Accelerating Sustainability with AI with Andres Ravinet - #689

0:00The result from a KPI perspective was incredible. I mean, they went down to a reduction of less than 1 % of food waste. And from a starting point of an average, what we're seeing in their industry, which is about a 15 to 40 % food waste.

0:34All right, everyone, welcome to another episode of the Twin Mall AI podcast. I am, of course, your host, Sam Charrington. And today I'm joined by Andres Ravinet. Andres is Sustainability Global Black Belt for Data and AI at Microsoft. Before we get going, be sure to take a moment to hit that subscribe button wherever you're listening to today's show. Andres, welcome to the podcast. Thank you, Sam. Thank you. I'm excited to have you on and for our conversation. Let's start out by having you share a little bit about your role at Microsoft. Sure, sure. Thanks. And also, thanks for pronouncing the name correctly.

1:11I know it can be difficult. Yeah, so I'm a sustainability global black belt, GBB, at Microsoft that focuses on our solutions under our Microsoft for Cloud for Sustainability platform. And yeah, the GBB is a very funny name that we have at Microsoft that really just helps customers understand that we're a D technical and industry expert within a specific, either a solution area or areas and specific industries. And so what I do really is my day-to-day is supporting customers on their sustainability journey as it relates to leveraging our different data and AI solutions, especially under our cloud for sustainability.

1:56Awesome. Awesome. And when you talk about sustainability as it pertains to your role and what you're seeing at customers, what all does that entail? Sustainability is a lot. It's like a multifaceted concept. I think there are different definitions. I think generally we like to think about it as the concept of addressing the needs of the present for everyone without compromising future generations and the needs that will be met then. You know, what really that means is just really everyone from an individual to families, to communities, cities, countries, and organizations trying to understand the impact that they have from three perspectives, environmental care, social well-being, and economic growth, and trying to create a balance through practices or strategies to adhere to those and set commitments and practices in place to kind of address that.

2:58And what's your background? Did you, have you always worked in sustainability? Yeah, I have quite the jack of all trades background. So I'll start from the kind of from the now. I'm actually new to role, not to Microsoft, but to this role. I think it's been like three months already, but I've been with Microsoft for six years. And in that time, I worked for a small part of the organization called the Microsoft Technology Center as a technical architect, cross-solution technical architect. And so what we did was we met with hundreds of customers a year in our New York facility where we would just help them ideate, strategize, and then build solution architectures around various different solutions that we have in helping them address their business objectives and challenges.

3:48And since the release of our platform and our commitments, I was immediately drawn to being able to talk to this because I've had a passion in this area for a long time. And so it was just so surprised, one, on our ambitious commitments, and two, that we actually came out with a platform that we can help customers leverage. And so for the last two years, I've been speaking to customers about that in our MTC locations and virtually around the world, since we had a bit of an interruption there. I finally was able to move into a role, thankfully, at Microsoft that is dedicated to this space. So now I spend my entire time just talking to customers in this space.

4:29And before then, very much of a technical background in building solutions around virtualization, cloud technology, cybersecurity consulting. It actually has served me well in having like a multi-solution or multifunctional architectural conversation. And I've just been able to use that same kind of drive and systems thinking into sustainability and how to kind of look at that holistically for a customer and how to think about solution architectures for customers. When you think about sustainability, what are some of the big threats that sustainability as a field of practice seeks to address? I mean, climate change is one, but I'm imagining it's one of a number that folks are trying to address with their sustainability initiatives.

5:24So when I think of these big threats, I immediately think of like the large ones that are interestingly enough interconnected. And so it's things like climate change and its impact on the increase to extreme weather events, to the collapse of biodiversity and the impact to us, and as well as the shortages of natural resources. right and just to give you some examples of like some of the gravity in terms of some of the facts related to around this right it recent studies have found that you know our negative contribution to the environment is a large contributing factor to the extreme weather events and what's unfortunate it is that you know when we think of transitioning we have a this idea of a just transition to kind of moving towards sustainable practices.

6:20Those who were the least in contributing to this are the most affected by these sorts of events and these risks. The other one is in terms of wildlife population. We're seeing a significant decrease in wildlife population, especially in South America. South America is seeing like a 94 % drop in wildlife population. And that, again, has a domino effect. To the domino effect, right, in terms of the ecosystem, you know, if you think of like the pollinator population, as an example, bees for one, right, that's being affected dramatically. And when you think of what happens when the decrease of population in pollination begins, you know, we see a direct effect to the global food supply.

7:1175 % of our global food supply is reliant on pollinators. So these are large, large, grave sort of facts that we have to deal with. And so that kind of sets the stage, you know, for a lot of strategies that the world, organizations, governments, right, are looking to populate. from reducing greenhouse gas emissions and moving towards renewable energy sources, to adopting systems thinking, to like the idea of the interconnectedness between nature, to protecting habitats and ecosystems, and to moving towards more regenerative and sustainable practices around industries like fishing, mining, agriculture.

7:57And so at a technical level, right when you when we think about trying to address these it's really not a new strategy it's collecting data and the fact that there's a lack of data right and then having to kind of bring that in right do some analytics on it it's kind of that old adage of right like i can't manage it if i don't measure it right and so right now you know we're seeing that as a big big gap in the industry is for anyone trying to set commitments is collecting data to understand what they can do and how to monitor progress towards those commitments. Let's dig into some examples of how AI can be applied in some of the diverse areas that you just spoke about.

8:49One example, that's an early warning system, early warning for all. Can you tell us a little bit about that initiative and what it's aiming to accomplish? Yeah, the United Nations put together this initiative called the Early Warning for All. And the goal was, or at least their commitment was, they're looking to protect all humans on Earth from hazardous weather events by 2027. And so to help, they started ideating, well, what can we do with the data that's available out there in the world as it relates to what could help in being able to advise communities around upcoming extreme weather events?

9:37And so a couple of thoughts were, well, how can we use satellite imagery, meteorological data, as well as real-time data streams like social media and put that together? And so we worked with an organization called SEEDS, which stands for Sustainability, Environmental and Economical Development Society, to help build an AI model called Sunny Lives. And the SunnyLies model was trained manually by a number of data scientists on 50 ,000 images of rooftops. And the idea was, how can we build a model that one can identify the different rooftops of specific buildings and categorize them into seven types?

10:29And then as well correlated to different geographical factors like bodies of water or elevation. And so with the result of that, they were able to use this model in being able to personalize the advisories to specific governments and specific populations of people or communities of people on upcoming events, as well as helping nonprofits and government agencies. essentially disseminate resources ahead of time when they detect an impacting event. And it was timely - Is the idea there that by identifying or classifying these different rooftops, they can characterize the building practices in general and how susceptible they are to different levels of storms?

11:21This was necessary because of a lack of urban development in a lot of these areas. And two, it was also just to identify, so to your point, it's a yes. And as well as identifying where there are communities of people living versus where there has not been. And so just understanding distribution of communities of people in certain areas. So it turns out it was timely, right, after piloting it and putting it into production. In 2021, Cyclone Yaz was nearing the eastern states of India. and with the model they were able to identify the populations at risk or the communities at risk and be able to send personal advisories immediately to over a thousand families and and and helping government agency as well as distribute resources appropriately and efficiently like into those different areas so it turned out to be a great project and it's still in place today and and really helping the areas that that need it right as i mentioned you know those are the ones that are most affected by extreme weather events.

12:26That's interesting. You mentioned that there was some custom model development involved here. Was this project a research project or research in partnership with other organizations, or was it more a case of organizations using off-the-shelf tooling to create that model? It was definitely not off the shelf. It was a model that had to be created by a partnership between that SEEDS organization and ourselves. There was no model at the time that would help in this specific use case. So it was very much a partnership where we supported them from a technology perspective and a consulting perspective on what technologies that could be leveraged and then helping scale it.

13:07You also mentioned one related to food waste. Can you talk a little bit about that use case? Yeah. Food waste, large, large problem. So pervasive issue right now affecting both not only the environment, but the economy, right? UN reporting 931 million tons of waste of food annually, right? That amounts to about 17 % of all food available to consumers, you know? And so lots of organizations are thinking about, you know, how can we help address this due to that and that sort of two-faceted result, right? And it's all across the supply chain for food. An entrepreneur years ago who was trying to tackle this in restaurants and hotels, and the approach was like putting a camera over the trash can and using that to develop reporting that fed to management about how food was used and wasted so that they could do things like tweak portion size and ordering and stuff like that.

14:13It's such a complicated issue because there's so many hands passing, right? Food, right? When you think of the supply chain and every hop something has to take. Last I checked, the majority of food waste results before it is received by grocery stores. And so yeah, grocery stores have the ability and starting to take measures like using either image or video recognition and being able to tie that to some process that can hopefully optimize their purchasing process. But on the supply chain side, they're trying to understand what they can do in sort of demand forecasting, optimizing their processes.

14:56And that has been traditionally something that hasn't been done in that industry. And so a lot of organizations are taking different approaches on how to address that. Yeah, there are all these different, when you think of a value chain for a specific food item, there's arguably countless of companies involved from the beginning to the end of the life cycle of food. And so lots of organizations are trying to understand what they can do on their part. But as well as, you know, interesting enough, we are seeing more of requirements for customers upstream, right? So if you think of a grocery store, a lot of grocery stores are starting to dictate data related to kind of the processes that they have of, you know, what their waste profile is, what their environmental emission data is, what their code of conduct is, you know, from an ethics and social responsibility perspective.

15:47So we're seeing a lot of that from a requirement standpoint at the private level, but also at the government level, we're seeing regulations sort of dictate the disclosure of information like that. I was going to ask about the specific food waste project that you mentioned. We have a customer called LMK Group. They're a Nordic food kit delivery company. And when you think of that food waste problem we're talking about, right, there is a player throughout that entire supply chain. They were looking to understand, you know, what can we do from a demand forecasting side, a supply chain optimization side, and how to start thinking about personalized customer experience.

16:29These aren't new, you know, strategies, right? I think they are new in terms of the outcome, right? So they're thinking about it from the sustainability lens versus, yes, arguably optimizing efficiency, right? In their processes. The end result is great because if I could reduce waste, I mean, that's better for the world. The thought was, okay, so how do we develop machine learning models that can essentially predict customer orders with accuracy, And so what they did was, one, try to build a model that allowed them to bring in not only historical data, but real-time data to forecast demand. The goal was to help them provide suppliers precise predictions, right?

17:19You know, up to 10 weeks in advance. You know, when they started in this process, you know, the challenges, you know, are not dissimilar to other use cases, right? You know, I need to integrate data in disparate sources and different formats. I need to build machine learning models on some platform that can ideally scale it. We help them with our Azure machine learning platform to help build this model that allowed them to bring in that data and allow them to bring in different factors like seasonal changes and consumer preferences. And again, bringing from other external sources to essentially correlate this data.

18:02They were successful in building this machine learning model. They used a variety of different algorithms from time series analysis to regression models to essentially help in forecasting the future demand. The result from a KPI perspective was incredible. I mean, they went down to a reduction of less than 1 % of food waste. And from a starting point of what? From a starting point of an average, what we're seeing in their industry, which is about a 15 to 40 percent food waste, at least in their part of the supply chain. Right. As I mentioned, the majority of this is is further downstream in terms of the impact of food waste.

18:48But, you know, from a pattern perspective in terms of, you know, how AI can be leveraged here, the challenges seem to be the same, except I'm finding that customers that are in the realm or in the responsibility of sustainability with an organization, we're finding that this is new to them. New to them that they can apply AI as a tool to address these issues or the challenges themselves are new? What aspects are new? So what's new to them is what can be done with AI and machine learning models or what needs to be done with the leveraging of AI technologies and machine learning models. Um, so tangentially we're finding a lot of cross collaboration between this new aspect of a business, this new part of the business and technology teams.

19:52It sounds like the technical process here was first identifying this challenge and then the application of, you know, what I might call traditional data integration techniques to kind of get all the information about their supply chain in the right place. and then imagining traditional ML models as opposed to anything fancy. Yeah. Sounds like kind of a bread and butter type of problem that I think of it as unsexy AI sometimes. Yeah, it almost feels like table stakes. Yeah, yeah. Right? It's just a new business objective, new outcome. It's AI and these models have been so publicly available now.

20:40You know, it's such an interesting time to have access to all of this and really easily deploy this. And so thinking about, well, how do I build one from scratch, you know, and do I even need to build one from scratch? Right. And how do I get started? Like, what are these platforms that we can use to get started here? and what are these data sets that I can leverage? There's so much available out there. You know, I look forward to where this becomes as a mature AI and ML use case as it is for other aspects of a business, like from financial operations, right, to market trends, to other things, which arguably is sexier.

21:22It's really nice to start seeing this part play a role here. I want to make sure we cover a really interesting use case that is focused on conservation in particular in the Amazon. Can you tell us a little bit about that one? Yeah. Yeah. That one, I'm pretty passionate about it. It's a pretty cool use case, you know, kind of to set the stage when you think of the Amazon rainforest, it's the largest rainforest in the world. It spans nine countries and, you know, not many people know this, but, you know, it's absolutely crucial to keeping the balance of the ecosystem. It is one of the largest carbon removal entities in the world.

22:10It as well - It's kind of like what you were referring to earlier, just the interconnectedness of all of these systems. And that is a huge piece that plays a significant role globally. Yeah. I mean, you know, on the bleeding edge side, a bit off topic, but a lot of tech companies like ours are investing in other companies who are looking at carbon removable technology because it's needed. But we've got like free carbon removable technology right now in the form of biology, you know, and it's at risk. And on top of it being one of the largest carbon dioxide removers in the world, it also dictates weather patterns.

22:53And so as it depreciates or as it's infected or impacted, you know, that dictates the weather patterns around the world. And this is not a new problem, right? Deforestation, but it has been increasing. And so what we did was collaborate with our research teams and a number of different universities and agencies down in South America to put together a project called Project Guacamaya. And Guacamaya means macaw, like the bird, macaw in Spanish. And so the idea was, okay, well, how can we start addressing the detection of deforestation trends that are illegal. And so the idea was, can we start using satellite imagery analysis, camera trap imaging analysis, as well as bioacoustic analysis to essentially monitor the rainforest's health and biodiversity?

24:00And so again, from a challenge perspective, it's a very interesting one because it's such a vast geographical area, right? And we're having to start implementing real-time monitoring. And then on top of that, the complexity of analyzing multimodal data, right? So you think of imagery from a photo perspective, imagery from a satellite perspective. Now we're talking bioacoustics, right? How do I put that all together? And so on the satellite front, right, we all started leveraging the Planet Labs high-resolution satellite imagery. And what we did was we worked through those photos on a daily basis to identify signs of deforestation and illegal mining.

24:46And the AI model that we built helped track the changes over time and then detect specifically unauthorized roads. And the reason to detect unauthorized roads is because that typically often preceded deforestation activities. On the camera trap analysis side, right, there were camera traps across the rainforest that were able to take photos on detection of movement. And so thankfully in that scenario, there was a model that could help here called the mega detector. And what it did is it helped filter and classify images of what they called bio indicators, which is essentially a way of saying, you know, species whose presence or absence might indicate an ecological change.

25:37And then lastly, on the bioacoustic monitoring side, which I thought was pretty fascinating, the idea of capturing sound, specifically animal sounds like birds, and differentiate between the different species to identify either migration of species away from a specific area or the introduction of an invasive species into an area. All this being put together into an open source model now. So now, not only can they leverage it for the initiatives in the Amazon rainforest, but now the hope is that it's going to be adopted by other parts of the world with ecosystems like that that could be affected, that could leverage such a model.

26:22The model created on the bioacoustic side is an interesting one. We'll drop a link to the paper into the show notes. The end result is this contrastive language audio pre-training model that is a multimodal model that aligns the audio with textual information to help enhance the process of identifying particular sounds. Really interesting on the research front, as well as the practical application here. You know, we should and we have kind of taken it for granted that this is something addressing sustainability is something that organizations should do. But are there particular drivers that are prompting organizations like Microsoft and the customers that you work with to pay attention to these things now?

27:09I know government initiatives, for example, are a big part of that. You know, how do you think about the drivers? Yeah, I mean, the first one I was going to mention was the first one you mentioned, right? I think that's one of the biggest drivers. Unfortunately, that's sometimes what it takes is kind of strict government regulations to have people move towards progress. So yes, we are seeing stricter government regulations as well as industry-related agencies that have regulations. We are starting to see rumblings of that here in the US with a regulation out in California and one coming from the SEC.

27:47These are all looking to essentially collect or require collection of environmental data, specifically around scoped emissions. From a terminology perspective, we have different types of emissions, right? You can think of carbon emissions. You can think of the use of water to waste to biodiversity loss. emission data for scope one is really the direct emission of an organization, direct operations. Scope two is emissions related to their utilities. So things like consuming electricity, gas, heat, steam, and so forth. Scope three is the most complicated one. There are two aspects to scope three.

28:37Scope three, from a downstream perspective, is everyone in your supply chain. All the emission that they emit on your behalf is your scope three emissions. Then there's upstream. If you create a product and the use of that product, as an example, Xbox, right? And every time someone uses an Xbox, that's our scope three upstream emission. And so how do I start considering collecting that data and reporting that? That is by far the most difficult one, but yes. So we're definitely seeing stricter regulation from governments, but as well as consumer demand shifting. So we are seeing more and more customers looking to do business with those who are actually being responsible as it relates to sustainable business practices.

29:27And that is why a lot of people are starting to publicize their commitments around sustainability. From an operational perspective, right, this also is about cost savings, right? So cost savings in terms of reducing waste and energy consumption. It's also a smart risk strategy, right? You know, addressing issues like climate change and resource scarcity. The other is really around like corporate social responsibility, doing good, right? Making sure that you're contributing to the social good and environmental stewardship, right? Microsoft specifically, thankfully, we're in a position to do more to help with climate change.

30:07And we believe we should. I mean, it's built into our mission statement. Our mission statement is to empower every person and organization on the planet to achieve more. Very different from our original putting a PC on every desk on every home. right but also like the reality is you know we do better as a business when economies around the world are growing and doing well so we don't do well if our customers are not doing well and i'd also be remiss to to kind of not also bring up our commitments in this space as well right we have been in this space for quite a while now we became carbon neutral in 2018 and a few years ago we set the most ambitious commitments by 2030 we are looking to be carbon negative zero waste and water positive by 2050 we are looking to remove all historical emissions since the beginning of microsoft in 1975 and so these are big commitments are part Have enough positive impact that undoes essentially the historical emissions?

31:22Yeah. Through a combination of reducing our environmental footprint and through the investment into carbon removal agreements. Yeah. And so lots of drivers nowadays, far less are those doing for the sake of doing good. more starting to do more because they're being forced to do more from either the government perspective or consumer demand perspective. Along those lines, I recently spoke with one of your colleagues, Laurent Benoit, in a conversation around AI and power and energy. And one of the key topics that came up was the growing need for power to drive data center growth and a big part of that was driven by AI workloads.

32:09How does Microsoft being kind of central to this growing use of power around AI get reconciled with a drive towards sustainability and being responsible in using resources? This is a massive challenge that's in front of us, right? The idea of having to decarbonize the delivery of AI to the global economy. And we are seeing the reality of that through our numbers. We've just recently released our sustainability report for 2023. And this has been one of our challenges. And we are committed to a full stack end-to-end approach to tackling this problem, right? This is really a once-in-a-lifetime opportunity for us.