Episode 52 Transcript: How Salesforce is Pushing for Greener AI

Molly Wood Voice-Over:
Welcome to Everybody in the Pool, the podcast where we dive deep into the innovative solutions and the brilliant minds tackling the climate crisis head-on. I'm Molly Wood.

When people are feeling some despair about how to tackle climate change, they tend to say things like, "It doesn’t matter what I do; it only matters what big companies do." You know my feeling on this: it matters what everyone does.

But it’s absolutely true that big companies, through their actions, their investments, the tech they buy, their business practices, can have a really big impact.

So today, I’m talking with a climate leader at a tech company that’s cleaning up its own operations and is also trying to push new rules to ensure future technologies, namely A-I, don’t make our current climate problems even worse.

Megan Lorenzen:
My name is Megan Lorenzen. I'm a director of climate and energy here at Salesforce. I'm definitely living the dream. This is my dream job.

Molly Wood Voice-Over:
Salesforce is not even close to the most giant tech giant, but it is a Fortune 500 company, worth a little over 200 billion dollars, that provides cloud-based customer relationship management software to companies all over the world. And the cloud, as Megan mentioned, is actually just a whole bunch of computers located in data centers. Data centers consume one and a half to two percent of global electricity currently, and that’s only expected to increase as A-I uses more and more and more computing power.

Molly Wood:
Yeah, let's dig into that a little bit more because I think people may not realize the power that big companies have already had to drive change toward the energy transition driven, in some cases, by outside pressure, but also just efficiencies, like data centers use a lot of energy for example, just one simple example, right? They use a lot of energy and that's expensive. And so the more sustainable they can be, the cheaper they are.

Megan Lorenzen:
Definitely. And this is really, you know, there's a cost perspective for sure, but this has been really integrated into the way Salesforce operates as a business from day one. And we've really operated with this concept that business can be a platform for change. It started with our founders who kickstarted this one, one, one model, which is 1% of the company's equity, 1% of its product and 1% of its employees' time given back to the community.

Megan Lorenzen:
And so when we think about parts of our business, like our data centers, of course, we're looking for reliability, efficiency, but we're also thinking about sustainability. And that's why we were one of the early companies to commit to 100% renewable energy back in 2013, which really kickstarted a lot of our work on trying to drive power sector decarbonization.

Molly Wood:
Yep. And then how is the progress toward that goal?

Megan Lorenzen:
It's been wonderful. We achieved 100% renewable energy in 2021. So what this means is we match an equivalent amount of renewable energy or renewable energy attributes to the amount of electricity we consume globally on an annual basis. And so this has catalyzed a number of projects all around the globe from the US to Australia to some new projects being announced next month, which we're excited about.

Megan Lorenzen:
And just last year, we were able to do some really amazing work across the global south in markets like Africa, India, and Southeast Asia. So really it's been a great opportunity to think about what role we can play to catalyze the energy transition in different markets.

Molly Wood:
Right. And I should, I don't want to gloss over. I mean, I should sort of say you're being humble. Salesforce is one of the very few companies in the world, right, to source 100% renewable energy. Like this is a really big accomplishment.

Megan Lorenzen:
There are definitely a lot of companies working on this and it's really fascinating to hear the different stages that each company is at and the different theories of change that they've embraced on how they can be most impactful with their procurements. So you have some companies that are hyper-focused on procuring as local as possible. For Salesforce, we're part of, again, a coalition of our peers called Emissions First Partnership where we're really focused on where can you have the highest emissions impact for your procurement? So we're not just continuing to procure in California, for example, where we're headquartered that already has a lot of renewable capacity. 

And instead we're asking ourselves, where is this additional project needed and where will it be most impactful? Because one of the very interesting things about the market today is that most renewable energy procurements, by corporates, are taking place in North America and Europe. I believe the stat from Bloomberg New Energy Finance was 29 out of 31 gigawatts were in North America and Europe. And so you're leaving the global South behind. And that was a large part of this ethos around our procurement with DRX across the global South was really, there's a gap and a need and we were trying to focus on where can we drive that impact.

Molly Wood Voice-Over:
To translate that a little bit, if you just build another solar farm in California or even Europe, you might be adding another 2 percent of renewable energy to a place that’s already generating mostly renewable energy. But if you build a solar or wind farm or hydroelectric plant in a place where the alternative was going to be a new coal plant, you not only get credit for the renewable energy you’ve helped generate, but you also displace dirty fossil fuel emissions, meaning fewer emissions and also cleaner air and water in the place you built.

The Emissions First Partnership also includes Amazon, Meta, Intel, Akamai, GM, Rivian, Heineken, and WorkDay.

Molly Wood Voice-Over:
Ok so, let’s get to AI. As I mentioned, in April, Salesforce announced its Sustainable AI Principles, which, in addition to all the other things people are worried about with AI, are specifically designed to help build best practices AND regulation around the environmental impact of AI.

Megan Lorenzen:
Because we're at this really interesting inflection point. We're seeing AI take off in everything we do. And in the face of this escalating climate crisis, AI has the potential to either exacerbate it or help alleviate and accelerate our work to address environmental challenges. And we see regulation as having a critical role to play in helping us navigate that future and really maximize the potential benefits while mitigating some of the potential impacts. 

And so that's where our new sustainable AI principles come into play. And this is really about guiding conversations with lawmakers, trying to get ahead of the conversation and help indicate where there might be support or opportunity. So there are two major buckets, I'll just touch at a high level and then we can dig in a bit more detail. But the first is really about better measurement in management of the environmental impact of AI. I mean, we really cannot talk about AI without also acknowledging its impact on climate, largely due to the amount of compute power it requires, especially with training. And there have been some really fascinating studies emerge.

Molly Wood:
Mm-hmm.

Molly Wood:
Right.

Megan Lorenzen:
About data centers which currently represent about 1.5% of global power generation. And that is expected to grow to about 4.5% by 2030, largely as a result of AI. So we're seeing this energy demand increase already. And so how can we think about better measuring and managing that impact? And then the second piece is really about not just mitigating the harm.

Molly Wood:
Mm-hmm.

Megan Lorenzen:
But also accelerating the innovation we can achieve with AI. We have the ability now to analyze vast data sets, identify patterns and predict outcomes that we previously required a lot of tools and resources to do. And so how can we use government policies to incentivize and enable those AI tools that can accelerate innovation on climate and unlock some of those critical social benefits? So those were really the two main buckets we wanted to tackle as a starting point with these principles.

Molly Wood:
Let's talk about the internal action part of that. And I imagine that's where the measurement and management comes in, right? What does that look like practically?

Megan Lorenzen:
Definitely. And so with everything we're calling for, we're really focused on also walking the talk. And so we are focused on driving change within our own four walls, but we know that's not enough. So we're using policy in our platform to drive change at scale. So as we think about within our four walls, we've been really focused on sustainable AI in our own development. 

We've had an amazing AI team for many years now. And this has really been built into the ethos of how they operate. I touched on a few of these earlier, but let me just reiterate it here. So one is optimizing our models, particularly with AI, and this concept that bigger isn't better. So looking at domain-specific models, really customizing them for their intended application, optimizing for cost and emissions impact. 

And talking about this ever-growing power of small models. I think we've seen this race for giant models as of late, and that's not necessarily where we need to go or where we should go.

Molly Wood:
Can we break that down too for people who just may not be that familiar with what we mean when we say models and training, right? It's sort of like the difference between training a chat interface on every piece of information ever and or saying, I just want it to parse the Inflation Reduction Act for subsidies I can get at my house.

Megan Lorenzen:
Exactly. And I think what's important to note is that when Salesforce is thinking about our own AI application, we are doing an enterprise AI software. This is not a general-purpose model that has all of the information you could possibly find and you can generate cat videos or whatever it might be. This is very specifically intended to enable your business and augment the work that you and your employees are doing. I think a great example of this is we have Einstein, which is our AI friend, AI partner, embedded into Slack, our Slack platform. 

And I was able to drop a new report that just came out that I was fascinated about reading and it's a hundred plus pages. I was able to drop it in my Einstein tool and get a one-page summary to determine what parts of it might be most relevant for me to read further on or use that to help me with scheduling or different tasks that take time away from the work that you want to be doing. And so when we think about Salesforce's AI product, it's really this enterprise AI designed to augment your business. It's not the general-purpose AI models that are anything and everything in the world.

Molly Wood:
And then why is that a lower-energy interaction?

Megan Lorenzen:
Well, there are a couple of different ways in which it's a lower-energy interaction. So part of it is how we design the training. So this is a piece where it's definitely you want the technical engineers to lean in, but we're optimizing our models for efficiency with the multi-epoch training and flash attention. And we're able to get our models to perform as well as larger models at half the size. And so there are different technical optimizations that we've been able to implement with our model.

Molly Wood:
Right, I think the key is knowing that those exist, right? That there is in fact a way to develop something like this so that you're not sifting through every piece of information in the world to get one simple answer when you just need this narrow-focus tool.

Megan Lorenzen:
Yes.

Megan Lorenzen:
Exactly.

Molly Wood:
And then, so that's the management internally and it sounds like you're also trying to sort of solicit and send out best practices.

Megan Lorenzen:
Yes.

Megan Lorenzen:
We are very committed. We're really committed to where we can, sharing externally these best practices and learnings. And so we were able to actually publish quite a bit of this just last year, talking about these optimized models, talking about the power of small models, as well as the opportunities with efficient hardware, the role of siting data centers in low carbon regions, all of these come into play in driving down emissions. And then really rounding that out with disclosures. 

And so we have disclosed the emissions impact of our models and those carbon footprints, both pre-training and those are located on Hugging Face and some of the other leader boards out there, as well as our own website. And so really walking the talk and putting forth that information so other companies can follow that lead.

Molly Wood:
I mean, it's pretty powerful. Like, I think what we find is that the measurement is sometimes the place where not everybody wants to go. Like, you're pretty exposed when somebody starts putting the information out there.

Megan Lorenzen:
Yeah.

Megan Lorenzen:
Definitely, and it's an important balance, and especially in this moment of, we're seeing really critical regulations around the globe, driving companies to more transparently disclose their emissions. That's critical information, but you also don't want to be in a position where you're seeing the dollars and the time a company invests in climate and sustainability, only go to reporting. So reporting is part of the puzzle. And that disclosure is part of the puzzle and should be the way in which you identify where you need to focus and take action. And so there are certainly sort of two pieces to the puzzle here.

Molly Wood Voice-Over:
All right, let’s save the second piece of the puzzle for the second half of the show. Time for a quick break. We’ll be right back.

Molly Wood Voice-Over:
Welcome back to Everybody in the Pool. We’re talking with Megan Lorenzen, a director of climate and energy at Salesforce, about its Sustainable AI initiative. Part of that is trying to engage lawmakers to consider climate goals in any future AI regulation.

Molly Wood:
Can you talk about the high-level ideals? Like what are some of the policy outcomes that you would like to see even just around sustainable AI, for example, or AI in general?

Megan Lorenzen:
Definitely. Our North Star has really been about reducing emissions and scaling nature-based solutions to reach that net-zero future. So net zero emissions by 2050. And we think about how we need to get there. And that's about reorienting entire economies and financial systems around that future. 

Some of the specific things we're calling for, requiring AI disclosures for all general-purpose models. I think this is a really important, again, starting point, because where there's visibility and disclosure, we can drive action. And a large part of that is going to be the development of standardized metrics. 

A second piece has been around adding the environmental lens as a risk factor. So as governments are thinking about AI systems and starting to identify which AI system is high risk, because a lot of what we do could be captured in a really generic AI description. And so we want to make sure we're really focusing that additional attention on models that might pose additional risk, or in the EU AI Act, they've called it systemic risk. 

And so we're looking at training that might exceed a certain compute threshold or cause a certain threshold of emissions. So once you exceed that threshold, then those AI models should be considered high risk. 

And for those high-risk models, let's layer on those additional efficiency standards and benchmarks, whether it's optimizing the model for efficiency, right-sizing it, ensuring we've got those right energy-efficient AI algorithms in place. This could be around the efficient hardware that's used to both train and run these models.

Molly Wood:
Right.

Molly Wood:
It's fascinating because there's obviously a lot of conversation all the time about attempts to regulate AI and they mainly focus on deep fakes or disinformation or copyright violation or privacy. And so what you're really specifically saying is as you're crafting regulation, please also include environmental impact in any future regulations. What is the reception to that?

Megan Lorenzen:
I definitely think that we've got opportunity in front of us. The EU AI Act, which has the initial foundation, they've included some of the training thresholds, and in their definition have left the door open to additional criteria. 

Emissions weren't explicitly stated, but there's a toehold there on which you can build. We're seeing legislation proposed in different markets in this vein. So I really do see the momentum building that folks are understanding that the energy and emissions impact of AI models is something we should be thinking about.

Molly Wood:
It does sort of feel like if you constrain it in a lot of really natural ways that are also good for privacy or good for copyright, you therefore create potentially smaller models that are not as impactful and that those regulations could all feed each other in a way that could be good for all of us.

Megan Lorenzen:
Exactly. And it's not just about how do you minimize the harms, but it's also how do you unlock the potential benefits of AI? And so there are some really exciting opportunities.

Molly Wood:
Right, you read my mind. That's my second, yes. No, totally, you have gone exactly to the next place, which is okay, cool. Now how do we make it do some good stuff?

Megan Lorenzen:
Yes, exactly. And AI has such capability to analyze data, predict outcomes, generate really robust models. And so how can we create and maintain some of those open-source databases really around environmental data and make it freely accessible to really accelerate the R&D process and enable action? And this is so exciting because I'm already seeing some government organizations lean in here. 

Some great reports came out of NREL in the last week on exactly this front, leveraging AI and datasets and open-sourcing it for, in that case, grid operators. And so I think there's a lot of potential here for the government to unlock a more cohesive effort. And then the government also has an amazing role to play in fostering collaboration and some of those public and private partnerships where each sector, whether it's academia, private sector or government, we have something to bring to the table. And so how do you convene us and then open-source some of those learnings and best practices so it's accessible to communities that may not have the ability to develop those resources in-house. 

And then last and certainly not least is how do you then spur innovation? So grants, tax incentives, the like around the environmental application of AI, tools to optimize critical systems like the electricity grid, or to leverage AI to detect or monitor wildfires or illegal deforestation. There's so much potential there. How do we really spur that innovation? We're doing as much as we can within our own sphere, but if you can achieve that at the government scale, we can unlock so many ecopreneurs.

Molly Wood:
OK, so be more specific about what government could and should do. It sounds like some version of open data, which many of us have been asking for a long time. But availability of data seems like a key component that you're describing.

Megan Lorenzen:
Definitely. Yeah, it's definitely the, how can you open-source those environmental datasets in particular so that we are enabling, whether it's researchers, early-stage development of new technology, how can those folks just jump ahead with this data we now have available and how can we train AI models on all of this really deep environmental data? And then how can we really...

Molly Wood:
What data are we missing? What would you like to see? You mentioned NREL releasing some data. Can you give us some examples, some more examples of what is in there that's really juicy we could work with?

Megan Lorenzen:
Yeah, it's a great question. I think it's all of it. I know it's a terrible response, but you've got data captured in so many different siloed locations. So how can we have a shared open-source database so that if you are, let's say you're in academia and you're researching a specific topic, so much of your work goes into tracking down the right databases that then you have to manipulate and bring in-house for your specific application.

Molly Wood:
Hehehe

Megan Lorenzen:
How can we cut out some of that work so that you have databases available and you can input them into your AI models to support your work? So I think all of it is the short answer, unfortunately.

Molly Wood:
Well, I would guess, though, that that's exactly why public-private partnership is so crucial here, right? The government may have a lot of data, but maybe not a lot of super-efficient high-tech systems to sort through and organize that. Just hazarding, I guess, at what might be happening. Got it. Check. And then when we think about probably, this is somewhat high level, but when we think about the...

Megan Lorenzen:
You said it.

Megan Lorenzen:
Mm-hmm.

Molly Wood:
Sustainability problems that AI is potentially uniquely suited to tackle. Like what is the view from Salesforce on what it could really do well in this arena, because it's attempting to sort of say like, Hey, let's, it's just going to like solve climate change and, you know, and then I'm always like, cool, how.

Megan Lorenzen:
Yep, yep. It has so much potential. And when we have already, we've already started on this journey with our ventures team and our Salesforce accelerator, we just recently completed an AI accelerator for climate. We're investing in a lot of startups and ecopreneurs who are really trying to leverage AI to benefit the planet. 

So whether it's a company like Pano AI who automatically detects wildfires with AI before they spread, or a company like Watt-Time, who's working on elevating the emissions impact of renewable energy projects around the grid, and really trying to access that small to medium market that they don't have the resources to access today. How can they use an AI assistant to reach more companies? There's so much potential.

Megan Lorenzen:
The AI Accelerator, which we just announced this week, actually identified five companies in particular as the winners that we felt were doing really amazing work on using AI for some of these climate outcomes.

Megan Lorenzen:
And in particular, a lot of nonprofits are feeling like they see the digital transformation as essential, but they don't necessarily have the means to really embrace that opportunity from technologies like AI. And so the accelerator was a really innovative model that is bringing forth unrestricted funding, technology donations and pro bono expertise from Salesforce to help those nonprofits maximize their impact. And so it's this really amazing full circle engagement. 

And I can brag about it because it's our wonderful philanthropy team. It's not my project, unfortunately, but I'm just so excited every time I see the work that comes out of that team. And so they identified several companies. One I already mentioned I might just elaborate on slightly, which was Wattime. And so we actually have worked with WattTime on a number of projects over the years on our own renewable energy journey. And they think really deeply on how do you strategically site wind and solar? And they've worked very closely with a number of large companies, just given the size of their organization. 

But there's this huge long tail of small and medium-sized businesses that are also embarking on this journey. And so how can they augment their very limited capacity as an NGO by providing self-service access, tutorials, FAQs with their AI assistant to help serve that broader market. And so really excited to see how a tool like that, which I have personally had a lot of benefit from, can reach a broader audience with AI.

Molly Wood:
Yeah. And then we should clarify, you mentioned it's coming from the philanthropies tool. These are all nonprofits that were funded through, sorry. You mentioned this is coming from the philanthropy side, which is funding nonprofits, which is somewhat not what you think of when you hear accelerator necessarily. Yeah.

Megan Lorenzen:
Yes.

Megan Lorenzen:
Yes, that's a very important clarification. So another company is Good 360. And so this is really focused on responding to disasters. So they truck, I think it's in the order of magnitude, around 16,000 truckloads of donated goods nationwide. And so it's this amazing network. But they're trying to figure out how can they minimize their environmental impact. 

And so they're using an AI resource to analyze their product data, to minimize delivery distance, fuel consumption, and just really optimizing their systems altogether to decrease waste, increase impact, and make sure that they're delivering the right resources to the right places. And they're expecting to see almost a 20% reduction in carbon emissions just through using AI to optimize their logistics. And so there's just...

Molly Wood:
Wow.

Megan Lorenzen:
So many different possible applications here.

Molly Wood:
I love it. And then Megan, tell me about your background. How did you end up with Dream Job?

Megan Lorenzen:
Yes, so I actually, I started my undergraduate in political science and then discovered that environmental studies was a potential pathway and I was like, wait, you mean I can pursue a career and a curriculum around something I'm personally very passionate about. I spend most of my time trying to get outdoors and enjoy nature and that really set me on this path. At the time, you couldn't take an environmental degree as a standalone major. It was still one that had to accompany a more legitimate degree. 

And so it was really, I think, early on there and just had the fortune of getting my foot in the door with some utility work. So working on water and clean energy through local utilities, and then really turn the corner with a company called Three Degrees, which is a renewable energy consulting firm. I like to say I grew up there. I had a lot of opportunity to learn fast and...

Molly Wood:
Mm-hmm.

Megan Lorenzen:
Dabble in a lot of different initiatives in the clean energy space before really turning the corner to Salesforce because at Three Degrees, I worked a lot with utilities and there's such an, it's such an important role to have the enabling environment with the utility. Are they offering the right options to their customers? 

But I consistently saw that corporations could come in and just change the dynamic because you're bringing such scale to the conversation. And so really saw the opportunity to engage in the corporate climate movement, because that's where so much innovation is taking place. You're able to operate outside of some of the regulatory bounds that utilities are constrained by and have found Salesforce in particular to be a company that is really lives and breathes sustainability as a core value. 

Mark, our CEO, is so personally passionate about this issue. I think he's one of the few CEOs I can think of who can probably out-talk me in terms of stats and really the deep science of climate, which is amazing to see as a leader. And then that leadership has trickled down throughout the org where business partners throughout Salesforce really understand that this is a core part of how we operate as a company. Again, it's embedding sustainability. It's embedding equity. These are our values and they really do show up in the decisions we make as a business.

Molly Wood:
Yeah, I want to put a fine point on this because I think it is easy for, you know, when people talk about climate action, they sort of simultaneously point to the need for corporations to have a big role and also don't believe them when they try to take that big role. And I think it's really, it's important to sort of say that people are choosing to work for companies that reflect their values and that if Salesforce or Google or Walmart, you know, makes big changes in the industry, that's how change happens.

Megan Lorenzen:
And if you think about it, just from the perspective of the clean energy, again, that's my bread and butter where I'm most comfortable. A decade plus ago, there were only a handful of companies who were thinking about procuring clean energy. And now you have hundreds who are part of this community and on this journey.

Megan Lorenzen:
And I think what's amazing is together we've catalyzed over 150 gigawatts of new renewable energy.

Megan Lorenzen:
But just the scale of impact that these corporates have been able to make to really take wind and solar from an emerging technology to now common practice and in many cases the least cost option in a market. It's amazing to see what change can happen when that focus, that unified focus of corporates.

Molly Wood Voice-Over:
Thanks to Megan and again, startups, individual action, policy, investment, we need all of it. We need companies to move and to encourage each other to move, whether it’s out of genuine care or simple economics or ideally both.

When I talk to company executives about say, a lack of regulation, or what will it mean if the political winds change and there’s an anti-climate administration, you know what a lot of them say? Something along the lines of, "Well look, once Walmart, back in 2017, announced it was going to try to decarbonize its supply chain, we all knew that’s where business was headed and you can’t stop that train."

By the way, back then, Walmart set a goal to reduce, avoid, or offset a billion tons of greenhouse gas emissions from its supply chain and achieved the goal this year, six years ahead of schedule.

You can’t stop this train.

That's it for this episode of Everybody in the Pool. Thank you so much for listening.

Email me your thoughts and suggestions to in at everybody in the pool dot com and find all the latest episodes and more at everybody in the pool dot com, the website. And if you want to become a subscriber and get an ad-free version of the show, hit the link in the description in your podcast app of choice.

Thank you to those of you who already have. See you next week.