Our latest dialogue, “Can AI and Digitalisation Accelerate India’s Energy Transition” featuring Priya Donti, Siddharth Sareen, Simran Grover and Angelina Chamuah, critically examined AI’s role in India’s energy transition.
The discussion looked beyond technology’s promissory hype, focusing on how AI can enhance grid management and energy optimisation, while also addressing deeper issues like governance, political economy, and data integrity.
You can read an edited transcript of the dialogue below:
Angelina (00:50)
Hello and good evening. Welcome to talking about transitions and today’s dialogue on the question of, can AI and digital technologies help accelerate India’s energy transition? I’m Angelina, and I will be moderating today’s dialogue. I’m a senior research fellow at Transitions Research, and my work focuses on the intersection of AI, society and sustainability.
For those who are joining us for the first time, Transitions Research is a social science research collective, based out of Goa. We are a team of highly interdisciplinary researchers working together to examine and impact the radical transitions that are taking shape at the intersection of tech, society and sustainability.Through our “Talking About Transitions” dialogues, we hope to seed critical conversations and reflections on these radical transitions.
Today’s dialogue focuses on India’s energy transition and the role and impact of AI and digital technologies within it. India, like many other countries, is currently navigating a significant energy transition, marked by a strategic shift towards a more diverse and sustainable energy mix. While India is still heavily reliant on coal, there has been a robust focus on expanding renewable energy sources, as well as improving last-mile energy access and delivery. As part of its long-term climate strategy, India aims to achieve net-zero carbon emissions by 2070. In the shorter term, the country has set an ambitious target to reduce the emissions intensity of its GDP by 45% by 2030 compared to 2005 levels. Additionally, India plans to expand its installed renewable energy capacity to 500 gigawatts by 2030. In line with these goals, India has implemented several initiatives to support its energy transition, including programs that promote solar power plants and solar pumps in rural areas, and the National Smart Grid Mission, which aims to improve the reliability of electricity networks and make the grid more amenable to renewable energy inputs. These efforts are complemented by the creation of large-scale solar parks, green corridors, and other initiatives like the National Green Hydrogen Mission. However, India’s shift towards renewable and sustainable energy also comes with challenges, including persistent legacy issues in energy provision, last-mile access, and inefficient systems. Additionally, growing urbanisation, population growth, and industrial and economic expansion will continue to drive up energy demand and consumption in India.
Now, in the midst of it all, AI and digital technologies are also being talked about as a critical force multiplier for energy. A lot of studies have been devoted to understanding the potential of AI for sustainable energy transitions, according to one that was conducted by the World Economic Forum, digital technologies are set to potentially help reduce global carbon emissions by 15 to 20% by 25 2050 through solutions in energy manufacturing, agriculture, land use, buildings and so on. The integration of digital and AI technologies can also help improve grid management, streamline maintenance, support the integration of renewable energy sources, along with reducing waste. However, what is less clear are questions regarding the impact and challenges of implementation of these technologies. So for example, while India has made significant strides in enhancing digital connectivity and access and even last mile energy access, studies suggest that there is still a long way to go. For instance, many parts of India, especially in rural areas, still lack a reliable power supply. Approximately 70% of the pop. Relation also has limited or no access to digital technologies, and over 60% of households remain digitally illiterate. These are 2022 numbers, but recent updates show a decrease by about maybe 10% in these figures.
Now the question for us today is – how do these twin transitions, that is, the digital and energy transitions meet in the Indian context? Will the integration of AI and digital technologies create new challenges of access equity? What are the opportunities? In what ways do they interact with existing issues? And is there an opportunity here to radically reshape a sustainable energy future for a vast majority of India’s population?
In order to explore some of these questions and understand how India’s energy and digital transitions intertwine and what the future of energy in India could look like, We are joined today by a great panel of speakers. And, without any further ado, let me introduce them to you.
We are joined today by Priya Donti, who is the co-founder and chair of climate change AI, which is a global non profit that catalyses impactful work at the intersection of climate change and machine learning. She is also an assistant professor and the Silverman family career development professor in the Department of Electrical Engineering and Computer Science and the laboratory of information and decision systems at MIT, her research focuses on machine learning for forecasting, optimization and control in high renewables power grids. Thank you, Priya for joining us. It’s a bit early, but thanks.
And next we have Siddharth Sareen who’s a research professor at the Fridtjof Nansen Institute in Norway. He’s an award winning Research Professor working on the governance of energy transitions, and has worked in seven countries and established an international interdisciplinary research profile on resource and energy governance. He has experience with electricity generation, distribution, end use, urban transport, digital energy infrastructure, resource extraction, and expertise on themes such as energy poverty, just transitions, accountable governance and questions of social and environmental equity and justice. Siddharth, welcome and thanks for joining us.
And finally, we are also joined by Simran Grover, who is the founder and CEO of Center for Energy, Environment and People. CEEP is a Jaipur based human centric research and policy advocacy initiative driving critical research and fostering democratic coalitions for low carbon transition and climate justice. Simran has substantive experience in the domains of renewable energy, power sector, applied research and entrepreneurship. Currently, he leads critical work in just energy transition and power sector reforms with an emphasis on pollution building and collective action. Thank you, Simran. And with that, let us jump in.
Angelina ( 00:80)
While I did set a little bit of context, largely from a general lens, let’s begin with you, Simran, if I may, in order to set the stage for this conversation, a little bit better.
Simran, can you elaborate a little on how you and colleagues at the Center for Energy Environment and people characterise and view India’s energy transition? What, according to you, are the key focus areas, and what role do you see digital technologies and AI playing in this arena?
Simran Grover (8:41)
Thanks, Angelina. So for us, I think challenges are multiple.
There is a larger push for greening the grid, increasing renewable energy deployment, but at the same time, we are struggling with reducing our dependability on coal and coal fired assets. And there are different social and political dimensions, which tend to be non technical and isolated from grid management challenges. Right at the institutional level, I think in terms of state capabilities, we are severely struggling to manage transition risks, and to even plan for a cleaner grid with less dependability on coal, or I must say, less dependency on the firm power. So those capabilities, I think, are also some fundamental capabilities within the institutions in our experience, have also been degrading.
One of the factors is increasing outsourcing of core business functions – whether it is the management of grid substations, billing and metering, operation, and maintenance. Pretty much everything which should be central to a utility is being outsourced right now, which means that, in effect, fundamental capabilities across the utilities are on a decline. The degree of decline may vary from state to state. I largely speak from experience of Rajasthan, but from our discussions, it’s true for many of the Indian states.
The second issue, of course, is an issue of planning and data. Much of transparency, and accountability issues are there. So while AI can potentially improve, and I think my colleagues and fellow panellists here will be able to comment on it better, I am no AI expert, right? And forecasting tends to be a critical bottleneck, or, I will say, a handicap, when it comes to Indian states, particularly. For example, bulk of the projections that we see in regulatory filing are still on a CAGR basis, which is a very rudimentary methodology, which a class 10- 12 pass student can also handle. You don’t need big four consultants to push those models for energy procurement planning, etc. So for us, the more fundamental issue is the data, transparency, legitimacy and accountability on which these models and forecast projections are built upon. I think that’s a fundamental issue which needs to be addressed, without which irrespective of the efficiency and accuracy of a forecast model, be it AI or otherwise, will be redundant.
Now, various political economy issues also come into picture. How much are the actual losses across different regions? What is the trajectory of loss management? And often in our doctor experience, we do see that losses are mismanaged and misattributed in the books. Much of the losses may also be happening for industrial consumers, but are attributed to rural consumers. There are management books when it comes to agriculture sales, which basically means that when it comes to modelling, you are playing with false data, and false data will give you false models.
So some of the arenas where we believe AI can really be very powerful, and I think it can be a phased approach, if policymakers and institutional decision makers are open to it. It should create, create, number one, a layer of integration which ensures data transparency and accountability, more legitimate data coming from sub-stations. Right now, I will visit substations and their metres are not calibrated, are not functioning etc. And some of this you can also see in third party auditor reports. The second level that I think, especially in the context of declining state capabilities, AI, can possibly play a role – an assistive role, across functions of operations, as well as decision making and management, etc.
And lastly, one important issue is that of planning for the future, where modelling capabilities and accuracy will play a huge role in ensuring that we are generating many more electrons for every dollar that is invested in generation and transmission infrastructure. So I’ll stop at that. Any questions, I’m happy to clarify.
Angelina (13:51)
Thank you, Simran. I think we’ll deep dive into some of the Indian context and the challenges, both in terms of state capacity, skills, infrastructure a little bit further. But, before that, I wanted to also ask Priya – your paper on Tackling Climate Change with Machine Learning that also led to the launch of Climate Change AI, identifies a few high leverage and long term machine learning applications for the energy space. Can you elaborate on some of these applications, and what you think are maybe some of the most promising and useful applications in the context of India?
Priya Donti (14:42)
I would say there are a couple of high level ways that AI plays out in energy related applications.
So, one of them is helping to distil raw data like satellite imagery into actionable insights. So for example, taking satellite imagery and using that to assess, where are solar panels installed? Where exactly is the grid infrastructure? In situations where we don’t already have that on the ground information, of course, gathering on the ground information is always the first choice. But, in situations where that hasn’t happened for whatever reason, these kinds of techniques can help fill in a gap.
AI and machine learning can also be used for forecasting. But one thing I’ll clarify or qualify is that if you’re going to do a pure data-driven forecast—taking historical data and predicting something—those predictions often only hold up in the short term. If you think about how electricity is consumed, you might assume that how electricity is consumed today or next month is similar in pattern to how it was consumed in the past. A machine learning algorithm that is learning patterns from the underlying data will identify the patterns that hold today and try to project those forward. However, much further into the future, these patterns may not hold.
When you have cities being built out, urbanisation, and people changing where they live, along with different macroeconomic and microeconomic trends shaping changes in those underlying patterns, pure data-driven forecasts no longer hold. In such situations, AI and machine learning can, in combination with economic models, techno-economic models, and so forth, enable better long-term forecasting. Often, this involves using a techno-economic model, which might be very complicated or expensive to run, and helping to approximate or slim down portions of it so that you can run more instances and scenarios. It’s important to note that AI and machine learning are really learning patterns, and if those patterns don’t persist, a pure machine learning approach will no longer work.
In addition to these applications, machine learning can play a role in helping us optimise operations in certain situations much more efficiently. For instance, think about controlling a heating and cooling system in a building, or scheduling power generators on the power grid in a world with more distributed devices and variable renewables. In terms of longer-term applications, AI and machine learning can also help accelerate the discovery of next-generation clean technologies. For example, when designing a better battery, there are many different combinations of chemistries to consider. AI and machine learning can learn from outcomes of past experiments, where you synthesised a battery and tested it, and use that information to suggest which experiments might be most fruitful to try next.
Building on my colleague Simran’s comments regarding data transparency, I believe AI and machine learning can be particularly useful in scenarios where you have a second source of data that might help pinpoint some of the numbers differently or where you have the ability to perform large-scale anomaly detection. For instance, if you have smart metre data from multiple sources, you can detect inconsistencies among them.
There are things like that, but in some sense, AI in general tends to be dependent on the quality of data input. There are some limited situations where it helps with the data transparency issue, but it itself requires good data and data transparency in terms of planning for the future. Again, I think it can play a role in forecasting, but there are qualifications: a pure data-driven approach doesn’t tend to work. It tends to be that AI isn’t assisting in the context of a techno-economic model.
Regarding the question of state-declining capabilities, this is really interesting. AI has been used to help synthesise literature or perform other tasks to provide more guidance. However, one challenge is that these syntheses are not always accurate. Sometimes AI is synthesizing out of context, so there might be some usefulness there, but care is needed to ensure that people using AI as an assistive technology have enough base understanding of the area of expertise – in order to contextualise whether the information is actually correct or whether it’s the whole truth as opposed to just part of the truth. So, there are some nuances to consider as well.
Angelina (20:10)
Yeah, thank you. That’s really interesting. Also, connecting this to the work that Siddharth has done, for instance in Rajasthan and Gujarat, as discussed in one of your papers—where you talked about how decisions on the ground regarding tech adoption and energy transitions often involve a confluence of different elements. It seems that capacity and the way we plan to integrate systems into various transition trajectories also matter.
Siddharth, could you elaborate on how you see AI and digital technologies influencing India’s pathway towards a low-carbon future? At the same time, could you tell us how, from your perspective, contextual factors are likely to impact the adoption of AI or other technologies for energy, such as smart grids, or any other kind of energy transitions?
Siddharth Sareen (21:29)
Thanks, Angelina, I think I’ll pick up on a couple of points that Simran and Priya have raised already along the way as well. One which Simran brought up was around the importance of accountability and transparency, and that, to me, really resonates with what Priya just said towards the end, about the need for AI not to become sort of a cure all, and, oh, this will fix it, and then just become another black box where there’s some input going in, some output, and we sort of magically think it’s going to solve everything. It’s not.
India has, of course, a number of different contexts within it. It’s got the complexity of a subcontinent and I think that it’s really worth keeping in mind that some of the challenges that we have, we don’t really need AI to solve at all. Some of the challenges in the electricity sector are questions of infrastructure planning, of implementation, of delivery. It’s about making sure that there is a logic at work that’s not politically modulated as this sole criterion, right? Quite often we come from a place of not acknowledging that formally. And, a lot that has to do with the energy sector, a lot that has to do with electricity is politically modulated, but, but then there are, there are targets, there are metrics that you can’t sort of just talk your way through, such as total electrification, right?
So now we, for some years, have lived in an India that has technically achieved that. What does that mean? What does that mean in terms of tracking the actual levels of energy security? Could AI be used to empower people with the kind of monitoring that we need to lay claims to the ones running these systems, to basically help create a crowdsourced set of parallel checks on the system. Right? How much power outage have we actually seen in what areas can we see a map of distributed real time availability, and then have a basically figure out what the patterns are, and visualise that in a way that you can’t hide behind national statistics, behind panel data that obfuscates the temporality of these things, right?
I think that would be a very powerful, quite simple thing to be able to do, and yet very powerful in terms of equipping people to really call on policy to be implemented in line with the targets, and with the achievements. And that’s not particularly, as I said, it’s not a technically innovative or novel way of doing things for its complexity, but more for its social situatedness. There’s a bunch of people who’ve been doing interesting work on Smart Metering in India. So Ankit Kumar at Sheffield is one name that comes to mind, and part of what he’s shown is that and others in this sort of work, that smart metres when they were rolled out in India, and that’s still very much ongoing in many places, but really in some of the urban contexts where there’s gone quite far, some of the places where there’s private utilities, for instance, that they’ve been serving a purpose that is much more related to monitoring and penalising for theft, for truncating commercial losses, as they call it, rather than for optimization and for increases in efficiency, which is the kind of discourse that one hears in many European countries that have gone through similar rollouts. If you look at the kind of leading places for smart metres in the world, there’s China, there’s some parts of the US, right? So there’s California, there’s Texas, Florida, but it’s really patchy. And within India, that patchiness is very evident as well.
And I think we need to think about what role AI is going to play within that patchy reality.
So, I think it’s probably going to be several different roles, and we need to differentiate this at sub national levels. We need to differentiate it for states, for urban versus rural contexts. We’re talking about quite often, realities where people are not equipped to know what’s going on. What are these models? What are these data flows that are being captured? And that’s a big conversation in places like the Netherlands, where you can sort of ask about whether people have control over their data for on ground supremacy, right? For what kind of consumption, whereas you also see examples of mini grids where you can really see the metres put outside the houses on display, almost as a transparency mechanism, right? And there are security risks that there are. These are, I think, the key thing with AI and this goes back a bit to some of the things that other speakers have brought up. There’s the demand perspective, which has been sort of putting forward a bit. And then there’s the supply side perspective.
And I should bring up a story from fieldwork in Rajasthan last summer? Where I spent a long time getting hold of somebody at an electricity distribution utility office towards sort of you know, one of the people you’d expect to be very much in charge of and competent about smart grids in Jaipur, which is one of India’s 33 smart cities. Well, this 100 smart city plan, but this is one of, one of the front runners, let’s say, and, and this was such an incredibly busy person, we were interrupted probably 10 times in the hour that I spent in that office after having, you know, spent a lot of time booking, booking time, because these are busy people. And I said, Well, what are the plans for being able to implement particular kinds of models? How would this roll out go? How do you envisage these targets panning out in practice? Looked at me and said, Are you crazy? Can you take these things seriously? These are just policy documents. You know, we’re barely coping. We have shortages of transformers. We have running losses despite, you know, today, and then the next scheme, and then the next scheme, very just about. And you can see this because the guy has people coming in and getting signatures on documents 10 times within the space of an art doesn’t have time to read through everything. It’s just damage control mode. So, we need to be very careful about who’s promoting this discourse for what it’s being promoted and why we’re talking about AI with, you know, a level of promissory potential that has been allocated to things like smart metres for a long time also, and that hasn’t really benefited people. This evidence is there because that’s been around for a decade, and even in a very progressive policy context, right as I’m sitting in Norway and and if you look at what’s happened with smart metres here, people have actually paid for them themselves. Most people don’t know this because costs are recovered over 10 years. The benefits have gone to the supply. The efficiency gains that are captured are not resulting in somehow magically lowered bills for households. If you lower system costs, if you can make better use, you can make your electricity stretch further. That’s not a bad thing, right? But then let’s talk about it in those terms.
So on the supply side, if AI has a role to play that is going to optimise systems, that is going to increase efficiency, that is going to create better control, then let’s talk about it as allowing us to make better use of a more variable load model, of, you know, decentralised generation of different temporalities of generation, than a base load model. And that’s a good thing, and I’m all for AI serving a purpose there. I don’t think that’s something that we’re going to have a lot of awareness of in the public domain. It’s not something we’re going to have people making very highly informed debates about, you know, making opeds in newspapers. So that’s, I’ll leave it there for at least this opening.
Angelina (29:13)
I think you brought forward a lot of important themes, right? I think starting from that question. So we often, you know, we often hear or engage with the AI discourse. So one hears about it mostly from like the starting point is usually AI benefits, right? What are the ways in which AI can help? But there is also a conversation, also a theme that you brought up, which is about assessing need, understanding where it’s needed, and also, then kind of going deeper into the complicated issue of who benefits ? How does that translate into questions of energy justice or energy poverty? It’s not even just access. It’s about affordability. As well. And I think alongside that, what I thought was really interesting in what you brought up in terms of, you know, what kind of use cases can one imagine, you know, in this space, that are not essentially about the supply side of it, or that are not essentially, in some ways top down, because the you know, policy example that you also gave is that, you know, these are at a certain level of discourse, and they function with that kind of, you know, promissory effect. But there are, there might be ways in which, like I said, we could increase visibility. There might be ways in which AI can actually be, you know, used in ways that empower citizens or empower communities in specific ways to kind of, you know, move past national statistics, like you said.
And I’m curious on this note, you know, and I want to kind of go hear more from you, Priya, on what you thought of, you know, what you think of this kind of this idea of, how does one assess need? How does one talk about AI that is not in this kind of broad strokes, you know, these are the benefits, and this is how it can be used, but really in a more sort of needs assessment oriented sort of a way.
And also, you know, what do you think of this potential for AI to be used in different kinds of, you know, bottom up approaches, for the lack of a better word, but ways in which that can actually help communities, you know, navigate energy futures a little differently. And just wondering what your thoughts are on that and, you know, and then I’ll come to the next question.
Priya (31:44)
Awesome, yeah, thanks for that question. And I guess what I’d say is, I think that we’re right now in a world where, as also my colleague Siddharth mentioned, right like AI, is being talked about with some kind of exceptionalism. If you took a lot of headlines about AI and replaced those with the terms civil engineering or political science, people would think your headline was absolutely ridiculous, right? Will political science save the world? Right? Will civil engineering save the world? And yet AI like these, you know, fields and areas of inquiry is a field and area of inquiry, it presents a set of tools, and those tools are useful in certain circumstances. They’re not in others, and they come with their strengths and limitations and risks. And so I think really, what we need to get to is a situation where, just like if you’re in a room trying to solve a particular problem, in some sense, you have an intuitive sense of do I need a civil engineer or an architect? Because my problem is well matched to that. Do I need a political scientist? We need to get to that level of literacy within the general population to really understand intuitively, is AI actually well matched to this problem? And also, when pursuing AI, what are the things I have to think about to make sure it’s done? Well, right?
With civil engineering, you have, for example, some certain types of permits and standards apply, and you know that you might put waste into the environment. So there’s regulation around that, right? You have a sense of what those risks are, and they’re codified ways to deal with them. And I think those kinds of standards and best practices, in addition to just kind of more crystallised and widespread mental models, are going to be really, really helpful here. And then, similarly to this question of right, can AI be used in ways that, you know, empower people, right, rather than only kind of, you know, government institutions. Absolutely, but it’s in some sense, sometimes the same technology, and even potentially the same application can be used in a different ways or used by a different set of actors that enables that. So for example, as other panellists have mentioned smart metres as a technology can reduce system costs, but whether that cost saving gets passed on to the consumer or to the supplier is sort of, it’s not the technology itself that caused that to happen either way, right? It’s sort of the policy surrounding the technology. And, so I think AI also has to be contextualised in this way, where we kind of use it for a particular technical function, but who are we enabling to actually use it? So who has access to data that they can analyse, for example, using AI? But then also even for a given application where, you know, multiple people could, could accrue, or kind of get the benefits of it on the other side. Often it’s a social and policy question of, you know, how are those benefits actually being distributed?
Angelina (34:50)
I think the question of when is a problem, a problem for AI is something that needs to be thought more about. And I’m curious as to what you think one might achieve that level of literacy in understanding this in a much more sort of, you know, as any other sort of tool that is at our disposal. But, before we come to that, I also wanted to just ask Simran if you want, wanted to add to this conversation in terms of you know, what you think are some of the challenges and risks in applying these technologies like both Priya and Siddharth pointed out, are there any additional social risks that you kind of see, or are there, you know, capacity, if you could delve a little bit deeper also into the capacity issue, I think that will be really helpful.
Simran Grover (35:51)
I think Siddharth reiterated what I was trying to say early, although I think Priya slightly disagreed on the issue of transparency, contactee, see some of the regulations. Let me just say that India has a comprehensive regulatory compliance framework across most states, right? But when it comes to the actual compliance of it, there are major issues. I’ll give you an example, right?
The Bureau of Energy Efficiency has something called a mechanism for energy audits, where reports are to be quarterly, and annually submitted for distribution. There are different distribution companies– there are about 60 odd companies within India, respond to it differently, right? Some, like in states of Rajasthan, will upload scanned copies of those forms, which will be very, very difficult to read and interpret. Many of those data points are completely empty. So the issue is in terms of, you know, governance, transparency, how much we want to empower AI is, I think, something that will be navigated and negotiated in the process of adoption of AI itself. Because eventually, I think, from a system design thinking, it also tries to cut across, and I agree with Siddharth, you know, the political economy of the of Indian states vary a lot, right, in many more ways than one and when it comes to systems designs, it does try to cut across those political, economic variations to some sort of convergence on standardisation, etc.
So in those processes, you know how much we do? Do we want to empower AI, for a lack of better word, as a compliance inspector, to enforce the compliance of policies, bring those data points out in public domain to ensure also for a more holistic discourse, uh, on transitions for citizens and for and definitely for CSO organisation like ours, and some of the issues are very basic, which a citizen does not even envisage, and many resources also, you know, done does not understand, unless you hit the ground, unless you look at data, more mandatory.
For example, if I look at some of the data in Rajasthan, I’ll see that some of the substations are energy generators without any energy generation unit in place, which is, which is physically impossible, right? So, how do you really bring accountability? I think AI does has potential in in in detecting, in highlighting these differences and and again, how much we empower it is will be a discourse on, on trade offs and negotiations, which will happen in policy places as difficult as it might be, there’s a potential possibility, in terms of, you know, benefiting and empowering consumers again, from a power sector point of view, there’s lot, especially in terms of, I think, safety, security, etc, there. There are issues on the ground, which I’m not 100% confident, but I think AI can detect, for example, absence of a relay or absence of a certain protector, and ensure that or raise it at least as a red flag, one of the accountability mechanism that is being pushed by the Ministry of power and across states and states have have adopted it through regulations, but not implemented in practicality, is the issue of compensation against SOPs, which is standard operating procedures, which which provides a benchmark of quality of services, so whether it’s any voltage variations which may lead to, you know, malfunctioning of your devices and which will be a cost to consumers, or any uninformed curtailment of power. So far, you know, it’s up to the consumers to document the evidence and put it as an app, as an application to claim SOP, which might be just 100 bucks and which is definitely not worth the effort. So it does lead to, lack of, lack of accountability by the electricity utilities towards the consumers. And coming back to the issue of possible challenges implementation. It’s hard, at least for me, to identify those right away, but one of the typical policy challenges is the issue of exclusion. And then I’ll try to bring out an example from a different space. I used to work on energy access before I founded seed angelina, you have, you have heard this example before.
So basically, we were working with a rural bank manager of a rural bank, and he was refusing to extend loans to a particular tribal community, saying that two people had defaulted on a loan 10 years back. That was information for decision making to deny certain services to villages to the community. Now I’m afraid that, you know, developers of AI may not sense. You know, sensibilities and sensitivity should not be assumed when it comes to technology, right? Our developers of AI really sensitive to the real, reality of ground conditions may be, you know, availability of networks in certain areas may may be the issue of tampering certain equipments, because the reality is, billions of dollars have have been pumped in for implementation of SCADA systems for substation and and 11 KV transformer metering. But at the end of the day, you see, the infrastructure is procured, but it’s not maintained, right? So what do you really do with that?
Angelina (42:21)
On that note, I think Priya, you’ve also talked about, you know, how, in this very paper, you talk about how machine learning applications are not a silver bullet. And you also talk about, you know, calling for collaborations, right? In this context, what in your mind is the kind of nature or shape that collaborations must take, because, as Simran also rightly pointed out, that, you know, the sensibilities may not match the considerations that are required to go into building the system, may not go into it.
And also, how do you see things differently? How do you see, kind of who is responsible for ensuring that you know sensibilities match that where the problems of equity access are being taken care of, right? How do you kind of understand this network of collaboration? And you know, also, I think to Siddharth as well, if you please, feel free to jump in on that. But yes, Priya over to you.
Priya (43:15)
Thanks so much. And before answering the question, I think Simran points on data transparency. I think a lot of those applications make a lot of sense. Personally. I think the terminology I would maybe use instead of data transparency for some of those is data actionability at scale, in the sense that for a lot of those, it’s a situation where, if you have uploaded reports, for example, from from an entity, it’s not that like, arguably, right, that someone couldn’t go and look at that information and extract it out. It’s that doing that at scale and actually making it easy for people to act on the information is, is the bottleneck where you can then leverage something like AI machine learning, but if the data is not open to begin with, right? Or if you don’t have enough information to even understand the rules that an algorithm should use to interpret that data, so if some of that information is not already out there, then there’s nothing sort of magic that the algorithms can do. Whereas, if there’s some understanding of how to analyse this data, but it’s a matter of just doing it at scale, then, then AI can help. So maybe that’s where terminologies wise, maybe yeah, like data, actionability at scale is what I would say. And I think that class of applications makes a whole lot of sense. Yeah. And so to get to Angelina’s question about, you know, yeah, who’s responsible for making sure that sensibilities match? And I 100% agree with this point, right? That many people from the technology sector come in with a little bit of a saviorist attitude without really understanding what’s going on on the ground. So I think there are, you know, several things that need to happen. So one kind of entities that are situated within the kind of AI space that are doing education of those who are kind of up and coming or already working in the AI field, trying to actually inject those values of humility, of really understanding and learning from the situation that you’re in, rather than assuming some lens on that, you know, there’s some educational responsibility.
And definitely, for example, in my role in a university as an AI and machine learning educator as well as kind of through the nonprofit hat of trying to educate the AI research community, that’s definitely something that I and we, you know, as a Climate Change AI organisation take very seriously. I also think that given that AI and machine learning are basically never an end to end solution to any problem, then it also means that the way we actually structure degrees in education around AI and machine learning should actually be deeply cognizant of that. I think, in the most extreme sense, I almost think that nobody should, for example, be able or allowed to graduate with only like an AI and machine learning bachelor’s or a master’s degree, and there’s these programs, but that there should be a mandatory minor, for example, at the very least, or kind of, you know, certificate or something in another area. So you study AI and political science, you study AI and electricity systems. You study multiple things. And in the process of doing that, when you’re in those formative years of, you know, building conceptual and mental models around those topics. You’re sort of jointly building it around maybe AI and another area, so you really understand how they interact. So I think that we really have to think about AI education as being situated within the broader space, and so then you should also understand what it’s situated within as a fundamental part of the education.
I also think that there are, you know, scenarios where you need someone who can innovate on AI and machine learning. But there are other situations, many situations where you really don’t, where it really is just a matter of ensuring that people are able to utilise the existing and, you know, open toolkits and kind of one great thing about AI and machine learning is that, you know, for many, many years we’ve there have been kind of open source, free to use, you know, code libraries that that people can can go forth and just and just use. And so I think also in those situations, it’s less a matter of making sure that AI people’s sensibilities match, but rather empowering the people who already have the sensibilities about a particular area to leverage various tools. And to your question of, how do we get to a place where people can do that, where AI literacy is there? I mean, I think that in some sense, there are a lot of kinds of open courseware and open tutorials. And so sometimes, I think it’s just in an hour, you can get to a place where you actually have a very solid mental model. What AI is, you might not be able to code it, but you might at least understand kind of what is going on there. And kind of, you know, one or one classes worth of investment, right? Can get you a lot farther than that. So I think it’s also just understanding, kind of who we can better match to those resources that maybe you know that in many cases already exist, and sort of allow them to bring that information into their existing context.
Siddharth (48:36)
So many great points I can just pick up on, a few things that have come to mind. The first is around this, you know, well, who should take responsibility? We often hear, well, we need green jobs. We need the green transition to deliver green jobs. And when you think about AI, you often hear people saying, oh, but this is going to, you know, make a bunch of people redundant. And I think it’s a kind of a flawed conceptualization of what a green job is.
A green job is not necessarily a job in the energy sector. A green job is not necessarily a job that says renewables around it. It’s a job in a whole bunch of sectors that are all going to decarbonize, right? And if you think about what that means for for things like electricity and where AI comes into it, then it means having people who are aware of the kind of context in which electricity is distributed and which is produced and what the needs are, right, deeply competent in that, but also able to then have this other layer that pre articulated or like, what is the use case for AI? And that’s not everybody who can spend an hour and glean things off of a course and apply it right? It’s somebody who needs to understand the context and be able to then glean and think this is directly applicable, or maybe indirectly applicable, to what I do. So I need to be the intermediary there. So building that competence, and that’s that’s just one sector. We need to do this in so many sectors, because when you have these. 10 year transmission plans, right? So you’re going to build out transmission infrastructure. We’re talking crores and crores of rupees and investments. What you have in terms of innovation today with AI is that you can basically monitor the usage on these lines and do machine learning on those to be able to figure out where you can optimise to save something like 20% of the waste capacity, so that you can just forego investments of crores or 10s or hundreds of crores and use this existing transmission infrastructure to be able to transmit way more electricity, right depending on the patterning. That means that we need spatial planners and the ones who understand catastrophe maps and work with those new processes, kind of licensing for land acquisition and so on. We need them to be able to digitise their ways of doing things, and bring AI into that in a way that syncs up with these electricity transmission investment plans. Now that’s not one kind of job. That’s many kinds of jobs with a particular competency added on. But we also need a bureaucratic, you know, bureaucratic leeway in these things, so that these people are not people who are being told you do this. These people are people who are allowed to add that level of managerial responsibility, allowed to push back and say, we need to do it like this. We can see these savings. We can see scope. And that means we need green jobs that are not just entry level. When I talk to developers in the desert, they say, Well, you know, people in academia and people who are training the next generation, they say, we notice that what what they want to hire when they’re building out solar parks, and what have you is people you can pay 15,000 a month on short term contracts. So they want somebody from an itI, they don’t want an IT graduate, they don’t want to. The thinking is, we just need to get this done at a level of logic that’s existing so that we can maximise profit. It’s short termism. And I think that with AI to really maximise benefits that you can get out of it. You need to think about, what are we moving towards in a five year perspective? In a 10 year perspective, how can we avoid doing that in a way that’s going to get us stuck in some costs, in ways of doing things that don’t make any sense anymore by the time we get there? And I’ll just give you one example of that, because I think it’s useful to concretize this. And that’s beyond the electricity sector energy generation, it’s transport. That’s one of the places where a lot of the decarbonization has to happen. Has barely begun happening in India, but clearly, you know, about a quarter of the emissions come from land based. It’s one of the easier sectors to solve, even if it doesn’t look like it.
Look at examples elsewhere, and you see that it’s pretty doable to electrify these modes. If you do it with a bunch of energy guzzling cars, it’s going to still require a huge amount of electricity demand. If you do it with public transport, if you do it with the different approach to transport that’s more collectively oriented, then you’re going to cut those energy requirements a lot and well, to wheel you basically already cut them into 1/3 right by electrifying rather than being on fossil fuels and running these fleets. And then, when you think about what, what urban transport for, just forget about regional networks, even. But urban transport looks like you’re talking about decongesting cities. You’re talking about a lot of hours saved in everybody’s lives. You’re talking about less exposure to pollution and public health, adverse outcomes and so on. But you’re also thinking about the spatial planning of those charging infrastructures. And you see things like, at her energy, right? You see things like these scooters that are going around and that can charge in a few minutes and that are electric and that have short range, 80 kilometres or whatever. They’re increasing. Maybe they’re under 50 now on a charge that takes a few minutes. And you think about who can benefit from this, you say, are there housing cooperatives? Are there a lot of electricity connections already where there’s distribution capacity on the network? Can these people benefit when they’re out of home, when there are times when all of that isn’t being used in a whole housing block? Can you plug in a couple of places, and is that going to save public investment that would otherwise be needed to expand the distribution network in the place that we don’t have in our congested cities?
Anyway, that’s an example, because it shows the usefulness of this, the utility of this logic of thinking cross sectorally. When you’re thinking about transitions, it has to come somewhere. And if you look at a place where that role has progressed, then you see that there’s a lot of investments going into it that could be avoided with a little advanced planning. And I think that’s where AI competence saves people from making the mistakes and the sub optimal investments that ultimately impact the public purse. So I think that’s a good argument for policymakers to understand and to push, because that’s where you’re going to save taxpayer money, that’s where you’re going to make responsible investments. I think too much of the onus is placed on trying to create some kind of democratic momentum for a bunch of things that are technocratically determined. Yeah. They need to be determined technocratically In sensible ways, and you need to equip people and build competence among people who are the key decision makers. And we can’t change that whole public culture. We’ve got to try, but we can’t change it in the time scale that we’re talking about, implementing some of these things.
Angelina (55:17)
Thank you for that. I think the last statement was quite impactful in terms of, you know, thinking through. I think there’s a lot of thinking also needed in terms of, then, how do we equip decision makers, like, how do we enable better decision making capacity as well? I think one strand of thought is, how do we make decisions more inclusive participatory, but the other strand of thinking is also in terms of, how do you make better decisions around it? And I think nowhere is the idea of, you know, here’s a problem and here’s an AI solution ready for it is not a settled question. I think it requires a lot of thinking in terms of understanding whether it’s the right way in which we frame the problem, and is it the right way to kind of proceed with any solution or not? And I think that requires far more deliberation, discussion and bringing in different actors into the mix. So thank you to all our panellists.
I’m just looking at the chat and Q & A in case some things come in. We are also at the last two minutes, which is also why I am kind of turning to chat now. But yeah, I think there’s one question that’s come in that says, can digital or AI driven solutions be tailored to support informal and decentralised energy systems that are prevalent in rural or underserved areas of India. I think we touched upon it a little bit when Siddharth was talking about the idea of sort of, crowd sourced ways of understanding how power is consumed or where the gaps are. But Would any of you like to elaborate on this or answer this question and get into it a little bit more?
Priya (57:19)
I was going to say yes, in the sense that, you know, AI is used for, you know, things like, you know better optimization of, you know, microgrids, or for kind of analysis of decentralised data on, you know, how infrastructure is used or could be used. So in some sense, it’s always a question of, you know what data is available to you? What are you trying to analyse? And is it something that you know either is sort of, you’re dealing with a large scale of data to analyse, or it has to be done. Yeah, scale is usually the word, right, in the sense that, if I have to, you know, home solar systems that are providing energy and I want to understand something about them, I can probably just look at the signal, that is, that is occurring. But if I want to do this for, you know, 30 systems, maybe it’s a bit harder. So it’s really always this question of, you know, is, again, their data to analyse, is there a pattern to be gleaned, and is there some reason that kind of manual analysis doesn’t doesn’t make sense? And a lot of those problems come up also in informal and decentralised energy systems.
Angelina (58:37)
I think also here is where I think maybe that the intersection with questions of digital literacy or digital divide might also come in, in terms of who is the digital subject that one is imagining at the end of the day, using these systems, working with these systems, right? I think there’s another question which is rather long. I’ll try and summarise it a little bit. So just in terms of decision making on energy and AI, how can we bring in community members? Priya mentioned talking about a use case of deploying AI to help manage renewable energy systems. But what about land related trade offs and water scarcity resources in places like Rajasthan. How do we bring communities into this conversation? And one what kind of institutions do we need to tackle this? Simran?
Simran (59:36)
That’s a tough one, and I think this is also a tough one because it goes beyond AI looking at land governance issues, which tend to be very complex. And Siddharth also has worked on it closely. I’m sure he has thoughts to please, but I think to look at the intersection of the discussion in the previous session, right. Uh. How communities can be brought in. So Dre, Dre is a spectrum of technology solutions and products. It’s not just a product, right? So if we are, if we are looking at just something which is off grid and which is not connected to the grid, then I will go back to what Pierre was saying. It cannot really. It can do some for short term forecasting, but not really. You know, long forecast, long term trends, right? And that’s one of the challenges in designing off grid systems to meet long term aspirations and local economic growth of the community meeting those energy demands through off with Dre systems now. And the second, I think more, bigger opportunity, which has challenges of notes, is integrating those Dre systems with the grid. Custom is one example. But there’s a lot more that can be done in terms of micro bits, etc, etc, right? Either, though regulatory frameworks are missing for it. Some of our research does show that if you take away the bidding subsidies and other discounts that are being given to large scale utility players in solar and wind, tail and integration of re tends to be cheaper in most cases, right? It also saves tons of money in terms of, you know, transmission infrastructure investments, and you can invest that money in strengthening, further strengthening, the distribution infrastructure. But of course, there’s a political economy of play, at play which favours centralization over decentralisation. And of course, it’s there’s another challenge of how fast we can build an ecosystem around it to meet some of the national aspiration and NDCs of 2030, and beyond, in terms of reducing our energy footprint and increasing renewables deployment, where I think centralization does have certain advantages on the issues of integration of risk on land, land use patterns, water scarcity, etc. Again, a lot can be done and that is one of the challenges, for, for you know, policy departments, also government departments, where they are often sitting on tons of data, but those data sets don’t speak to each other, right? But the challenge will also remain on the issues of collaboration, how to open this and easily those data sets are available. We work in the distribution space. And you know, we had a collaboration with IIT Kanpur, and despite MOUs in place, and over two years, we were not able to access some basic data sets. So I think there has to be, there are political economic challenges. When you look at these and again, these are again, trade offs and negotiations which need to happen in policy space. So that over to you, I think, I’m sure you have some comments on this.
Angelina (1:02:56)
I think we are also going a little bit above time, and I know that both Siddharth and Priya have to jump off. Maybe we can stop here. And I think, you know, because the note on which Simran ended, I think, is my larger kind of takeaway for this conversation is also that, you know, there’s a lot of political questions at stake. And you know, it’s one thing to have the technology. It’s another to make it work for the people. So I think a large part of our conversations today has focused on decision making, and, you know, the kind of actors that need to be involved with that. Said, I want to thank all our panellists for making the time to join us. It is truly an enlightening conversation. I learned a lot. And thank you to all the attendees who’ve also stayed on for the full hour, and also those who jumped off and might come back in later and check out the full conversation. Thank you again. I hope you all have a good evening and a good day. Thank you. See you. Thanks everyone.
You can also watch the dialogue here:
