KUMUDAVALLI SEETHARAMAN : Today's day and age it's safe to say that AI is powering everything. It's an AI powered world. You know when I say everything, everything, even businesses are powered by AI. You know we keep hearing things about you know, algorithmic pricing and machine learning and recommendations, all of these things. What is the basis of all of this right? It is data at the end of the day. And there is an adage that says data is the new oil.

Which is, pretty much it has found its true meaning in today's world. When AI is powering the world in terms of using data and you know it is the technological medium which is there to stay like you said in the previous episode, what are the guardrails that you know that one needs to keep in mind when we talk about AI in the sphere of competition law?

ABIR ROY : That's very interesting. I was just thinking before you, you took a very right word, guardrails. Now, before we analyze what is the guardrail, whether we should have a guardrail in this day and age of AI, which is at that stage, we need to first understand the value chain, we keep on talking about. We need to understand that there are two different phases. One is the development of an AI model and the deployment thereof. Obviously in development, you need inputs in terms of data. You need computational technology. You need the engineering talent. And it takes a number of years and a lot of expense to develop a model. Deployment thereof, the marginal cost may not be that high. So we need to appreciate the fact that there are a lot of stages which are there. What we are seeing is a lot of AI tools, which is at the deployment stage, perhaps. And the other stage is obviously preceding that is the development stage. 

So I guess the entire value chain, again, has to be developed, or rather has to be understood, to make out a case whether we need guard rails in the first place.

AMAN SHANKAR : I think especially when you talk about competition law as one of the primer here, for guardrails it is more important to understand what is the functionality of AI, right? Because there may be a theoretical notion of harm that we may talk about, but in competition law as the Supreme Court has settled and effects analysis is required, whether you do analysis of section 3 or section 4, right.

So the theoretical notion of harm that yes there may be a vertical integration or there may be economies of scale or there may be inherent barriers to entry all those are theories. So before that let's step back and try to understand also that what is the value chain that we are talking.

KUMUDAVALLI SEETHARAMAN : Sorry to cut you guys, but what is this value chain that you are even talking about?

AMAN SHANKAR : I'll just give a very brief example here. For example, let's take the example of a search engine that is operating brilliantly. Now that search engine has a humongous data set within its realm. Today, it decides to integrate an AI model within its system. Now that AI model will obviously have an advantage of that access to network, right? And with the efflux of time, economies of scale will play into its favor. For example, the marginal cost of production, that is the data and other computing resources that will be required, will go down. And the ultimate output of the revenue model that we will focus upon will go up. So this is the value chain that we are talking about right now.

VIVEK PANDEY : Precisely, see to understand the value chain and I will say overall market, one of the best ways is the market study conducted by the competition authorities themselves. So that is being done by India and across the globe and it's very relevant because if there is any new technology or there are new issues popping up in the market, it's better to assess how that affects everyone in the value chain that includes the producer, supplier, distributor and the consumers also. Because if there is any new technology, we may feel that there are new issues in the market. There is actually a harm in the market, but perhaps there may be a case that the nature of market itself is changing. And that is bound to happen. In this situation, it would be best if the competition authorities do their studies and come up with some conclusion and identify where the issues are. There may be issues like you mentioned vertical integration. There may be AI algorithm related issues. There are a lot of issues which is required to be assessed at first, see verifiable harms in the market and then perhaps take an action. It is very similar to that e-commerce study. So, the CCIA conducted the e-commerce platform study and then when it identified the issues, it took some of the enforcement actions. So, we will have to perhaps wait and see what are the issues.

KUMUDAVALLI SEETHARAMAN : The idea what I am understanding is that have a light touch approach, see how this know this system pans out and then take the call which is pretty much what MeiTy also did right. And the other thing I picked up from you is algorithmic issues right. So, when we are talking about algorithm there is this thing about surge pricing right. Especially when you talk about airlines or even food aggregators, like there is one drop of rain outside and the price skyrockets. So when you talking about algorithms, how is this surge pricing algorithm, what is the competition law angle here?

ABIR ROY : I think we have to really go to the basics and slightly not over complicate the entire issue. What is prohibited under that? What is prohibited under the act is collusive behavior. Independent actions are perfectly permissible. So, I may decide as a cement company for example, that my price of cement per kilogram is 1000 rupees, may be high, may be low does not matter. I have decided myself. Now that decision may be based on multiple factors. Perhaps an AI tool is helping me out. AI tool may be helping out depending on the demand and supply patterns. They may say you can have these many margins. And the customer, in this case the builders, will pay for it. That's perfectly fine. What is prohibited is, for example, again the cement example. I'm a cement player. Aman is a cement player. We decide and have a handshake agreement. OK, fine. Let's have it at 1,000 rupees per kilo. Now that is what is prohibited.

So that is the conceptual understanding. Now whether AI helps in that or not, that's something which has to be looked into. Now depending on an AI algorithm from a business perspective, in this day and age where like you said data is a new oil, it may be actually be a very eschewed business strategy. Because what does an AI perhaps will do? They will look at the demand patterns. They will perhaps look at say, fine, Amman is charging 1,000 rupees. So Abir, can also charge 1,000 rupees. That's perfectly fine. Because AI is developing its model on public available data. But the problem what will happen is where this entire algorithmic collusion has come in is the AI model is developed in such a manner wherein both the companies in this case say, in this exam me and Aman have adopted an AI and the AI suggests the price wherein please price it both a 1000 rupees and we agree to it. So, that is an issue because there is actually collusion.

So in that example, AI is the hub. So that is the entire concept. So we have to see from that angle, what is the AI doing? I think we have fundamentally said, I think earlier also AI is the medium. So what is the AI doing? So that is algorithmic collision. There's no human collision which is happening, but it's an algorithm machine-read collision which is happening.

KUMUDAVALLI SEETHARAMAN : So you just spoke about hub. Can you explain legally what is this?

ABIR ROY : So again rudimentary example, what is prohibited? Prohibition is two players in the same level of market. In this case, example that I was going to, me and Aman, cement players cannot collude. The law was silent on the part that what if say Vivek is a common distributor. He gives information to me also, he gives information to Aman also. And then we collude. So there's no agreement between us. The agreement is through him. So he becomes the hub and we are the spokes.

In that example which I was telling the AI is prompting so AI is the hub and we are the spokes. Hence the concept of algorithmic collusion algorithm driven collusion so that is the entire concept of

AMAN SHANKAR : I think Abir, the examples that we are discussing right now is more about AI-facilitated collusion. What happens in a situation wherein AI is capable enough to take autonomous decisions in your business? So I remember reading one of the cases of US, I think it's still ongoing, of Yardi systems. I think Yardi system developed an AI model wherein through which you can develop the rental inputs of the particular area, like what rents should be there and few property management firms took that technology with them. So there was some case that arose that there's algorithmic collusion because the AI was after a certain point in time capable to take independent decisions also. And it was then fixing the rent all around. And the people who were, have licensed that were acting also on it. Correct. So there was a case of algorithmic collusion. It went to the court. There was a motion advanced by these defendants saying that the case should not proceed to which the court said that AI should also receive same kind of condemnation which a traditional cartel or Hub and Spoke cartel would have received. Because even if it's providing those information or taking autonomous decision, you are acting on it. So there's active participation on your part. So that's also one of those aspects that will crop up and will throw up more complex examples. This can be one of the examples, but I'm sure there will be more complex examples to this.

VIVEK PANDEY : Technically, AI can be used as a transparent tool also for signaling. But if there is no evidence of collusion, then it is perfectly fine. It is like the classic cases merely because there is a price parallelism that does not mean that there is a cartel. So, there has to be an evidence of collusion. So, perhaps in that case, they would have found some evidence. But I would presume that if merely because the technology is involved and everyone is acting on that technology, that by itself should not be the violation.

ABIR ROY : I think what is basically at the end of the conceptually is it an independent slash interdependent action, or a collusive action? See we are living in a lot of industries are oligopoly. They will mirror each other. Now you may have AI, may have distributor, etc. as the case may be. The CC and other jurisdictions have said it's perfectly fine. What is prohibitive is collusion. So that is that I think is the line that is has to be has to be drawn. Like in the example which Yardi that you mentioned, AI was prompting tools perhaps on confidential data of the landlords. And people are following it. So there was a decision to follow. So there is a meeting, so there is a collusion at some level. So I think the principles still apply.

AMAN SHANKAR : But again that test of evidentiary standards that will obviously be more nuanced in these cases.

KUMUDAVALLI SEETHARAMAN : So, essentially what I am getting from all of you guys is that the classic rules of cartel behaviour still applies in some form of, or the other in these like recent changes of hub and spoke sort of model. But what about abuse cases right, because data like we spoke about is a critical element here. And not all users have the wherewithal to have that kind of data or to use an AI model you know. So how does, for example, the traditional market is a capital intensive market right and this market that we are currently speaking about is both capital as well as data intensive market. So, then how do we sort of see the abuse patterns, how do we see this scenario from the abuse perspective of things?

ABIR ROY : What the Supreme Court has basically said is that you have to show effects. In fact, Supreme Court has gone a step ahead and says not only actual effects but likely effects. Now, in this entire environment of AI-driven world that you were saying, there are certain theories and these are theories which have some backing. I will not say they have no backing whatsoever. They are concerns about self-referencing. They are concerns about tying. They are concerns about buddling. But having said that, first we need to really understand, I think ties back to what Vivek was mentioning, our market studies. Is it a digital market in a broader scheme? Perhaps yes. But that doesn't mean that all the concerns of digital market have yet come into the AI world. It may come in the future. I'm not saying no. But is it too soon to perhaps say, first we need to understand which level are we talking about? When we say we have AI tools everywhere, that means there's vast deployment which is happening. Now obviously, AI is also generating output. So that generation of output, it's basically, it may be amounting to search. So we have to see what are the various issues which are there.

VIVEK PANDEY : See, merely because there is an input and an output does not make an AI platform comparable to a search engine because the output is completely different. is actually generating something which a search engine cannot. So, we will have to go back to the principles of relevant market under the Act that what is substitutable. It has to be in terms of characteristics, usage and price. That is the point. We have to analyze the market thoroughly and see what is actually, what falls in the relevant market also that is also a key factor.

AMAN SHANKAR : I agree that ipso-facto we just can't term that a foundational model is a search engine. Obviously more intricate analysis on part of the regulator has to be there as to what is the substitutability of the product in the relevant market, right. And for that purpose I think market studies becomes the key, as the CCI is doing or as the METE is doing because then only you can gain more knowledge that what is the technology all about and then only you can come down to assessing and determining that how do you meet the standards of law and how do you come to a conclusion? So the market studies will be the key.

KUMUDAVALLI SEETHARAMAN : This is such an interesting topic, and within this topic I want to touch upon, you know, tech deals that we hear. We hear so much about, so many crores worth of a tech deal and this and that. So, how do you look at that in terms of a competition, regulatory perspective?

ABIR ROY : So what you said is so interesting. So what I'll do is I'll divide this response into two parts. First is, do these tech deals which you read in the newspapers, do they require the blessing of the CCA? The answer is yes. And the reason is because there's been a recent amendment in the competition regulation, which essentially says that if your deal is more than 2000 crores and the parties have what they call it as a substantial business operations in India, you need to go to the CCA for approval.

So fine, so you will go to the CCI for approval. Now how will the CCI approve that transaction? I'm more concerned with that because like we are all saying we have to do market studies to understand. Now CCI when they are seized of a matter has to give an approval or to give a remedy or to not give the approval for that matter. So what are the principles that the CCI will look into? Again, how do you define the market like Vivek was mentioning? What is the structure of the market? You will have situations wherein you are acquiring a company which is rich on data. So how is the licensing regime works? That is back to the copyright examples that we are talking about. So, and then say suppose if a fund house is investing and a fund house already has similar companies in their portfolios, how is the portfolio effects function? So these are the various nuances with the CCA will have to gather in such AI tech deals. Going to CCA is one thing, but approval is another.

VIVEK PANDEY : Yeah so, I guess there is one more facet that CCI itself has to be well equipped and aware about how the technology works. So, if they pass any remedy be it in combination or otherwise also it has to be a constructive remedy. After understanding the technology I am sure the market study will come in handy.

KUMUDAVALLI SEETHARAMAN : Interesting, I mean it is a fascinating topic because there so many discussions that are in play, right and with AI and so many laws you talk about competition, privacy, copyright, BNS that you are saying. So, it indeed a learning phase like you saying you have to understand what is happening in the market, the businesses have to understand, the policy makers have to understand. So, I think right now is it safe to say that we are in a learning phase?

ABIR ROY : We are always in a learning phase, AI, yes definitely.

VIVEK PANDEY : We are in a wait and watch sphere I will say .