishi company
23:45or Schneider Electric
23:46it's actually our
23:47job because we
23:49are both two feet
23:51deep in the ground
23:52talking about
23:54electricity energy
23:55production and at
23:57the same time
23:57focusing quite a lot
23:58on AI.
24:00And the gap is not
24:01that big.
24:02You've heard about
24:03the edge, you've
24:03heard about the far
24:04edge, the things
24:05are becoming more
24:06and more intertwined
24:09one of the gaps
24:10and then I will let
24:11Hiroshi continue
24:12one of the gaps
24:12that is still
24:14super important
24:15for the industry
24:16is what should
24:17run in the cloud
24:18what should run
24:19on premise
24:21and there is still
24:22work being done
24:23but even on the
24:24LLM which are
24:24complicated we see
24:26now LLM that
24:26can run on premises
24:28so from my point
24:29of view this gap
24:29is getting closer
24:30and closer
24:31and closer
24:31but definitively
24:33it takes effort
24:34to make sure
24:35we speak the same
24:36language.
24:36If I take even
24:37internally in a
24:38company like Schneider
24:39we decided to
24:40build a central
24:41team of AI
24:41of more than
24:42300 people
24:43one of the
24:44biggest challenges
24:44is how we would
24:45make that team
24:46able to speak
24:47understand and
24:48work with the
24:49rest of the
24:50company which is
24:50140,000 employees
24:52and sometimes
24:54I categorize the
24:55AI people in
24:55three categories
24:56when I meet
24:57them people in
24:57Schneider
24:58some we say
24:59yeah it's another
25:00hype we're going
25:01to disappear
25:01soon the one
25:03who believe it's
25:03going to solve
25:04everything that
25:05they have been
25:05trying to solve
25:06since 50 years
25:07and then the
25:07more realistic
25:08our job is to
25:10help people to
25:10go to the
25:11realistic side
25:11yes it can do
25:13a lot of things
25:13no it will not
25:14solve the thing
25:15I've been trying
25:15to solve since
25:1650 years.
25:17Hiroshi I see
25:18you smiling and
25:18nodding there
25:19when we talk
25:21with the customers
25:22they are not
25:25concerned not
25:25just electricity
25:26but also heat
25:27so they would
25:28like to manage
25:29entire electricity
25:31apart energy
25:31electricity and
25:33heat and power
25:34optimize
25:35so we are
25:37proposing the
25:38customer to
25:38utilize AI
25:39the energy
25:40management
25:41total energy
25:42management
25:42systems
25:43using grid
25:44electricity
25:45in house
25:46PV
25:46battery storage
25:48how you
25:49use heat
25:50steam
25:50etc
25:51we have
25:52I can say
25:53kind of a
25:54total solution
25:55proposing the
25:56customer
25:56using AI
25:57got it
25:58I want to
25:58go to
25:59questions
25:59from the
25:59audience
25:59so if you
26:00have a
26:00question
26:01please raise
26:01your hand
26:02and wait
26:03till Mike
26:03comes to you
26:04and then you
26:04can stand up
26:05and ask your
26:06question
26:06who has
26:07questions
26:08please raise
26:09your hand
26:09if you have
26:10questions
26:10if not
26:10I will
26:11oh there's
26:11a question
26:11there in
26:12the back
26:13gentleman
26:14there
26:14keep your
26:14hand up
26:15until Mike
26:15gets to
26:16you
26:16so they
26:16can see
26:17you
26:17hopefully
26:18we can
26:18get a
26:18mic
26:18to this
26:18person
26:19there's
26:19another
26:19one
26:19there
26:20but here
26:20we go
26:20let's
26:21go to
26:21this
26:21gentleman
26:21first
26:22thank you
26:23if you
26:23could
26:23stand
26:23up
26:24and
26:24please
26:24state
26:24your
26:25name
26:25and
26:25where
26:25you're
26:25from
26:26hi
26:27so I'm
26:28Charlie
26:28I'm a
26:29computer
26:29scientist
26:31from
26:31Insalion
26:33and I
26:34would like
26:34to ask
26:34you a
26:35question
26:35now
26:36it's
26:37have
26:38you ever
26:38heard
26:39of
26:39rebound
26:39effect
26:41and
26:41I
26:42think
26:42you
26:42avoiding
26:43a bit
26:44the
26:44question
26:44with
26:45AI
26:46here
26:46because
26:46I
26:47think
26:47the
26:48real
26:48challenge
26:49here
26:49is
26:49how
26:50to
26:51use
26:51AI
26:51for
26:52sobriety
26:53so to
26:54use
26:54less
26:55energy
26:55in
26:55the
26:55end
26:56and
26:57not
26:57just
26:57optimize
26:58things
27:00so I
27:00guess
27:00the
27:01question
27:01is
27:01how do
27:01you
27:02how do
27:02you
27:02use
27:03AI
27:03not
27:03just
27:03to
27:04optimize
27:04things
27:04but
27:05to
27:05can
27:06you
27:06say
27:06the
27:07question
27:07again
27:07what's
27:08the
27:08question
27:08again
27:09yes
27:09so
27:09it's
27:10how
27:10to
27:10use
27:10AI
27:11to
27:12be
27:12more
27:12sober
27:13using
27:14less
27:14energy
27:15for
27:15to
27:16use
27:16less
27:16energy
27:16not
27:17just
27:17to
27:17optimize
27:17the
27:18processes
27:18we
27:18have
27:19Claire
27:20do
27:20you
27:20have
27:20a
27:23I
27:23think
27:23it's
27:23important
27:24to
27:24state
27:24that
27:25AI
27:25is
27:26we
27:26cannot
27:26rely
27:27on
27:27AI
27:28to
27:28address
27:29the
27:29climate
27:29crisis
27:30it's
27:30not
27:31a
27:31panacea
27:32we
27:33actually
27:33need
27:35changes
27:35at the
27:36policy
27:36level
27:37and
27:37changes
27:38in
27:39lifestyle
27:39you know
27:40reduce
27:41the flights
27:41that we
27:42take
27:42and
27:42our
27:43reliance
27:44on
27:44petroleum
27:46reduce
27:47driving
27:48etc
27:48so
27:49yeah
27:50I agree
27:50that
27:51it
27:51can
27:52only
27:52be
27:52used
27:53in
27:53concert
27:54with
27:55major
27:55lifestyle
27:56and
27:58sort
27:59of
27:59economic
28:00changes
28:01because
28:02the
28:02climate
28:03crisis
28:03is
28:04up
28:05and
28:05running
28:05and
28:05we
28:06needed
28:06solutions
28:06yesterday
28:07so
28:07we're
28:08not
28:08going
28:08to
28:08have
28:09a
28:10technological
28:12safety
28:13net
28:13to
28:14save
28:14us
28:14great
28:16I
28:16know
28:17there's
28:17another
28:17question
28:17somebody
28:18else
28:18had
28:18their
28:18hand
28:19up
28:19if
28:19you
28:19raise
28:19your
28:19hand
28:20again
28:20we're
28:20going
28:20to
28:20get
28:21a
28:21mic
28:21to
28:21you
28:21there
28:22at
28:22the
28:22end
28:22if
28:23we
28:23can
28:24please
28:24stand
28:24up
28:25and
28:25state
28:25your
28:25name
28:27I'm
28:28working
28:28for
28:29Le Monde
28:29I'm
28:29a
28:29journalist
28:30this is
28:30a
28:30question
28:31maybe
28:31for
28:31Mr.
28:32Spencer
28:32you
28:33said
28:33that
28:33overall
28:34there
28:34would
28:34be
28:36maybe
28:36more
28:37save
28:38of
28:38energy
28:38through
28:39AI
28:39than
28:40there
28:40will
28:40be
28:40rays
28:41of
28:41consumption
28:42but
28:42will
28:42there
28:43still
28:43not
28:43be
28:43problems
28:44on
28:45the
28:45grid
28:45just
28:46to
28:46get
28:46enough
28:46electricity
28:47for
28:47those
28:47data
28:48centers
28:48we
28:48are
28:48now
28:49pushing
28:51and
28:52you
28:52can
28:52see
28:52that
28:52there
28:53will
28:53probably
28:53be
28:54not
28:54enough
28:54energy
28:55in
28:55certain
28:55places
28:57that's
28:57a
28:57really
28:58good
28:58question
28:58and
28:58it's
28:59one
28:59I
28:59was
28:59going
28:59to
28:59ask
29:00as
29:00well
29:00about
29:00how
29:01you
29:02know
29:02where
29:02do
29:02you
29:02put
29:03industry
29:03where
29:03do
29:03you
29:03put
29:04data
29:04centers
29:04do
29:05we
29:05have
29:05enough
29:05grid
29:06capacity
29:06to
29:07actually
29:07bring
29:07this
29:08technology
29:08online
29:08Thomas
29:10yeah
29:11thanks
29:11very
29:11much
29:11I
29:12think
29:12this
29:12is
29:12a
29:13great
29:14question
29:14and
29:16in
29:16a
29:16way
29:16I
29:17think
29:17you
29:17answered
29:17it
29:17yourself
29:18in
29:18the
29:18way
29:18that
29:18you
29:18framed
29:19the
29:19question
29:19because
29:20you
29:20said
29:20not
29:21enough
29:21energy
29:22in
29:22certain
29:22places
29:23and
29:24I
29:24think
29:24that's
29:24exactly
29:25the
29:25way
29:25that
29:25we
29:26would
29:27see
29:27it
29:27what
29:28we
29:28see
29:28is
29:29that
29:29data
29:30centers
29:30for
29:30various
29:31reasons
29:31are
29:32quite
29:32geographically
29:33concentrated
29:34so
29:34you
29:35have
29:35a
29:35huge
29:36data
29:36center
29:36hub
29:37in
29:38northern
29:38Virginia
29:39you
29:40have
29:40a
29:41large
29:41data
29:42center
29:42industry
29:43in
29:43Ireland
29:43and
29:45in
29:45those
29:45places
29:46certainly
29:47what we
29:48see
29:48we
29:49track
29:49very
29:50carefully
29:50the
29:51projects
29:52that are
29:53coming up
29:54for new
29:54data
29:55centers
29:55we
29:56look at
29:56the
29:56reports
29:57of
29:57the
29:57transmission
29:58system
29:58operators
29:59around
29:59the
29:59world
30:00and
30:00certainly
30:01what
30:01we
30:02see
30:02is
30:03the
30:03risk
30:03of
30:04local
30:04bottlenecks
30:05that
30:06there's
30:06not
30:07enough
30:07transmission
30:07capacity
30:08there's
30:09potentially
30:09not
30:10enough
30:10electricity
30:11generation
30:12capacity
30:13in
30:13these
30:14localized
30:14places
30:15my
30:16point
30:16was
30:17only
30:17that
30:18if
30:18we're
30:19looking
30:19at
30:19the
30:19overall
30:20balance
30:21of
30:21AI
30:21whether
30:22it's
30:23positive
30:23or
30:23negative
30:23on
30:24the
30:24negative
30:25side
30:25we
30:25have
30:25the
30:26footprint
30:26question
30:26and
30:27so
30:27all
30:28I
30:28was
30:28saying
30:29is
30:29that
30:29if
30:29you
30:29look
30:30at
30:30it
30:30in
30:30the
30:30context
30:31of
30:31the
30:31huge
30:32scale
30:32of
30:33the
30:33global
30:33energy
30:33system
30:34that
30:35footprint
30:35question
30:36for
30:36now
30:37starts
30:38to look
30:38quite
30:39small
30:39but
30:39certainly
30:40at the
30:40local
30:41level
30:41we are
30:41seeing
30:42concerns
30:42and
30:43challenges
30:43about
30:44connecting
30:44data
30:45centers
30:45to the
30:46grid
30:46fast
30:46enough
30:48that's
30:48an
30:48interesting
30:49point
30:49and I
30:49also
30:50I
30:50started
30:51out
30:51the
30:51session
30:51by
30:52mentioning
30:52these
30:52statistics
30:53from
30:53Microsoft
30:53and
30:54Microsoft
30:55saying
30:55that a lot
30:56of that
30:56is
30:57coming from
30:58the
30:58manufacturing
30:58of the
30:59chips
30:59most of
31:00those
31:00chips
31:00are
31:00made
31:01in
31:01Taiwan
31:01by
31:02TSMC
31:03TSMC
31:04if you
31:04look
31:04they're
31:04trying
31:04very hard
31:05to
31:05decarbonize
31:06but they
31:06have this
31:06problem
31:06that in
31:07Taiwan
31:07there's
31:08not
31:08actually
31:08that much
31:08green
31:12wind
31:12farm
31:12that
31:13may
31:13make
31:13a
31:13big
31:13difference
31:14to that
31:14but
31:14there's
31:15probably
31:15a couple
31:15of these
31:16key
31:17choke points
31:17around the
31:18world
31:18where
31:18if you
31:19could get
31:19more green
31:20energy
31:20it might
31:21make a huge
31:22difference
31:22again
31:22on
31:23maybe
31:23certain
31:24places
31:24where
31:24steel's
31:24being
31:24made
31:25where
31:26if we
31:26could
31:26just
31:26get
31:27a little
31:27bit
31:27more
31:27green
31:27power
31:28in
31:28some
31:28of
31:29those
31:29places
31:29it
31:29would
31:29make
31:29a
31:30huge
31:30difference
31:30I
31:30don't
31:30know
31:30if
31:31this
31:31is
31:31something
31:31you
31:31guys
31:31have
31:31looked
31:32at
31:32Schneider
31:33or
31:34Hiroshi
31:34again
31:35as I
31:36said
31:36I
31:36think
31:36that
31:38reduction
31:38of the
31:38consumption
31:39of energy
31:40in data
31:40center
31:41is one
31:41of the
31:42target
31:42or challenge
31:43we have
31:43to do
31:44so as I
31:44said
31:45we are
31:45working
31:45on that
31:46got it
31:47and Felipe
31:47there are
31:47places
31:48you look
31:48around
31:48the world
31:48and say
31:49if we
31:49could
31:49just
31:49make
31:50a
31:50difference
31:50in a
31:50few
31:50of
31:50these
31:51spots
31:51it
31:51actually
31:51would
31:52be
31:52a
31:53huge
31:53step
31:54forward
31:54yeah
31:54we are
31:55not
31:55producer
31:56of
31:56energy
31:56per se
31:57but I
31:57think
31:58that
31:58we need
31:58to look
31:58at
31:58the
31:59problem
31:59at
31:59the
31:59different
31:59scales
32:00what
32:00Thomas
32:00was
32:01mentioning
32:01is the
32:01high
32:02scale
32:02of
32:03the
32:03transmission
32:03and the
32:04production
32:04and most
32:05of the
32:06electricity
32:06and the
32:07energy
32:07will come
32:07from that
32:08no doubt
32:09we also
32:10need I
32:10think
32:10to develop
32:11a lot
32:11of
32:12what we
32:13call
32:13prosumers
32:13which
32:14means
32:14take
32:14data
32:15centers
32:15of course
32:16it can
32:16rely
32:16mostly
32:17on the
32:17grid
32:18but it
32:18can
32:19also
32:19rely
32:19on
32:20its
32:20own
32:20solar
32:21panel
32:21farm
32:22on
32:22its
32:23own
32:23wind
32:23generation
32:24and
32:25therefore
32:25remove
32:26some
32:26of
32:26the
32:26load
32:26of
32:27the
32:27grid
32:27and
32:27that
32:27are
32:27things
32:28on
32:28which
32:28we
32:28work
32:28much
32:29more
32:29which
32:29is
32:29how
32:30can
32:30we
32:30equip
32:31those
32:31people
32:31whose
32:32main
32:33job
32:33is
32:33not
32:34to
32:34produce
32:34electricity
32:35their
32:35main
32:36job
32:36is
32:36to
32:36manage
32:37a
32:37data
32:37center
32:38or
32:38building
32:39like
32:39this
32:39one
32:39or
32:40a
32:40factory
32:41how
32:41can
32:42we
32:42automize
32:42thanks
32:43to AI
32:43the
32:45optimization
32:45and
32:46coming back
32:46to the
32:46question
32:47optimization
32:47is
32:47most
32:48of
32:48the
32:48time
32:48reducing
32:49the
32:49overall
32:50energy
32:50need
32:50or
32:51optimizing
32:51the
32:52percentage
32:52of
32:52green
32:53energy
32:53so
32:53back
32:54to
32:54my
32:54point
32:54how
32:55can
32:55we
32:56help
32:56them
32:56with
32:56AI
32:57by
32:57forecasting
32:58by
32:58optimizing
32:59maximize
33:00the
33:00amount
33:00of
33:00energy
33:01used
33:01and
33:02produced
33:02locally
33:02versus
33:03what
33:04they're
33:04going to
33:04need
33:04and
33:04buy
33:04from
33:05the
33:05grid
33:05and
33:06I
33:06think
33:06that
33:06the
33:06combination
33:06of
33:07all
33:07these
33:07efforts
33:08will
33:08bring
33:09us
33:09into
33:09a
33:09capacity
33:10to
33:10provide
33:10enough
33:11decarbonized
33:12electricity
33:12to
33:13everyone
33:14including
33:14data
33:15centers
33:15great
33:16we have
33:17time
33:17for
33:20oh
33:20we've
33:21got
33:21a
33:21bunch
33:21now
33:21this
33:22woman
33:22in
33:23the
33:23front
33:23row
33:23here
33:24if
33:24we
33:24can
33:24get
33:24a
33:24mic
33:25to
33:25her
33:26and
33:27please
33:27state
33:27your
33:27name
33:30good
33:30morning
33:31I'm
33:31Lily
33:32I
33:32studied
33:33earth
33:33sciences
33:34and I'm
33:34going to
33:34study
33:35data
33:35science
33:36and
33:36artificial
33:37intelligence
33:37now
33:38and
33:39I had
33:39a question
33:40because
33:40we
33:40talked
33:41about
33:41well
33:42you
33:42talked
33:43about
33:43the
33:46energy
33:47consumption
33:48but
33:49you
33:49haven't
33:50talked
33:50about
33:50the
33:50carbon
33:51footprint
33:51that
33:52much
33:53is
33:53it
33:53a
33:54really
33:54big
33:55negative
33:56point
33:56or
33:57is
33:57it
33:57something
33:57that
33:58can
33:58be
33:58compensated
34:00that's
34:00a good
34:00question
34:01so the
34:01question
34:01is
34:02on
34:03the
34:03carbon
34:03footprint
34:03as opposed
34:04to
34:04just
34:04the
34:05energy
34:05consumption
34:06and
34:06there
34:07is
34:07this
34:07issue
34:07that
34:07many
34:08of
34:08these
34:08data
34:09centers
34:09as I
34:09mentioned
34:09at the
34:09beginning
34:10are
34:10run
34:10on
34:11renewable
34:11power
34:11so the
34:11carbon
34:12footprint
34:12and
34:12the
34:12energy
34:12consumption
34:13may not
34:13be
34:13the
34:14same
34:14thing
34:14but
34:20look at
34:20all
34:20the
34:20people
34:21working
34:21in
34:21data
34:22science
34:22or
34:23studying
34:23data
34:23science
34:24use
34:24the
34:25right
34:25size
34:25tool
34:25for
34:26the
34:26right
34:26problem
34:26when I
34:27hear
34:27people
34:28coming
34:28and
34:28saying
34:29great
34:31news
34:31we can
34:32now use
34:32large
34:33models
34:33to make
34:34time series
34:34prediction
34:35I say
34:36bad news
34:36because
34:36you're
34:37going to
34:37use
34:37and
34:38generate
34:38thousands
34:38more
34:39carbon
34:39to get
34:40the
34:40same
34:40outcome
34:41so one
34:42of the
34:42things
34:42as a
34:43data
34:43scientist
34:43when you
34:44work
34:44on an
34:44AI
34:45algorithm
34:46to
34:46optimize
34:47or reduce
34:48energy
34:48consumption
34:49reduce
34:50therefore
34:50carbon
34:50footprint
34:51you need
34:51to think
34:52in a
34:52frugal
34:52way
34:53what do
34:54I really
34:54need
34:54and you
34:55were saying
34:56that Claire
34:56a few
34:57minutes ago
34:57that very
34:58simple models
34:58give excellent
34:59results
35:00do not go
35:00to a very
35:01complicated
35:02model that
35:02would cost
35:03a thousand
35:03more in
35:03carbon
35:04to just
35:05get a
35:050.1%
35:06of improvement
35:07and my
35:08belief is
35:09let's not
35:10fall in the
35:10hype of
35:11Gen AI
35:11Gen AI
35:12is great
35:12but not
35:13for everything
35:14and don't
35:14use a
35:15hammer
35:15as we
35:15say in
35:15French
35:16to crack
35:16a nut
35:17if you
35:17can do
35:18it with
35:18a very
35:18simple
35:19tool
35:20that would
35:20be my
35:21strong
35:21recommendation
35:21and always
35:23you are
35:23very right
35:24always think
35:24what is
35:25the carbon
35:25footprint
35:26of what
35:26I do
35:26versus
35:27what I
35:27save
35:28usually
35:28as Thomas
35:29was saying
35:29super
35:30positive
35:30in Schneider
35:31we say
35:311 to
35:321000
35:32for most
35:33of the
35:33energy
35:34savings
35:34applications
35:35but
35:35if you
35:36don't
35:36use
35:36the
35:37right
35:37AI
35:37model
35:37you
35:38can
35:38reduce
35:38advantage
35:39very
35:39quickly
35:40that's
35:41a good
35:41point
35:41Philippe
35:42but as
35:42you know
35:42you probably
35:43have your
35:43CEO saying
35:43but if we
35:44tell the
35:44market
35:45that we're
35:45using
35:46Gen AI
35:46our stock
35:47price
35:47will go up
35:4820%
35:49and we get
35:50that big
35:50gain
35:51tell it
35:51don't do
35:51it
35:52right
35:53other questions
35:54who else
35:55has a
35:55question
35:55I saw
35:55a couple
35:55there's
35:56another
35:56question
35:56in the
35:56front row
35:57if we
35:57can get
35:58a mic
35:58to the
35:58gentleman
35:59here
36:05hello
36:06I'm
36:07Adrien
36:08I work
36:09in
36:09consulting
36:09and my
36:10main
36:11customer
36:11at the
36:12moment
36:12works
36:13in
36:13electricity
36:14and
36:15we see
36:16a lot
36:17of
36:18struggles
36:19with the
36:19industry
36:20because
36:20as of
36:21today
36:22there's
36:22a lot
36:23of
36:23decentralization
36:24in the
36:25production
36:25of
36:26electricity
36:26through
36:27intermittent
36:28ways
36:28such as
36:29wind
36:30turbines
36:30and
36:30solar
36:31partners
36:31I have
36:32a question
36:34for our
36:35industry
36:35panelists
36:36with the
36:39intermittent
36:40character
36:41of the
36:41electricity
36:42production
36:43that arises
36:44as of
36:44now
36:44do you
36:45have
36:45any
36:45means
36:46to
36:46address
36:46the
36:48surplus
36:48of
36:48production
36:49when
36:49the
36:50supply
36:50of
36:50electricity
36:51out
36:51balance
36:52the
36:53demand
36:53that's
36:55an
36:55interesting
36:55question
36:55so
36:56yeah
36:56if you
36:56have
36:56intermittent
36:58surplus
36:58from
36:59industry
36:59is there
37:00and
37:00you know
37:00I guess
37:01it's a
37:01question
37:01about
37:01how you
37:01manage
37:02the
37:02grid
37:02to
37:02some
37:02extent
37:03how
37:03that
37:03handles
37:03it
37:04but
37:04Hiroshi
37:04why don't
37:05you
37:05he
37:06wants
37:06he always
37:06wants
37:07Philippe to
37:07go first
37:07it's
37:07very polite
37:08of him
37:08but
37:09Hiroshi
37:09why don't
37:09you go
37:10and then
37:10we'll
37:10go to
37:24Philippe
37:25one
37:25one of
37:25one of
37:25ways
37:25that
37:25store
37:26electricity
37:27storage
37:28that
37:28is
37:28an
37:28important
37:29point
37:29but
37:30another
37:30way
37:30that
37:31you
37:31optimize
37:34generation
37:35and demand
37:36that
37:37is
37:38one of
37:39the
37:39areas
37:39we
37:39will
37:40think
37:40about
37:40and
37:40AI
37:41can
37:41work
37:41on
37:42that
37:42area
37:42I
37:43think
37:43Philippe what
37:44are you
37:44guys seeing
37:44at Schneider
37:45I think
37:46I will give
37:46some answers
37:47I think
37:47that Thomas
37:48could do
37:48a better
37:48job
37:48because
37:49it's
37:49really
37:49a
37:49grid
37:50problem
37:50but
37:50from
37:51an
37:51industrial
37:51point
37:51of
37:52view
37:52I would
37:53say
37:53two
37:53things
37:54again
37:54forecasting
37:55if you
37:56are able
37:56to say
37:57that
37:57you can
37:57wait
37:58for
37:58when
37:58the
37:59energy
37:59is
37:59cheap
38:00because
38:00you
38:00are able
38:01to
38:01forecast
38:01your demand
38:02your
38:02production
38:02then you
38:03can
38:03optimize
38:04your
38:04cost
38:04reduce
38:05your
38:05impact
38:05on the
38:06carbon
38:06reduce
38:07your
38:07ecological
38:08footprint
38:08so
38:08a lot
38:09of
38:09the answers
38:09will be
38:10around
38:10the
38:10forecasting
38:12on the
38:12other
38:12side
38:13and here
38:13now I
38:14step out
38:14of
38:14Schneider
38:15and I
38:15speak
38:15and I
38:16will let
38:16maybe
38:16Thomas
38:17compliment
38:17but
38:17I was
38:18yesterday
38:18at a
38:19conference
38:19done by
38:20Luc Raymond
38:20the CEO
38:21of EDF
38:22he was
38:23mentioning
38:23those
38:23things
38:24called
38:24step ups
38:25what they
38:26do is
38:27when
38:27electricity
38:27is
38:28negative
38:28when
38:29it's
38:29super
38:29cheap
38:30they
38:30use
38:30it
38:30to
38:31bring
38:31the
38:31water
38:31back
38:31in
38:32the
38:32dams
38:32and
38:33by
38:33doing
38:33that
38:33they
38:34can
38:34generate
38:34the
38:35equivalent
38:35of
38:35two
38:35to
38:35three
38:36nuclear
38:36reactors
38:37when
38:37energy
38:38is
38:38needed
38:38so
38:39I
38:39stop
38:40here
38:40but
38:40I
38:40think
38:40part
38:40of
38:40the
38:41answer
38:41will
38:41be
38:41on
38:41this
38:41big
38:42application
38:42on
38:49it's
38:51a
38:51grid
38:51optimization
38:51question
38:52again
38:52and
38:53what
38:54policies
38:54need
38:55to be
38:55put
38:55in
38:55place
38:55and
38:56how
38:56different
38:56places
38:56around
38:57the
38:57world
38:57are
38:57doing
38:57on
38:57this
38:58question
39:01surplus
39:01energy
39:02is
39:02going
39:02to
39:03become
39:03and
39:03is
39:04already
39:04a
39:04big
39:05problem
39:05in
39:05high
39:05renewables
39:06systems
39:06we
39:07see
39:07it
39:08in
39:08Europe
39:08with
39:09this
39:10year
39:10long
39:11periods
39:12with
39:12low
39:12or
39:12negative
39:13prices
39:13on
39:13the
39:14electricity
39:14market
39:14the
39:15market
39:15is
39:15sending
39:15us
39:15a
39:15signal
39:16that
39:16there's
39:17surplus
39:17energy
39:18there's
39:18basically
39:19three
39:19things
39:19that
39:19you
39:20can
39:20do
39:20you
39:20can
39:20curtail
39:21the
39:21energy
39:21not
39:22dispatch
39:22it
39:22that
39:23is
39:23actually
39:24often
39:24an
39:25optimal
39:25strategy
39:25particularly
39:26when
39:26renewables
39:27are
39:27very
39:27cheap
39:28the
39:29second
39:29option
39:30is
39:30you
39:30can
39:30store
39:32so
39:32storage
39:33technologies
39:34like
39:34batteries
39:34like
39:35pumped
39:35hydro
39:35but
39:37also
39:37new
39:38emerging
39:38technologies
39:39like
39:40for
39:40example
39:41using
39:41surplus
39:41electricity
39:42to
39:42generate
39:43hydrogen
39:44store
39:44the
39:44hydrogen
39:45and
39:45then
39:45use
39:46the
39:46hydrogen
39:46to
39:47generate
39:47electricity
39:47at
39:48times
39:48of
39:48deficit
39:49and
39:50the
39:50third
39:50thing
39:50that
39:50you
39:50can
39:51do
39:51is
39:52transfer
39:53that
39:53surplus
39:54to
39:54an
39:54area
39:54with
39:55a
39:55deficit
39:55via
39:55transmission
39:56of course
39:57and we
39:58will need
39:58all of
39:59those
40:00options
40:02while I
40:03have
40:03to
40:03say
40:05one
40:05thing
40:05that
40:06has
40:06not
40:06been
40:06raised
40:07but
40:07if
40:07you
40:07allow
40:07me
40:09go
40:09ahead
40:09we
40:10have
40:10a few
40:11minutes
40:11I
40:11just
40:12wanted
40:12to
40:12add
40:13we've
40:13been
40:14having
40:14a
40:14conversation
40:14that
40:15is
40:15looking
40:16a lot
40:17at
40:17the
40:17issues
40:18faced
40:18by
40:19developed
40:19countries
40:19how to
40:20meet
40:20data
40:21center
40:21demand
40:21there's
40:22also
40:22an
40:22issue
40:23the
40:23developing
40:24world
40:24has
40:24big
40:25energy
40:25problems
40:25energy
40:26access
40:27affordability
40:28etc
40:28and
40:29we
40:30at
40:30the
40:30international
40:30energy
40:31agency
40:31are
40:32also
40:32looking
40:32at
40:32how
40:33AI
40:34can
40:34be
40:34applied
40:35to
40:35these
40:35problems
40:36so
40:36we
40:36worked
40:37with
40:37MIT
40:37to
40:38build
40:38a
40:39model
40:39essentially
40:39to
40:40forecast
40:40energy
40:41demand
40:42at the
40:42household
40:42level
40:43in
40:43Africa
40:43to
40:44enable
40:44more
40:45optimal
40:46access
40:47strategies
40:47on the
40:47part
40:48of
40:48utilities
40:48so
40:49I
40:49just
40:49wanted
40:49to
40:49bring
40:49to
40:50the
40:50table
40:50that
40:50this
40:51is
40:51not
40:51just
40:52the
40:52developed
40:52world
40:52issue
40:53we
40:53also
40:53need
40:53to
40:53think
40:54about
40:54how
40:54this
40:54technology
40:55can
40:55be
40:55applied
40:55in
40:56the
40:56developing
40:56world
40:56and
40:57Clara
40:57when
40:57you
40:57do
40:58sort
40:58of
40:59predictions
40:59on
41:00weather
41:01and
41:01on
41:01climate
41:02change
41:02are
41:03you
41:03are
41:03most
41:04of
41:04these
41:04just
41:04for
41:05developed
41:05places
41:06or
41:06are
41:06you
41:06also
41:06looking
41:06globally
41:07at
41:07the
41:07emerging
41:08markets
41:08and
41:08what's
41:09happening
41:09there
41:09yeah
41:10I
41:11think
41:11I
41:12like
41:12to
41:12view
41:14AI
41:14as
41:15a
41:15way
41:15to
41:15go
41:16from
41:18data
41:20abundant
41:21regions
41:21to
41:22data
41:23scarce
41:24regions
41:25we
41:26have
41:26worked
41:27on
41:27this
41:27with
41:28some
41:28partners
41:29in
41:29India
41:29and
41:30the
41:30idea
41:30is
41:31that
41:33generative
41:34AI
41:34now
41:34I'm
41:34not
41:34talking
41:35about
41:35chat
41:35GPT
41:36but
41:36probabilistic
41:37AI
41:37that
41:38can
41:38train
41:39itself
41:39self
41:40supervised
41:40AI
41:41unsupervised
41:42AI
41:42can
41:43be
41:44validated
41:44against
41:45supervised
41:47models
41:47and
41:48other
41:48sorts
41:49of
41:49well
41:50calibrated
41:50models
41:51in
41:51the
41:51regions
41:52where
41:52we
41:52have
41:52abundant
41:53data
41:53which
41:54is
41:54often
41:54global
41:55north
41:56regions
41:56Europe
41:57North
41:57America
41:58and
41:59then
41:59they
41:59can
42:00be
42:00applied
42:01in
42:01the
42:02data
42:02scarce
42:02regions
42:03in
42:03the
42:03global
42:04south
42:04and
42:05fine
42:06tuned
42:06with
42:07the
42:07available
42:08data
42:08for
42:09the
42:09application
42:10so
42:10this
42:11is
42:11a
42:11really
42:11exciting
42:12approach
42:13again
42:13it
42:14is
42:14not
42:14a
42:14panacea
42:15we
42:16need
42:16to
42:16see
42:17some
42:17market
42:18levers
42:19helping
42:20in
42:21the
42:21developing
42:21world
42:22and
42:22this
42:22is
42:22not
42:23even
42:23just
42:23in
42:23the
42:23developing
42:24world
42:24like
42:24in
42:24the
42:25US
42:26we
42:27have
42:27a
42:27history
42:27of
42:28climate
42:28injustice
42:29and
42:30AI
42:30is
42:30not
42:31the
42:31silver
42:31bullet
42:31but
42:32we
42:33can
42:33do
42:34some
42:34things
42:34of
42:35going
42:35from
42:35data
42:36abundant
42:37regions
42:37to
42:37data
42:38scarce
42:38regions
42:38using
42:39AI
42:39models
42:39guys
42:40you've
42:41been
42:41a
42:41very
42:41good
42:42audience
42:42and
42:43we
42:43are
42:43unfortunately
42:43just
42:44out
42:44of
42:44time
42:44but
42:45I
42:45think
42:45we've
42:46heard
42:46some
42:46reasons
42:47to
42:47have
42:47some
42:47optimism
42:48about
42:48how
42:49AI
42:49can
42:50help
42:50with
42:50the
42:51green
42:51energy
42:51transition
42:51but
42:52I
42:52think
42:52the
42:52overall
42:52message
42:53is
42:53it's
42:53no
42:53silver
42:53bullet
42:54it's
42:54not
42:54going
42:54to
42:54solve
42:55climate
42:55change
42:55for
42:56us
42:56and
42:57we're
42:58not
42:58going
42:58to
42:58be
42:58able
42:58to
42:58click
42:58our
42:58fingers
42:59or
42:59wave
42:59a
42:59magic
42:59wand
43:00with
43:00AI
43:01and
43:01suddenly
43:01have
43:02a
43:03net
43:03neutral
43:04future
43:04but
43:05it
43:05can
43:05definitely
43:05help
43:06in
43:06a
43:06variety
43:07of
43:07ways
43:08which
43:08you've
43:08heard
43:08some
43:09about
43:10today
43:10and
43:10hopefully
43:11giving you
43:11some
43:11food
43:12for
43:12thought
43:12anyway
43:12thank you
43:13to my
43:13panelists
43:13and
43:14thank you
43:14to all
43:14for
43:14listening
Commentaires