Jon Krohn: 00:00
This is episode number 459 with Vince Petaccio II, data scientist and climate advocate.
Jon Krohn: 00:12
Welcome to the SuperDataScience Podcast. My name is Jon Krohn, a chief data scientist and bestselling author on deep learning. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. Thanks for being here today, and now let’s make the complex simple.
Jon Krohn: 00:42
Welcome back to the SuperDataScience Podcast, I am your host Jon Krohn. We’re very lucky to be joined today by the engaging and fact-filled Vince Petaccio II. By day, Vince is a data scientist at Amazon, or more specifically, for the gigantic and hugely profitable AWS Amazon Web Services cloud platform. However, that is not the topic of today’s podcast. Today, we’re focusing on green machine learning. Based on his experience as a climate advocate, Vince will be filling us in on specific ways that data science can be used to fight climate change, as well as guidance on how you yourself can take action and have a big impact.
Jon Krohn: 01:24
During today’s episode, Vince and I do delve into some technical aspects of machine learning and data science tools for a few minutes here and there, but by and large today’s episode should be appealing to anyone whose interested in knowing how data and technology can be leveraged to rein in climate change. All right, tons of practical information for you in this episode, let’s jump right in.
Jon Krohn: 01:54
Vince, welcome to the show. It’s such an honor to have you on.
Vince Petaccio: 01:59
Jon, it’s a pleasure to be here today and can’t wait to really dig into everything with you today. It’s always a pleasure to speak with you.
Jon Krohn: 02:05
Yeah, this is such an important topic, we will get to all of that. But first, tell us where in the world you are. I think I know, I have a feeling that you are in New York, in the Bronx, New York. Am I right?
Vince Petaccio: 02:19
The beans have been spilled. I am in the Bronx, here in New York City. That’s right, in the Boogie Down.
Jon Krohn: 02:25
Nice. To the listeners who have listened to this podcast many times, they probably know that I live in Manhattan, we don’t live very far away from each other. And, we actually recently, we went to the New York Botanical Garden together.
Vince Petaccio: 02:42
That’s right, right here in beautiful Bronx.
Jon Krohn: 02:46
We have met many times before. Vince and I used to work together at a company called untapt where I still work. We also, in 2020, started a podcast together called A4N, the Artificial Neural Network News Network, which was a fun podcast. It was a news show on AI, machine learning news, it was intended to be very lighthearted. We only made four episodes, but in that time, our second guest on the podcast ever was one Kirill Eremenko. And, I suspect that Kirill’s experience on the A4N podcast is related to why you’re now stuck, everyone listening to me right now, on the Super Data Science Podcast, listening to me. I hope that at least some of you are enjoying that transition.
Jon Krohn: 03:42
Anyway, Vince is very experienced with podcasts and I wanted to bring him on because he knows a ton about the climate, and how machine learning and data science can be used to help with the climate. Let’s dig into this green machine learning topic. I know that there are some applications of data science and machine learning to tackle climate change, and I really wanted to have that be a focus of an episode. But, you are the person who is most knowledgeable, by far, about that topic so I wanted to have you on the show and provide I know it’s going to be a wealth of knowledge on this topic.
Jon Krohn: 04:33
Before we do that, how long have you been involved in climate science, or learning about it? And, what do you do today in that space?
Vince Petaccio: 04:44
Yeah. It’s hard to identify a time when it really started for me, but when I think about that, I have been asked that in the past, the moment that really sticks out for me is actually when I was in first grade. I went to a kindergarten where we celebrated Earth Day for entire week. This involved activities like watching a caterpillar form a chrysalis and turn into a butterfly, everything to just studying the trees out in our playground, and just in general just being very observant of the natural surroundings that we enjoyed. And really paying attention to them, and how we interacted with them, and how we shared a space on the plant with them rather than thinking of them as just a product that we could consume, or a resource that we could consume for our own benefit.
Jon Krohn: 05:34
I love to eat caterpillars.
Vince Petaccio: 05:36
Oh, so delicious. So delicious.
Jon Krohn: 05:38
Right as soon as they make the chrysalis, it’s so crunchy.
Vince Petaccio: 05:40
I’m a big fan of the legs, myself. They have all these extra ones, once they turn into butterflies. I actually don’t even know if that’s true. Where do they go?
Vince Petaccio: 05:55
In general, I think that laid the foundation for my thinking about our interaction with and engagement with the natural world.
Jon Krohn: 06:03
But you said in the first grade, and then you … Did something happen in the first grade that you were … This whole story was kindergarten.
Vince Petaccio: 06:14
Oh yeah, it was first grade. I’m sorry.
Jon Krohn: 06:15
Oh, it was the first grade.
Vince Petaccio: 06:16
It was the same place I went for kindergarten.
Jon Krohn: 06:19
Oh. I thought the story was going to go to something like, “In kindergarten, we had Earth Day for a full week, and we really observed Earth. And then when I got to the first grade, we just studied math the whole time, through the entire Earth Week and we didn’t even acknowledge Earth Day, and that’s when I realized I needed to be a climate advocate because people weren’t listening. We were just doing math.” I don’t know, something like that, I thought that was going to happen and I don’t know why. Anyways, you meant the first grade. Okay, cool.
Vince Petaccio: 06:48
Yeah. And then, when I was in 10th grade actually, we had a park in my town, still have a park in that town, to an extent. That park was previously underwater in the Delaware River in Pennsylvania, and it was formed as a park when, I believe in the early 1900s, a creek, a canal was formed off the Delaware River and that channeled water away from this low-lying land later emerged. There was a series of locks that would prevent that canal from overflowing its banks. When I was in 10th grade, not first grade, some flooding occurred which caused the locks to break and all that water flooded our town’s park.
Vince Petaccio: 07:38
This park was the hub of social activity for young people in my town. It had a swimming pool, and a giant wooden playground for the kids. It was under, I want to say eight to 10 feet of water, which decimated this shared public space for entire town. It collapsed the swimming pools, destroyed all of the dugouts for little league. And that was a moment when I realized, “Wow, this is a very tangible example of how a changing climate can really destroy resources that are vital to a community.” From then on, going into college and being involved with climate activism in my university, at Drexel University in Philadelphia, I always carried that passion with me.
Vince Petaccio: 08:25
After school, I actually had a career for about six years work in surgery, in brain and spine surgery, and was really interested in machine learning and AI, just from reading pop science books mostly. I thought, “Wow, this seems like a really powerful tool, and I want to study this and learn more about this, and find ways to apply this to the climate problem.” That’s what eventually led me into the space, and a career change, and led me into your arms.
Jon Krohn: 08:50
Yeah. Your first data science job was with us at untapt. Yeah, you’re a brilliant data scientist so we are very grateful to have had that time with you. And, it’s great to hear some of the background on your inspiration behind your climate work. I didn’t know a lot of that backstory before, I just knew that you were very passionate about the climate.
Jon Krohn: 09:12
Today, on top of your day job as a data scientist at AWS, you moonlight as it were, as a climate advocate for the Citizen’s Climate Lobby, right?
Vince Petaccio: 09:24
Yeah.
Jon Krohn: 09:25
What does that mean? What is the Citizen’s Climate Lobby, and what does a climate advocate do?
Vince Petaccio: 09:29
Yeah. The Citizen’s Climate Lobby is an organization, it’s a volunteer, non-partisan lobbying group. We are a non-partisan group of just citizens in the United States and abroad, and we lobby state, local and Federal officials for a particular policy, namely a carbon fee and dividend. We are really advancing a particular policy proposal, where we’d like to see carbon priced in the economy in a way that is equitable for everybody who participates in the economy and that’s where the dividend part comes in.
Vince Petaccio: 10:10
As a climate advocate with Citizen’s Climate Lobby, what we do is we spend a lot of time building relationships with community leaders in our communities, and with elected officials, so that we can advocate for this policy and really build the political will to take the action that we feel will be a very durable climate policy. There are all kinds of climate policies you could advance, some of them are more politically divisive than others and we feel that this particular policy is sticky, as you might say, because it’s really consistent with the views of most folks, regardless of their political leanings.
Jon Krohn: 10:50
Nice. So we’re talking about having a price on carbon, right?
Vince Petaccio: 10:53
Yeah.
Jon Krohn: 10:53
If you emit a ton of carbon, you have to pay a certain amount for emitting that carbon. And typically, I think the way that these schemes work in places where they have been implemented, there’s typically a scale over time. There’s a floor on the price that typically starts low, so that provides heavy emitters, like factories, to start coming up with solutions at that low price knowing that, in the future, the floor in these carbon credits is going to go up, so better be innovating to avoid unnecessary carbon release.
Vince Petaccio: 11:32
Yeah, exactly. Exactly. The whole idea is that, right now we have this major externality that is totally unpriced in our economy, which is carbon emissions. The idea is to start pricing that externality, because it has genuine costs, both economic and otherwise, for everybody on the planet. This is an attempt to correct a market failure.
Vince Petaccio: 12:03
This particular policy prices the carbon at the point where it enters the economy, so at the coal mine, at the oil well, at the port when the oil is imported. The assumption is that those costs will be passed on to consumers, so 100% of those fees that are collected are returned as a flat payment to everybody with a social security number in the United States. In some ways, you can think of this as being redistributive or progressive, and in fact the bottom three Quintiles of income in the United States would actually come out ahead. They would actually profit from something like this. That’s a result of the fact that those upper two Quintiles are very disproportionately more likely to consume more and to generate more emissions.
Jon Krohn: 12:59
Great. That part, I am less familiar with. The climate price, that’s something that I’m used to reading about a lot. I read The Economist every week, and they are huge fans of carbon pricing as a solution, as one solution in climate change. Or, for avoiding climate change. But, I didn’t know much about this dividend, so that’s very interesting to hear. I’d never heard that from you before. Cool to know about that.
Jon Krohn: 13:27
But now, let’s jump to the topic that I think our listeners really want to hear about, which is how machine learning and data science can be used to tackle climate change, alongside policy initiatives like carbon pricing schemes.
Vince Petaccio: 13:43
Yeah. Before I get into any specifics, I just want to make really clear that AI and machine learning, both extremely powerful tools in this battle, but ultimately climate change is such a big problem that it’s going to take all hands on deck. And, anything that we do in the AI space will need to be in collaboration with subject matter experts in other domains. That allows us to accelerate the pace of advancement and development of new capabilities, within those existing domains, so that they can be more readily capable of dealing with and tackling climate change.
Vince Petaccio: 14:28
Some of those domains where we can apply them might seem somewhat obvious, places like transportation, electrification of the transportation system, or reducing transportation activity by optimizing routes and things like that. But, some of them are actually pretty surprising, at least to me, some of the ways that AI could be applied. Just to list some of the places we could be applying AI and machine learning are things like electricity systems. We could be enabling low carbon electricity by optimizing solar generation or wind generation, and power delivery using predictive methods, using hyper local weather predictions, for example. Buildings in cities, optimizing buildings by controlling their HVAC and lighting systems in a very predictive way, or an optimized way, is another great example. Google has done amazing work in this field, optimizing the energy usage of their data centers using some AI and machine learning techniques.
Jon Krohn: 15:37
Some of the research out of the DeepMind team at Google was helpful.
Vince Petaccio: 15:42
Yeah, exactly.
Jon Krohn: 15:42
Some deep reinforcement learning involved in saving some money in those costs in the centers. I didn’t realize if that also had … Of course, it does. I’ve only ever thought about that as something that saves them money, I was thinking about how that’s also going for the climate, to obviously be emitting less.
Jon Krohn: 16:00
Hey everybody, hope you’re enjoying this amazing episode. We’ve got a quick announcement, and then we’ll get straight back to it. The announcement is that DataScienceGO Virtual number three is approaching quickly, it’s happening on April 10th to 11th, and you can get your free tickets today at datasciencego.com/virtual. We’ve got incredible speakers, hands-on workshops, and an expo area that you can virtually attend. And of course, we’ve also brought back one of the most popular parts of DSGO Virtual, the networking sessions. These sessions are the best way to become part of our global data science community. Over the course of the conference, there will be several three minute speed networking sessions in which you connect with a randomly selected data scientist from anywhere in the world. After the three minutes, if you like each other and you’d like to remain connected, you hit the connect button and you can stay in touch. Once again, every aspect of the DataScienceGO conference is absolutely free. Register for your ticket today at datasciencego.com/virtual and we’ll see you there. And now, let’s get back to the episode.
Vince Petaccio: 17:06
I think it’s very common for us to think of saving money and saving the planet as being mutually exclusive, but it many cases those two ideas are completely consistent with one another and totally compatible with one another.
Vince Petaccio: 17:22
A great example of this is things like supply chain optimization. How can you optimally pack the most packages into a single delivery truck and take the shortest possible route to deliver all of those packages, and that’s a great application of AI. Not only does that minimize the emissions that that delivery truck will generate, it also reduces the cost associated with operating it. There are tons of examples that we see with this.
Vince Petaccio: 17:51
Another example is with precision agriculture. It’s important, as the human population continues to grow, that we’re able to produce enough food to feed everybody around the world. And, the field of agriculture is already incredibly sophisticated with its automation, and with its ability to scale technology in order to meet the demands of a growing population. But, to be able to continue to do that and continue to push that envelope, AI is a great and very powerful to do that. Using computer vision to help identify sickness in plants, or be able to extract plants from the ground to reduce manual labor, would allow you to scale your farming operations.
Vince Petaccio: 18:38
And in a similar vain, vertical farming is another place where there’s a lot of opportunity for us to leverage automation and AI. There’s a lot of work to do there, and a lot of that is not necessarily related to AI or machine learning, but that’s just another example of where cost and environmental responsibility are completely consistent with one another.
Jon Krohn: 19:00
Those are all super cool examples. I am fascinated by ideas around vertical farming and precision agriculture, I think it’s really cool to think that, even in urban areas, you could have so much grown in the urban area in unused buildings, or on the roofs of buildings.
Jon Krohn: 19:22
When I first heard about how machine learning was being applied to vertical farming and agriculture from you, I went and found a documentary on it and I was blown away. I was like, “Oh, this is some totally cool new thing! I wonder if this is completely uncharted territory,” and discovered that there is a huge business already existing in the New York area. So New Jersey, Brooklyn, there are huge vertical and indoor farming ops, grow ops.
Vince Petaccio: 19:56
I think that’s something different.
Jon Krohn: 20:01
They’re growing plants.
Vince Petaccio: 20:02
Yeah, you’re right. You’re not wrong, you’re not wrong. Yeah, it’s already an industry that’s experiencing explosive growth right now, and I think it’s fantastic.
Vince Petaccio: 20:12
There’s even a startup in California that is trying to 100% automate the process. Not just the growing process, but also the harvesting process and the packaging process, so that you could imagine just a giant warehouse where no one even works in there, but you just are able to receive produce from it. It’s a giant produce dispensing machine. I think there’s a long way to go still. Vertical farming right now has been very successful.
Jon Krohn: 20:38
I’d be surprised if you told me, “They’re almost done.”
Vince Petaccio: 20:43
They’re opening next week.
Jon Krohn: 20:43
Pretty much finished, yeah.
Vince Petaccio: 20:46
No. But, I think vertical farming, right now it’s sophisticated at producing micro greens, and premium lettuces at a premium cost. I think that the next step for vertical farming will be, “Okay, how can we scale this technology both in size, to make it accessible to more people and more places? But also, how do we expand the types of foods that we can grow here?” Because something like lettuce is going to be a much easier challenge to tackle in a warehouse than growing a row crop, like corn.
Jon Krohn: 21:22
Right.
Vince Petaccio: 21:24
That’s the frontier for vertical farming, and I’m really excited to see where that goes.
Jon Krohn: 21:31
Nice. I don’t want to spend too much time on vertical farming because I’m sure there’s tons of topics to cover. But, why is it more expensive?
Vince Petaccio: 21:38
That’s a good question. I think the major expensive input for vertical farming is energy. Lighting is generally going to be more expensive than sunlight, and that is also the main point where advances are continuing to drive the costs down to be competitive with traditional agriculture. As we see more efficient LED lighting and more research around the exact bandwidths of light that a plant needs to grow, and the exact amount of light energy that a plant needs to grow, we’re seeing further and further improvements in the efficiency of lighting.
Jon Krohn: 22:17
Nice. And, that’s a potential predictive modeling opportunity right there.
Vince Petaccio: 22:20
Absolutely, absolutely.
Jon Krohn: 22:23
Cool. All right, so vertical farming, optimizing energy delivery. What else have we got? What else can we do with machine learning?
Vince Petaccio: 22:30
Yeah. This is a scenario that I think is interesting, is just social impact. A great example of this that I love is Dr. John Cook, who … Many folks have probably heard the stat of 95% of scientists believe that climate change is real and a big problem. That’s a bit of an outdated statistic at this point, but when that data point started actually being publicized, they were quoting a paper by John Cook. He has since-
Jon Krohn: 22:58
Presumably even more scientists … Yeah, right.
Vince Petaccio: 23:03
Correct. Correct, yeah. Thank you.
Vince Petaccio: 23:06
But, I had an opportunity to meet him, actually at a Citizen’s Climate Lobby conference in DC. And I told him, “Hey, I’m really interested in machine learning and AI,” this was before I worked in the field. I said, “How can I make an impact in this space?” And he told me about a project he was working on called The Cards System, C-A-R-D-S. Basically, it was a natural language processing model that could ingest articles, or tweets, and that could classify them in terms of how likely they were to be misinformation about climate change. The idea here is how can we use AI to identify falsified or bad information about climate change, so that we can raise public awareness and give people correct information in an automated way.
Vince Petaccio: 23:53
There are a lot of other social impact opportunities for applying AI. Things like managing infrastructure is just a fantastic opportunity for applying AI and machine learning, that would have an enormous impact on societies everywhere.
Jon Krohn: 24:14
Nice. That is a cool one. It had never occurred to me in the past that natural language processing could also be used in climate change. There isn’t an obvious link there, but now that you mention, that is yeah, identifying misinformation is definitely a good use of NLP.
Vince Petaccio: 24:37
Yeah. There are a lot of surprising ways that AI and machine learning can be applied here, and perhaps some that are not so surprising. I think a good example there is in climate modeling. We have a ton of existing models, that some use machine learning and some don’t, to try to predict what the climate will look like depending on different input parameters. Generally, what we do is we build a big ensemble of all of those models and try to use those as a way to form some sort of consensus over time. But, being able to ingest larger and larger amounts of more hyper local data to get better predictions over time would really help us with adaptation, which is one part of dealing with climate change that’s a little different than mitigation. But, being able to predict what the climate will look like in the future will play a huge role in letting us adapt to it.
Vince Petaccio: 25:32
A lot of that also could be related to forecasting extreme events. How much more likely are we to experience super storms in the future? So getting good climate prediction models would be a great resource there, in helping us predict that and really be prepared for them in the future.
Jon Krohn: 25:48
Yeah. How high do we need to build the flood walls around Manhattan?
Vince Petaccio: 25:52
Exactly.
Jon Krohn: 25:53
Or New Orleans, whatever.
Vince Petaccio: 25:55
Yeah. And, one of the specific advancements in that area that I’m very excited about is what I would call colloquially physics layers in neural networks, which basically constraint the parameter search space that you’re exploring when you’re trying to train a neural network, to that region which is consistent with the laws of physics. Rather than having an infinite search space over your parameter space, you have a smaller infinite search space where the laws of physics are respected. And so, there so really promising results coming out of that area now, that I think will have a big impact in this field.
Jon Krohn: 26:36
All right, I want to talk about that for a second. It isn’t directly related to climate change, but I’m unfamiliar with this. Do you mean that the outputs of the model are physically possible? Or, something to do with the model weights respecting the laws of physics?
Vince Petaccio: 26:59
The former, mostly. There are a number of different techniques for this.
Jon Krohn: 27:03
Phew. I was like, “Oh man, what does that even …” My mind just bent there. I was trying to think of neural network weights could behave in the laws of physics and I was like, “Man.”
Vince Petaccio: 27:16
They tend to be pretty good about that. I think the simplest version of this is something like regularization or a penalty applied to your loss function if the result of your model’s output during a training run violates the laws of physics, so you heavily penalize the model for making predictions that are impossible, basically.
Jon Krohn: 27:44
Nice. It could never be that hot.
Vince Petaccio: 27:47
Exactly.
Jon Krohn: 27:50
Oh, geez. All right, so climate modeling, you’re right, I guess that is a relatively obvious use of machine learning in the climate space, but that’s a very cool innovation nonetheless. You got anything else for us on your list there?
Vince Petaccio: 28:04
Yeah. This is something that’s near and dear to my heart. As much as climate change is definitely a systemic, global problem, and we need action on a social and societal level, personally I believe that individual action is also very important because many individuals taking individual action result in society level action. So I think that we could really help people make good decisions in their personal lives by giving them better information to inform their decision making day-to-day.
Vince Petaccio: 28:39
An example of that is simply finding a way to help us make better consumption choices day-to-day, in terms of environmental impact. If I go to the grocery store today and I pick up an apple, or I pick up a steak, there’s nothing there telling me what the environmental footprint might be to consuming one of those.
Jon Krohn: 29:04
Oh cool, I hadn’t thought of that before. So it would be like nutrition information. In New York City for a while now, it has been required that if you’re a restaurant of a certain size, you have to publish the calorie count next to any food item. So you could have the same thing for carbon emissions or something, right?
Vince Petaccio: 29:22
Exactly. It doesn’t need to be perfect, but having some sense that this Macintosh apple, two-thirds of the crop is destroyed on its way to the grocery store, versus this Fuji apple which is much more efficient. Just having some qualitative way to compare these two might make me choose one apple versus the other, which is a great way to tie financial incentives to environmental outcomes. But also, just helping me as a consumer choose between an apple or a steak, or a steak and tofu, if I’m into tofu. Just understanding what the actual effects are on the planet for my consumption decisions, and just with food, with anything. We could be using AI to model these phenomena, and just get a qualitative or quantitative understanding of what the impacts are of our consumption choices.
Jon Krohn: 30:16
Nice, that’s cool. I think that brands, some aware brands, particularly in areas that might have questionable green standards … So for example, something that might surprise listeners, and you might actually know this stat much more than me, Vince, but almonds use up a crazy amount of water. So there’s a lot of almond farming in California, and California has a lot of droughts, and almond farmers take a lot of the blame for those droughts. I noticed, just this week, that on my particular brand of almond milk it says right on the bottle, there’s one entire side panel of the almond milk container says that they somehow offset all of that water use.
Vince Petaccio: 31:03
Wow.
Jon Krohn: 31:03
That’s kind of interesting.
Vince Petaccio: 31:03
How?
Jon Krohn: 31:05
Yeah, that was my next question.
Vince Petaccio: 31:09
Well good for them, that’s great. But, wow.
Jon Krohn: 31:14
They ship in a bunch of water bottles, into California.
Vince Petaccio: 31:19
Just industrial scale dehumidifiers, capturing water from the air.
Jon Krohn: 31:23
Exactly. I don’t know, that’s a good question. They use cold fusion to have hydrogen and oxygen invade the air, combine together to form water. They haven’t thought of any other better uses of that cold fusion system yet.
Vince Petaccio: 31:37
There are none, there are none.
Jon Krohn: 31:39
Yeah, that’ll do it. Offsetting almond farming. Cool.
Jon Krohn: 31:45
All right, those are all awesome use cases of machine learning, data science and AI for climate space. But, are there any dangers or risks associated with using tools, do we need to be careful? I think coming up next in the show, we’re going to provide people with tips on how they, as a data scientist or as just a concerned citizen, or what have you, how they can be involved in the fight against climate change, maybe how they could even use data science and machine learning to make an impact. But, before we get to that, is there anything that they could think about but they do?
Vince Petaccio: 32:25
Definitely. Ultimately Jon, climate change is a global problem that has local impact, that disproportionately affects disenfranchised and vulnerable communities. Any strategies that we want to use to address climate change need to be very careful to avoid reinforcing existing inequities, and perhaps should even focus on targeting vulnerable people and places. Because otherwise, we could end up in a situation where we’re creating solutions for the areas that might not need them, and avoiding creating solutions for places and people that need them most. AI and machine learning, like any other tool in our toolkit, are tools that we can and should wield for solving this problem, but it’s important that we’re thoughtful about collaborating with the subject matter domain experts that I talked about earlier, so that we can make sure that we’re having a big impact.
Vince Petaccio: 33:33
Some of the risks that are associated with that come down to things like making sure we’re not using too complex of a tool for what might actually be a simple problem. Just as an extreme example, imagine if I wanted to train GPT3 in order to perform sentiment analysis on tweets about climate change, and just imagine the carbon emissions associated the training a model of that size. What am I actually getting from that, when I could have just used a logistic regression model over some word embeddings, or something like that.
Jon Krohn: 34:06
Yeah. In case you aren’t aware listener, GPT3 is an enormous natural language model that has hundreds of billions of parameters. Yeah, so the carbon footprint is actually something that it’s a contentious issue. I’ve been thinking about having this be a specific topic on the podcast, I should do that sometime soon. But, there’s been a huge paper …
Jon Krohn: 34:31
At the time of recording, we’re recording in early March. Just a few weeks ago, there was a big paper released called Stochastic Parrots. It’s a paper from researchers at the University of Washington, and controversially former as well as current Google employees, talking about some of the negative aspects of these giant language models. So GPG3 has been hailed as a huge achievement in the natural language space. And haven’t read all the paper, but I know the idea of the Stochastic Parrots title is implying how parrots don’t really have an understanding of natural language, but they can squawk some things back at you that sound impressive. And, these stochastic parrots can be a parrot with a little bit of randomness in it, stochastic just means randomness, but that they’re not actually intelligent in the way that a human is intelligent.
Jon Krohn: 35:41
Anyway, that was a long winded way of saying that maybe these huge natural language models aren’t as effective as some media splashes have suggested. And on top of that, there could be negative side effects, like huge impacts on the climate. Sorry to take away from your time there, Vince.
Vince Petaccio: 36:02
No, thank you for adding to the conversation.
Jon Krohn: 36:06
We’ll [inaudible 00:36:07] your time on at the end of the podcast.
Vince Petaccio: 36:07
Okay, great. To your point, it’s just a giant hammer for what might end up being, in some cases, a small nail. Or inversely, there’s a risk in the inverse which is that, as a computer scientist, we might have an oversimplified understanding of a particularly nuanced or complicated problem, and our attempt to solve it without collaborating with domain experts could cause more harm than good, in some way or another, in some nuanced way or through some downstream impact. Just in general, the failure to be able to predict unforeseen risks is a major risk in this kind of work, when you’re not collaborating.
Vince Petaccio: 36:56
Basically, the theme of everything I’m saying here is that collaboration is really key. Listening to people in other domains, recognizing that, as computer scientists working in AI and machine learning, we have a particular domain of knowledge that is powerful at accelerating other domains will be critical to making a big impact and making a positive impact.
Jon Krohn: 37:22
Nice. So understood, machine learning is only a tool. We do need to be careful how we wield it, no doubt. How can people wield machine learning to make a difference themselves? How can our listeners make an impact?
Vince Petaccio: 37:39
Yeah. I think ultimately, a lot of us who want to be involved in using the skills and the tools that we have to be helping with this. And, coming off of what I said a moment ago, I think the best way to do this is to collaborate with folks from other domains, and finding ways to do that is key. I would recommend the folks who are interested in this check out Climate Change AI, which is an organization started by Priya Dante, who is a PhD student at Carnegie Mellon University. That is an organization that is historically oriented towards researchers in the field, but is rapidly expanding to encompass and include folks from industry as well. There are a lot of opportunities there to network, and connect with collaborators, and find ways to have a great impact. And, to just learn more about where the need is.
Vince Petaccio: 38:39
Speaking about that, about Climate Change AI and Priya Dante, she was one of the key authors behind a seminal work in this space called Tackling Climate Change With Machine Learning. It is 100 plus page paper.
Jon Krohn: 38:53
Oh, man.
Vince Petaccio: 38:56
Yeah.
Jon Krohn: 38:56
I hadn’t made that connection.
Vince Petaccio: 38:57
Yeah, it’s all coming together.
Jon Krohn: 38:59
There is so many names on that paper.
Vince Petaccio: 39:02
I know, I know, and some big names, too. Some household machine learning [crosstalk 00:39:07] names.
Jon Krohn: 39:07
Yoshua Bengio’s on there, right?
Vince Petaccio: 39:09
Yeah. Yes, he is. Andrew Ng. Yeah, that organization is open to folks who are interested in this space and finding ways to collaborate with others. But definitely, check out that paper. It’s long, at over 100 pages, and that’s because it has an enormous amount of extremely valuable information on everything we talked about today. We touched on a few areas where machine learning and AI can have an impact, and this paper is just basically the bible or the encyclopedia of possible avenues to explore. There are even business opportunities that can be made out almost any single one of these that we’re interested in.
Vince Petaccio: 39:53
In fact, there is a website that I would recommend your listeners to check out called The Diamond List, thediamondlist.co, which basically is a list of companies that are making significant climate impact. That’s a great place to check out, what activity is happening right now in the industry that’s focused on having a positive impact.
Vince Petaccio: 40:20
Just in terms of private corporate action, there’s also a website called climatevoice.org, that is a great place for those who are in industry to find ways to come together and to advocate for climate action and climate policy together.
Jon Krohn: 40:43
Nice. Great choice of final word.
Vince Petaccio: 40:46
Thank you.
Jon Krohn: 40:49
That’s brilliant. We’ve got Climate Change AI, diamondlist.co, climatevoice.org. I love it Vince, thank you for those straightforward tips for how people could wield ML themselves. In addition to the paper that you just cited, which is the encyclopedia for wanting to apply machine learning to tackling climate change, do you have any other books that you recommend that people should read?
Vince Petaccio: 41:18
Yeah. Personally, I like to take a very pragmatic approach to addressing climate change. Sometimes, when we dig into the numbers and do the math, we find that our expectations are slightly violated in terms of what we think will have the biggest impact. There’s an amazing book, I have here actually, called Drawdown, let me get it in the frame. This book basically lists … It’s a comprehensive plan of ways to reverse global warming, is the subtitle of the book. Basically, this goes through dozens of different actions that we can take to try to mitigate climate change, and then ranks them quantitatively by the impact they’ll have. There are some pretty surprising outcomes here.
Vince Petaccio: 42:06
In fact, the number one most impactful issue, according to the numbers in this book, is refrigeration. This book suggests that refrigeration and modernizing the way that we cool things is perhaps the single most important climate issue. I would definitely recommend readers to check out this book, because it’s a great way to reframe our understanding of what does and does not have a big impact.
Jon Krohn: 42:35
Brilliant. We’ve learned a lot from you in this episode, tons of resources to check out, things to read on our own. But, how can we follow you or get in touch with you if we’d like to learn more from you about machine learning applied to climate change, or just climate change generally?
Vince Petaccio: 42:53
Yeah, I would definitely invite people to add me on LinkedIn, Vince Petaccio II on LinkedIn. Yeah, I’d be happy to connect with folks.
Jon Krohn: 43:02
Nice. All right, well thank you so much for being on the show Vince, and hopefully we can have you on again soon to give us an update on other big news in the green machine learning space.
Vince Petaccio: 43:14
It was absolutely my pleasure, Jon. Thank you so much for giving me the opportunity to come and talk to you today about this.
Jon Krohn: 43:20
My pleasure, catch you soon.
Jon Krohn: 43:26
Wow, Vince sure knows a ton about cutting edge data science as well as how machine learning can be used to fight climate change, doesn’t he? In today’s episode, we covered tons of green machine learning applications, including optimizing energy delivery, precision agriculture, identification of misinformation and climate modeling. We also discussed the importance of working with subject matter experts when designing climate change solutions, in order to ensure you’re maximizing global impact and avoiding the reinforcement of historical inequities. And, Vince provided us with a number of specific resources to enable us to move forward and take action ourselves, ranging from the highly technical and detailed Tackling Climate Change With Machine Learning paper, to organizations we can get involved with, like Climate Change AI, diamondlist.co and climatevoice.org.
Jon Krohn: 44:19
As always, you can get all the show notes including the transcript for this episode, any materials mentioned on the show, and URLs for Vince’s LinkedIn profile at www.superdatascience.com/459. That’s www.superdatascience.com/459. If you enjoyed this episode, I’d of course greatly appreciate it if you left a review on your favorite podcasting app or on YouTube, where we have a high-fidelity smiley face filled video version of this episode. I also encourage you to follow or tag me in a post on LinkedIn or Twitter, where my Twitter handle is @jonkrohnlearns, to let me know your thoughts on this episode. I’d love to respond to your comments or questions in public and get a conversation going. You’re also welcome to add me on LinkedIn, but it might be a good idea to mention you were listening to the SuperDataScience Podcast so that I know you’re not a random salesperson.
Jon Krohn: 45:08
Since this podcast is free, if you’d like a hugely helpful way to show your support for my work then I’d be very grateful indeed if you made your way to the Data Community Content Creator Awards nomination form, the link is in the show notes. Of course, we’d love you to nominate the SuperDataScience Podcast for category seven, the podcast or talk show category. I’d also love my name, Jon Krohn, nominated for category eight, the text book category, for my book Deep Learning Illustrated. And finally, I’d also love my name, again Jon Krohn, nominated for category two, the machine learning and AI YouTube category, for my YouTube channel which contains tons of free videos on deep learning, linear algebra applications and machine learning libraries.
Jon Krohn: 45:54
All right, thanks to Ivana, Jaime, Mario and JP on the SuperDataScience team for managing and producing another great episode today. Keep on rocking it out there, folks, and I’m looking forward to enjoying another round of the SuperDataScience Podcast with you very soon.