Episode transcript The transcript is generated automatically by Podscribe, Sonix, Otter and other electronic transcription services.
Hi everyone. Welcome to the 5 Minutes Podcast. Recently, Antonio Nieto and I released our second article at around me. And this time, we decided to understand AI, but in a much broader context, the context of sustainability. And every time we talk about sustainability, the first thing that comes into mind of everyone is environmental sustainability. It's how you reduce CO2 emissions. And this, for me, is absolutely perfect. But sustainability goes above and beyond that. It's social, it's environmental and it's economic sustainability. This was the topic when I did my Ph.D. that I was very keen on studying and understanding that we cannot just think about sustainability from a simple and pure environmental perspective. And what we did in this article, we made an analogy with our magic triangle, in which we talked about project scope, time, and cost. And we put complexity, cost, and carbon as the three C's why we need to balance these three things. Because complexity comes from several interdisciplinary demands, regulatory requirements, and data overload, this all brings a massive complexity that I need to manage. The complexity is not only to develop but the complexity to manage on a daily basis. This brings me to the second aspect, which is the cost because managing complexity requires money. So all these direct interconnections and all these different disciplines that you need to put in place, this costs a lot of money. Life cycle shorter life cycle, a lot of volatility, and a lot of changes bring costs up.
So this is, for example, if you go on the web and talk about OpenAI and ChatGPT, you'll be surprised by the fortune and the insane amount of money they spend just to keep their product moving and working. I'm talking about millions of dollars every single day. The third aspect, it's all about the environment. And I just want to make an analogy. For example, when we think about cryptocurrency like Bitcoin, the mining of Bitcoin means the process for you to generate a new cryptocurrency is everything but sustainable. It consumes a massive amount of energy, it's extremely demanding, and it produces a massive amount of CO2. If we make the same analogy of AI, just to give you an idea, one research of the University of Massachusetts Amherst calculates the CO2 emissions just to train a large language model. This is just to train. This is not to operate. It's around 626,000 pounds. Just to give you an analogy, this is all the CO2 that a car emits during its full life cycle. I mean, it's not per day or per year. It's until it exists. And this is just to train. I'm not saying that this is what, for example, OpenAI or Google is expanding on Bart or ChatGPT, respectively. No, this is just for you to train. Just to give you an idea of a motto. We had to train several times. So on each of these, of course, our database is not as big, but probably we spend a relevant amount of this.
So this is something we need to be ultra mindful of. And we finished the article with a five-step approach. We work very hard on defining these approaches for how you can build sustainable AI applications. The first is to understand what your sustainable objective is and what your end game is not in terms of business perspective but in terms of managing the economic aspect of, for example, the solution of AI, the environmental aspect of the solution, the social and the complexity aspect of attending all the regulatory managing bias and all these human interface aspects. After that, we need to understand that data must be centralized. You need to identify ways of reducing the complexity of your data management. Because when we talk about large language models, we are talking about billions and billions, even trillions of data. So this is not a normal amount. This is an insanely large amount of data, so you need to centralize two things: to reduce complexity, to reduce energy demand, and to reduce cost. After that, you need to customize the algorithm. There are hundreds of different ways of connecting the dots. You need to identify, based on your business case, which is the best way, which is the one that reduces complexity, reduces cost, and is environmentally sustainable. For example, TensorFlow you can find in TensorFlow several examples of algorithms that can be more sustainable than the others.
And there is a whole set of research on that. On trying to identify how you can optimize these machine learning algorithms to make them more reliable and less demanding of energy and less demanding of data and, of course, money. After that, you need to start small; you need to do a pilot, and you need to test. For example, today, most of you know I'm working heavily on the MotoE project, and I know every single person asked when this will be available, when I can do that, and why it's not so simple. This answer is because we are doing a pilot, and we are testing. And many times when we test, we see, okay, it works now, but it's not long-term sustainable because the energy that is is demanding to do. For example, one request is too large. How do we optimize the database? How do we optimize the different sources? Otherwise, we may release something that will have to be turned off in a couple of months because it's just not sustainable. Maybe it's extremely expensive; maybe it's insanely complex to update. For example, when new information about projects comes up. How do I train the model? Imagine that I will produce 100,000 pounds of CO2 every time I do that. I cannot do this training every single day or every single week. Otherwise, it's just not sustainable. This is why the pilot is so important because the pilot gives you the chance to understand what is going on.
And then, last but not least, you will implement so you don't do this big bang. For example, many times, we see a lot of AI applications failing because they don't think long-term, and then they become unsustainable. You create something that is very hard to update. You create something that is very costly to maintain. For example, even OpenAI, OpenAI, there is a massive criticism. I'm saying, look, I love OpenAI, but there is a massive criticism about the financial sustainability of OpenAI, of course, with the money of Microsoft and this. But I'm talking about a business because it spends millions and millions. It produces a massive amount of CO2, and it's extremely complex to manage every single day. If someone knows how to create a more sustainable product, it will just win the game. It's not just because it's the right thing to do; it's because it's the intelligent thing to do. Think about that. Take a look at the article. It's available on HBR. We are also doing webinars to discuss this in an informal way about this aspect. So do not miss the opportunity to go a little bit deeper on that because it will make a massive difference in your project management role and also in the projects you are doing, specifically in the times of AI. I hope you enjoyed this podcast, and see you next week with another 5 Minutes Podcast.
