- 11 minutes ago
At the AI Summit, experts debated India’s challenge of protecting citizens’ data while keeping pace with global AI leaders. CP Gurnani, Co-Founder and Vice Chairman of AIONOS, stressed governance and cybersecurity. Umesh Sachdev, CEO and Co-Founder of Uniphore—one of the world’s largest AI-native, multimodal enterprise-class SaaS companies—warned against overregulation and potential “data colonisation.” Saurabh Kumar Sahu, Managing Director and Lead for India Business at Accenture, advocated human-led AI governance to ensure innovation, security, and strategic competitiveness.
Category
🗞
NewsTranscript
00:00There are back-to-back events all over town and I am glad you could make time in Delhi's
00:06traffic today. So great to have you with us. How do we govern AI? If you kind of simplify
00:13it for the masses, CPI, I start with you. Does AI even need a governance regulatory framework?
00:22Europe has done something, other parts of the world are considering it. We are looking
00:27at various forms of, you know, amending our privacy, approach towards privacy. Does AI
00:33even need any framework of this kind or should entrepreneurs and entrepreneurship be allowed
00:38to do its job? The mic to your right.
00:43You know, Siddharth, when you designed an automobile, you did put an accelerator, clutch and a brake.
00:56When you designed an highway, you planned for growth in the number of vehicles that will
01:04ply on that highway. Now, would you say just the city planning is for the sake of planning
01:13or it is part of the governance? So my personal view is, you know, if data, which is really
01:24what AI is, and if data and machine learning, if we let it remain without any governance or
01:35without any cyber security, then we are getting, exposing ourselves to trouble. You know, you
01:42have seen it on some simple laws that have come in social media, you say, Australia says not
01:49below 15. Right? I mean, you remove, now this is what I call waking up a little too late, but
01:56at least you woke up. Here we have an opportunity as AI consumption is only touched of what the
02:05potential is, it probably touched only 1.5%. Yes. So wake up now and bring in the right guard
02:13rails. Bring in the right guard rails. Umesh, as I come to you, you need to be 25 years in
02:20many domains to buy a alcoholic drink. You need to be a certain age to get a driving license.
02:27You need to be eligible to vote. There are eligibility norms for everything. The point
02:34that CP Gurnani made, very well taken. In the world of AI and just before you came into the
02:39hall, I asked this entire hall of many, many students and I asked how many of you use AI?
02:45Every single hand went up. Are, is there need for any guardrails similar to the examples
02:51that I gave for the use of AI and therefore, consequently, the protection of privacy and
02:56data? So, since CP has taken one side, let me take a contrarian view. Um, my company is
03:05headquartered in Silicon Valley. We see what's happening in US and in close competition with
03:11China in terms of the pace of innovation that those countries are leading with, whether it's
03:15LLMs or GPUs or data center. And India clearly wants to be in the race. This week here, all
03:21of us are here to make sure that India is part of the race. And so, you know, while it's
03:29absolutely
03:30necessary that we all take cognizance of protecting our citizens against a very powerful technology
03:37which could have negative ramifications, but it's a very thin line. If we over-regulate
03:43at this moment when, like CP said, we're not only scratching the surface, if we over-regulate
03:49in India right now, we run the risk of being left behind from the two giants in AI, which
03:54is US and China. Now, should there be guardrails? Of course there should be guardrails. My company
04:00serves 2,500 other businesses that are using AI. Not one Fortune 500 company is using AI
04:07without guardrails within their own setup. It's not like White House or the US government
04:11came out with a regulation or paper. This is just common sense. You want AI to use a company's
04:18data? There has to be guardrails. You don't want a person working in HR department looking
04:24at financial data because the AI is now connected. So of course there are guardrails. Of course there
04:29are architectural ways of protecting information, protecting citizens. But I think we should
04:34not get confused in that there is this big role of the government. The government needs
04:39to protect the citizens. Let's over-regulate ourselves because the risk on the other side
04:44is that India might become colonized by US and China in terms of AI, which we don't want.
04:50And so we have to play by the rules that the leaders are running at.
04:53Saurabh, where do you stand? Two of the three panelists, he spoke about guardrails, he is
04:59speaking about data colonization and the associated risk. Where do you stand on this debate?
05:04Saurabh, I guess somewhere in the middle then.
05:08Saurabh, I think simply put the future of governance, for any governance will have to
05:15be human in the lead and powered by AI. So I think the combination of human plus machine
05:21would be the right optimal combination to put in the right safeguards and to interpret and
05:25apply AI well. Where does this entire discussion about, you know, lack of authenticity or guardrails
05:33being put in place come in? It essentially comes in because there is a general sense of
05:37feeling that there could be certain bias, there could be certain inefficiencies in the data
05:42that we carry. See, we hit the nail on the head when he said data is the foundation for
05:47any AI. So data can come with certain biases, certain unfairness and that needs to be therefore
05:52checked and applied clearly. And hence, therefore, a regulatory framework with the human in the
05:57lead to make the right decisions, reverse or overrule AI systems wherever required to apply
06:02in the right manner is, I think, the best way forward.
06:05Saurabh, I come back to you. You spent five decades and you've seen every era, every pivotal
06:11moment in technology, in software, in software applied services. The world that we are racing
06:17towards and in morning panels here, the world of AI agents was discussed several times. In
06:23fact, that is perhaps going to be the future along with so-called AGI. When both of those
06:30combine for a country where, as I see it right now, AI is getting mass deployed, both on the
06:37consumer side as well as on the B2B side, what's the ideal framework that you would advocate
06:43to the government, the policy makers?
06:47You know, so let's, again, I believe in governance and at, after all those decades, I still came
06:57back to set up a company called IONOS, which is really AI for enterprises. Now, and some of
07:05the products that we have developed, we're planning to launch it on my second anniversary, a product
07:14called Unistack, which is really about governance of all the AI agents that will be floating around.
07:21Okay.
07:21I mean, ultimately, they need harmonization. They need orchestration. You need to have somebody
07:28who's managing different sets of tools coming into play into your enterprise. Now, for a
07:36policy maker, on one end, I strongly believe that my data should not be misused. Elon Musk
07:47might say that, you know, even if you put the guardrails, he will take it away.
07:54Yes.
07:55But the point I'm again trying to make out here is that data protection for an individual. I was
08:03actually surprised when the Ministry of Telecom came up with the secure application on every
08:09mobile phone, and there was this hue and cry as if government wants to control the data. I thought it
08:15was a necessity that on my mobile phone, which is a gateway to anything and everything, I mean,
08:23there should be a, you know, secure application. And we, in the same atmosphere, we live in,
08:33every third day, we come across a digital fraud.
08:37Yes.
08:37I mean, trillions of probably rupees being siphoned away by some very smart people in, you know,
08:45the districts called Nu or wherever they are. Now, as far as I am concerned, the governance
08:51is first to protect my data, protect my savings, and make sure that anything that is available
09:00in public domain is not being misused against me. Now, I can go on in terms of policies.
09:07If you'll go and look at all the data that we have, I mean, on one hand, we are very
09:13proud of 1.4
09:15billion digital records. I mean, amazing effort when we did the Aadhaar. Amazing effort when we did the
09:23GST. Amazing effort when we did the DBT. Now, I again, the second question again out there is that is
09:32the new
09:32application that we will come up with. Is there a law that we measure the outcome, not just go away
09:39and let
09:39somebody else manipulate the data? So, I think there is a framework requirement here, and that framework,
09:47we could be, actually, I don't think Europe has done a great job. Europe tends to be black and white.
09:52And excessive, perhaps? Excessive regulation in Europe? And that they go in for extremes. My belief is that
10:00India can show the way and bring in more intelligent frameworks, more intelligent and smarter policies,
10:08and that adopts itself to the growth of AI.
10:11Umesh, take that point forward. If we agree with what Mr. Gurnani says that India can develop a framework
10:19and take the lead, I would add to that, that in India, experience has taught me that one size does
10:25not fit all.
10:26Because we are a very diverse country. If you were to take that forward, how would we be able to
10:33do it?
10:33Because the policy challenges in one state are very different from the policy challenges in another state.
10:40So, Siddharth, very rightly put, that world over the experience is showing us that AI is not one size fits
10:47all.
10:48It's not just an India thing, it's a global thing.
10:50Let's take some examples. Not every use case of generative AI or agentic AI is well suited with a large
10:58language model.
10:59There's this, you know, lot of talk in US and Europe that for specific applications such as healthcare or education,
11:07smaller fine-tuned language models are outperforming larger models both on accuracy and big difference on cost.
11:16Alright? So, thinking through this architecturally, my company Unifor provides this architecture to large enterprises
11:24where they are able to use their own data safely, fine-tune their own small language models for specific domains,
11:32and then build AI agents. Insurance companies are now processing claims using AI agents which are powered by small language
11:40models.
11:41We are working with the Fortune 10 oil and gas major who has 35,000 contracts and 1.5 million
11:49invoices that have to be matched.
11:51And the cost of not matching contracts with invoices is a leakage of about a billion dollars a year.
11:58And now AI agents are correcting this mismatch using small language models.
12:04Now, on the flip side, how do you put governance or regulation around this?
12:10Again, we don't want to burden innovation at this point.
12:14We want startups, we want students, we want universities to start innovating.
12:18We want companies, large and small, to start innovating.
12:21And again, the US experience is showing us that if there is a need for regulation,
12:26let that be centrally driven and not in 51 states in the US, or over 20 states in India,
12:33because that's one recipe of stifling innovation.
12:36You burden the innovator so much with state and national regulation that they, you know, can't take the next step
12:43forward.
12:43And especially as we are so nascent, India can be the third wheel globally in AI, behind US and China,
12:51but lead in AI diffusion, AI applications, AI adoption.
12:56And for that to happen, of course we need some safety and guardrails,
13:01but we need to keep enough freedom for innovation to run very fast here.
13:07You know, Saurabh, I am reminded of how we have dealt with previous challenges of governance
13:13and catching up with technology, crypto being the most recent example, in a different context really.
13:21But the point that both of our speakers made, population scale mass deployment,
13:26you know, people sometimes don't get it.
13:27Even for the AI summit, if there are 400,000 registrations in India, because you have that population scale,
13:34there are that many people who are interested in seeing what this summit is all about.
13:39For a policy maker who is dealing with 400 million database, let's say for health,
13:47or any other database in India, and we have so many of them.
13:51What is the ideal mix at?
13:53What point does he trust a private sector solutions provider in the AI space?
13:59Or does he go only with the Sarkari setup?
14:01And we have seen examples of that in the past.
14:04You yourself must have had experience there.
14:06Well, Siddharth, fair point.
14:07And for a country like India, and I agree with both the panelists,
14:10that we are at the cusp of really leapfrogging in the AI, you know, revolution.
14:14I think to answer it, I will give you an example.
14:17Accenture today has developed the world's largest AI multilingual omnichannel platform
14:22to address citizen grievances with the Department of Administrative Reforms
14:26and Public Grievances for India.
14:28When we are doing that, very clearly, guardrails have been put into place to say
14:31what kind of data should come in from the consent of the consumer or the complainant,
14:36visibility to be given to the complainant to say what is the tracking of the particular complaint as to where
14:41it goes,
14:41and then ultimately what is the resolution process behind it.
14:44So, essentially net-net, I am a firm believer of the fact that AI can be very well used
14:49if the ultimate end consumer or the beneficiary is given visibility of how the information is used.
14:55So, consent plus confirmation.
14:57To answer your question specifically, I think the point about private sector or sovereign does not matter as much
15:04as it matters to say there are three typical entities in which AI regulatory frameworks have to operate in.
15:11Sovereign, enterprise and consumer.
15:13For sovereign, the guardrails will have to be different because you are dealing with sovereign data, sovereign applications
15:17like the one that I just explained.
15:20For the enterprise applications, as I think CP himself is working in that space,
15:24there are areas in terms of use cases in which certain guardrails will have to be put into place.
15:29And lastly for the consumer, let us try to see if we can give the power back to the consumer
15:33to say,
15:34like for example in your phone, there are multiple apps which take your data.
15:38How many of them really tell you what are they doing with the data and how you can therefore benefit
15:42from it?
15:43If we as consumers of AI are able to get that particular visibility to say,
15:47I am consenting to my data but also give certain power back to me to say how the data is
15:51being used and how I can benefit from it.
15:53In that case, that's a combination of human plus machine wherein it can be regulated well and more importantly,
16:00AI is being used for benefits for the larger consumer base or enterprises.
16:05If they therefore are a participant rather than just a consumer, that is where I think the game really begins.
16:10You know, CP, on a lighter note, hundreds of millions of Indians have for years now been providing their data
16:19to social media platforms.
16:21They got virality, they got views and many of them also were able to monetize their efforts on those social
16:29media platforms.
16:30In the case of AI, you know, applications that reside on our phone, anything above the entry level, you people
16:41are paying up.
16:43But let's turn the focus away from governing AI to AI in governance for a country of the size and
16:51issues that we have.
16:53I want all three panelists as we come to the end of this session to give us maybe one or
16:58two examples
16:58where you fundamentally feel that AI will make a huge difference and outperform in the Indian context in solving Indian
17:08problems.
17:10I think it's a wonderful question because it changes the focus to output and results.
17:18It changes the focus to not saying, you know, what the digital infrastructure will do.
17:27It changes the focus to not the number of internet subscribers to actually output.
17:35So let's take example.
17:37Legal, we have a backlog of five crore cases.
17:41Can I use AI?
17:43And will I be able to measure the results that year on year I am able to reduce the backlog
17:50by, let us say, a rate of 30%.
17:53Number two, on healthcare, we all know National Health Mission is doing an incredible job because they are creating one
18:02of the largest health databases in the world.
18:05That the reality is that in comparison, we handle COVID a lot better.
18:13That in comparison, even the, you know, Krishna Ella's Bharat Biotech was able to create a vaccine in the record
18:23time, which was designed, developed and produced in India.
18:28Now my question, if I were to relate it to an output, then I want to know why am I
18:35the generic capital of the world?
18:37Why am I not a solution or a life sciences capital of the world?
18:44Because that's what data leads us to.
18:46We can go on and on, but every department, every sector, if we start measuring the output, I think we
18:55will be in a better shape.
18:58So, before I answer the AI and governance question, to your point of, you know, citizens giving data to social
19:04media platforms,
19:04I'm reminded of, you know, when I travel to Europe these days, there is no major corporation that doesn't slow
19:13me down and say, but you're an American company.
19:16And what if President Trump one day comes out with trade policy and says, we can't use you.
19:22So, sovereign AI in Europe is a big topic.
19:26But you leave the room and you ask the engineers in that company, what are you doing for AI?
19:31They say, chat GPT.
19:32So, they talk of sovereign AI, but they're sending all their company data to this public LLM.
19:38And so, you know, to your point…
19:40No, that's a great point that, you know, there are some people who are debating that why should India put
19:44in billions of dollars of money.
19:46Even corporations or the government, regardless, money in India being put into doing this.
19:51But you make that brilliant point that in the world that we live in, we need this capability.
19:56There is no doubt.
19:57There is no doubt that sovereign AI and controlling one's destiny as a country is very important,
20:05especially as geopolitics is getting mixed with technology.
20:08It's unfortunate, but it's a fact.
20:09We all saw in Davos a month ago how geopolitics and AI were coexisting in that one week.
20:16But coming to AI and governance, again, think about population scale.
20:21Who is the fintech leader in the world where 40%, 40% of the global digital transactions happen in one
20:28country?
20:28That's India.
20:29Yes.
20:30A bank like JPMorgan Chase, which is the number one bank in the world,
20:34has 1,600 people employed just to do anti-mully laundering and compliance checks on their transactions.
20:42Imagine the scale of compliance India needs to keep all our digital transactions secure.
20:48And those are low value transactions.
20:50JPMorgan has trillions of dollars of transactions.
20:54AI agents with fine-tuned models is what JPMorgan is using now at their scale.
21:00And that would be a gift to India at the scale of transactions that we are doing.
21:05Example one.
21:06Example two, think about the world's potentially largest health insurance scheme called Ayushman Bharat.
21:13Imagine the time it takes to file for a claim where somebody is in front of a hospital trying to
21:19file a claim so that they can get the healthcare service that they need.
21:24And at population scale, the number of people would be needed to check each claim and approve or disapprove a
21:31claim, etc., beyond a certain limit.
21:34Again, the biggest insurance firms in the world are now using AI agents with fine-tuned models to approve or
21:41disapprove claims in minutes, not hours.
21:44And at India scale, those examples, those learnings can really help scale what we are really, you know, set out
21:52to do, which is digitize this whole country.
21:55And I think AI is going to be a big accelerator for all those efforts.
21:59Sorry, the last word to you.
22:01The thing about the last is most of the examples have been already done.
22:04So, I have to go through something which is – but I agree with the health insurance part for sure.
22:09I think final words on crop productivity and yield output in agriculture.
22:13Clearly, especially with satellite, you know, connectivity coming in and with more powerful LLMs and AI, you know, technologies coming
22:21in.
22:21The ability to, therefore, give a power back to our farmers in terms of looking at their data or rather
22:26their crop yields and output will be, I think, something which we can make a huge difference in.
22:30You know, this is a subject that can be debated endlessly.
22:33But if you've been to the summit, I noticed so many innovative ideas of entrepreneurs, Indian entrepreneurs using something that's
22:43been developed somewhere else,
22:45but wanting it to be applied in the Indian context, and I think that's where the future lies.
22:51Each of these panelists here, experts in their own specific areas have given us thoughts and ideas to comprehend on
22:59and dwell on.
23:00Thank you very much, all of you, for being here with us today.
23:03With that, it's a wrap on this set.
Comments