AI agent is becoming a co-worker: Nitin Mittal, Principal at Deloitte Consulting
Category
🗞
NewsTranscript
00:00At the India Today Conclave, it's time now for what should be one of the most insightful
00:07and also one of the most important sessions of this entire conclave.
00:12So I want to introduce you to our two guests, joining us to talk about the trends in AI
00:18that you're likely to see through 2025 and I'm not going beyond that because of just
00:24the sheer speed at which things keep changing, so trying to project too far into the future
00:28is actually quite hazardous.
00:30Allow me to welcome first Nitin Mittal, he is the Principal at Deloitte Global, basically
00:36heads Artificial Intelligence Practice for Deloitte and from what I've seen, a lot of
00:41the world's biggest companies consult with him very closely on how best to use AI.
00:48So we're delighted that he's made the effort of coming down from Paris to be here at this
00:52conference.
00:53Thank you very much.
00:54I have Puneet Chandok, Puneet Chandok heads Microsoft in India as its President for India
00:59and South Asia.
01:00He also heads up CII's AI Council, so thank you very much gentlemen for joining us.
01:06You know, I want to start by asking Nitin about India's approach to artificial intelligence.
01:13We've just seen China over the past few weeks upend a lot of global calculus around AI and
01:22large language models by producing DeepSeek, which supposedly was produced at a lower
01:28cost and works more efficiently than the large language models produced by Microsoft, OpenAI
01:34and several others.
01:35Why couldn't, why has China been able to do this or one Chinese company been able to do
01:40this and despite the plethora of IT talent that India has, all these big behemoths that
01:46make billions of dollars, why did that innovation happen in China and not in India?
01:51Nitin, why don't you start from there?
01:53So the way that I would kind of maybe characterize it as opposed to comparing it between India
01:58and China, a lot of the talent in India is on the application side, applying technology.
02:07How does it work in a corporate setting?
02:09How does it work in society?
02:11How does it work for employees?
02:13How does it work for customers?
02:15That's why we have one of the, I would say kind of largest and perhaps one of the most
02:19progressive IT services industry, but it is very much on the application side.
02:26Yes, I will kind of admit that there's a lot more that needs to be done from a product
02:33research perspective, product innovation perspective, product orientation perspective.
02:38China has certainly kind of focused a lot more with respect to product innovation and
02:43that's kind of perhaps where the differences come in.
02:46China focusing a bit more on product innovation leading to R&D and innovation like…
02:52Which is very awesome from my lens and I'm sure from the lens of a lot of other people
02:55as well because at Davos, for example, we saw all the big Indian companies talk about
03:00use cases of AI and that's where they were concentrating their energy and here's this
03:04small unheralded Chinese research startup which essentially goes and relooks the entire
03:09model bottom up.
03:11It requires audacity, it requires ambition and there are a lot of very creative, technically
03:18sound people.
03:19Why didn't that happen?
03:20I mean by…
03:21I'm asking you this.
03:22I know you stay in the United States now, but this is where your heart I suppose is.
03:28It will happen.
03:29It is inevitable.
03:30We should not be looking at essentially just one episode, one instance of innovation, one
03:35moment in time.
03:36This is a continuum.
03:38Frankly, I'll say this.
03:39The pace of innovation with respect to what is happening in AI is unlike anything else
03:45we have ever basically imagined.
03:47We are not talking of innovation in years, we are frankly not even talking of innovation
03:52in months.
03:53We're talking of innovation in weeks and days.
03:55So just one event in one moment of time doesn't basically kind of tell the entire story and
04:00there's a lot actually that is yet to come from India.
04:03There's a lot that we experience and I will actually say this.
04:07Instead of just a big bang event like DeepSeek, we see a lot of basically incremental innovation
04:13applied out of India by Indian companies, by Indian talent in the corporate world and
04:19in businesses.
04:20So you…
04:21Can I build on this as well?
04:22Yeah.
04:23Just I think…
04:24Just to build on what Nitin said, I think Satya was in India in Jan and he said something
04:26which is fascinating.
04:27He said tokens per dollars per watt.
04:30That's the new equation for AI.
04:31Tokens is the intelligence of the output we produce out of these models.
04:34Dollars is the capex that we put in.
04:36What is the energy we consume?
04:38I think just building on Nitin's point, India is not short of ambition, audacity and capabilities.
04:42I think our opportunity, Rahul, is to get this equation right which is can we build
04:46these models, improve on these models with lesser dollars and lesser energy consumption
04:51and that's where India innovates, right?
04:52I think that's coming.
04:53I don't think it's too far.
04:54So you're saying picture abhi baaki hai, don't think of it as one big bang moment.
04:59Let's spend some time talking about what India is doing on sovereign AI.
05:03One of the big trends in artificial intelligence is sovereign AI where the debate is about
05:07whether countries should build their own large language models or should they just use the
05:14large language models because the LLMs are now commoditized and therefore do you really
05:17need to build your own?
05:19I'm sure you've all read that Minister Ashwini Vaishnava, IT Minister is now buying up a
05:24lot of compute power and essentially what we're trying to do is fuel Indian companies
05:28into building their own large language models.
05:30I'll start again with you, Nitin, where do you come out on this?
05:33The requirement, the sovereign imperative to build your own sovereign LLM models versus
05:40thinking of these models as commodities, everybody virtually has access to the same data sets
05:44and therefore do you really need to go down that route?
05:47A lot of countries similar to India are absolutely going down the path of sovereign AI and here's
05:54kind of one fundamental aspect of it.
05:58Say we manufacture physical goods.
06:02Many countries want to have a manufacturing ecosystem, perhaps not at the scale of what
06:07China has but all countries are looking to kind of figure out their own manufacturing
06:12ecosystem to produce and manufacture physical goods.
06:17What is AI?
06:18AI is about manufacturing intelligence.
06:21It's not a physical good, it's a digital good and that digital good is intelligence.
06:27What's the manufacturing capacity to actually manufacture that digital good aka intelligence?
06:35That's your sovereign AI infrastructure.
06:37If a country like India does not have its own sovereign AI infrastructure, it essentially
06:42means it does not have its own means of manufacturing or an ecosystem around it for the digital
06:50product of today which is intelligence.
06:53Do you have a different perspective because you also wear your Microsoft hat?
06:56Are you saying, leave it, we can do it, you just use our co-pilot or chat GPT and you're fine.
07:02Ashwini ji, Vaishnavi ji, why are you wasting time trying to build your own LLMs?
07:05I'm absolutely not saying that.
07:06I think what India needs to do is, I think both innovation and diffusion, right?
07:11And if you look at the history of countries that have used technology to drive development,
07:16it's not just innovation.
07:17Obviously, we got to do innovation but the power of diffusion, right?
07:20How fast can we take AI from classrooms to boardrooms in India, like literally, right?
07:24And as Microsoft, we committed to training 10 million people in India on AI in the next
07:28few years.
07:29I don't think it's enough.
07:30We got to keep doing more.
07:31But how fast do we diffuse this, right?
07:33From commerce to communities, right, just from finance to farms, every part of India,
07:38every business, every school, every institution, can we truly become an AI first nation by
07:43diffusing power, the power of AI is something that I'm excited about.
07:47But let's stay with building our own models, Nitin, because the minister said that we'll
07:51have 18,000 GPUs that we're putting together as a collective bank and then whichever company
07:56wants to build, they can take from a national collective rather than trying to buy on their
08:00own.
08:01They've got their own targets.
08:02But looking at it from your expert lens, Diloyd, how soon do you think it'll be before India
08:07can effectively produce its own large language model, DeepSeek, ChatGPT, whatever it may
08:13be, how soon do we see one made in India, they see DeepSeek?
08:17So I think we have a little bit of over-fascination with large language models.
08:22There's many of them.
08:24What actually tends to be a lot more useful is contextualized small language models, something
08:31that could be used for a particular purpose, a particular task, a particular workflow,
08:36a particular innovation, a particular kind of design.
08:39That tends to actually fuel a lot more innovation than just having one huge kind of large language
08:45model.
08:46Many companies who already do that very well, Microsoft obviously kind of being one of them,
08:50OpenAI, Anthropic, and Google, et cetera, we can sort of go down that list.
08:55But how do you actually apply them for a particular context, in a particular industry, in a particular
09:03domain, in a particular business, for a particular task?
09:07That's where a lot more innovation is perhaps going to happen in the small language space
09:14that are contextualized models for specific purposes to achieve specific outcomes.
09:19That didn't answer my question.
09:20Yes, of course, SLMs are important, and a lot of Indian companies, startups, and the
09:24big companies are focusing on them, and some of them have been very successful.
09:28But let's look at that foundational model and your perspective on the approach that
09:34India has taken at this moment, the effort that is being made, and the timeline which
09:39would be entailed in trying to see a successful outcome.
09:42A little bit difficult to predict the timeline, to be honest, I mean, because there's a lot
09:46of things that have to come in place.
09:50It's entrepreneurship that meets investment that is made available, meets computing, meets
09:58engineering prowess, meets even basically kind of the need for it.
10:01A lot of that has to come together.
10:03It's very difficult to kind of predict when a country comes up with its own sovereign
10:06LLM.
10:07Okay, so let's now move on to the second part of this conversation, and this is to do with
10:12agentic AI.
10:14And just for context, what we're seeing so far is a lot of search queries being run and
10:19responses being thrown up through generative AI.
10:22We're also seeing a lot of predictive AI, which is basically to use data to predict
10:27what may happen, and you're working, Microsoft is working a lot on agentic AI, and we know
10:32that that's something that Satya and the entire team at Microsoft is very big on.
10:36Do you want to talk us through where we're at at this moment and the trends that you
10:39think will be available publicly over the next few months and give everyone a sense
10:44of how their lives may be better because of, say, co-pilot and the other trends in agentic
10:49AI?
10:50I think, I think Nitin said this very well, right, we're entering a world of model diversity
10:55with multiple large, small, my team, I've seen teams build tiny language models now.
10:59So I think these models are getting smaller and larger at the same time, which is fascinating.
11:03If you look at the history of tech, right, last 70 years, it was all about Moore's law,
11:06which is every two years, we double performance.
11:09But with these models now and deep neural networks, we're doubling performance every
11:12six months.
11:13So that's what's going on behind the hoods, if you will, on AI, every six months, we're
11:16doubling performance.
11:17And that has three implications for all of us, everybody in the crowd today, each one
11:21of you.
11:22First, more intelligence is coming.
11:25And there was this beautiful quote by somebody who said, intelligence is a beautiful property
11:28given only to humans.
11:31I don't think that's true anymore, right?
11:33We're truly manufacturing intelligence now.
11:34And there's more intelligence, information, expertise, judgment coming at your fingertips
11:39and through co-pilots and these AI assistants.
11:42Second is more empathy is coming.
11:44Right?
11:45Technology for the sake of technology doesn't mean anything.
11:48And I can imagine a future where these co-pilots and these AI assistants will have a lot more
11:52empathy, they'll have a lot more context than you.
11:55Now, your agent will be different from my agent.
11:58And three or four or five years down the line, when somebody hires you or me or looks
12:03at you, our profile, they look at Rahul plus Rahul's agent and Puneet plus Puneet's agent
12:07and Nithin's agent.
12:08Right?
12:09So I think there's more empathy coming, more context coming.
12:12And then third is more agencies coming, right?
12:14Which is where this gets really magical.
12:16When these AI assistants start working with your permission, but not your involvement.
12:21And I think this, I genuinely believe 2025 is becoming the year of agents.
12:25And these are not the James Bond kind of agents.
12:27These are autonomous software systems that understand your context, they understand your
12:31objectives and they get stuff done for you.
12:35And I did this.
12:36My team built a custom agent for me the other day where I was taking a flight from Delhi
12:39to Bangalore.
12:40Before taking the flight, I put in a prompt in my agent saying, I have two CEO meetings.
12:44For one, I need a proposal ready with a data center proposition that we have.
12:48For the other one, there's some legal clause that needs to be reviewed.
12:51Can you please get this done?
12:52By the time I landed in Bangalore, two hours, 20 minutes, I had a prompt back.
12:57I had the document ready, the proposal ready, the presentation ready.
13:00But by the way, the co-pilot gave me advice saying, on the legal side, you need to talk
13:04to your legal team and deal desk team before you go and talk to the customer because a
13:07couple of things that are not clear.
13:09Now, this was intelligence.
13:11And by the way, it said, best of luck for the two meetings.
13:13Hope you have a good day and have a safe flight back.
13:16So this was intelligence, this was empathy, and this was agency and actions.
13:19All three are coming.
13:20And this is where I think it becomes really, really fascinating.
13:22And by the way, it's not just co-pilots which are giving you this intelligence.
13:26We're seeing vertical co-pilots being built, right?
13:28So we're building co-pilots for doctors with the Polar Hospital.
13:31It's fascinating innovation happening.
13:33For first-year doctors, can we give them the same information, same expertise, same judgment
13:37like a 20th-year doctor?
13:39And that's when this becomes really, really magical.
13:40So he's largely spoken of what they're doing at Microsoft, which is fair, that's reasonable.
13:45You give us a broader sense of what to expect from agentic AI this year and give some smart
13:50tips for people about how they can best use it.
13:53So let me kind of draw this picture.
13:57You kind of started by innovation from China, DeepSeek, et cetera.
14:01The bigger innovation is exactly kind of where Puneet was going.
14:04We have gone from gen AI telling something to doing something.
14:11That's the bigger innovation that's actually kind of happened.
14:14In the first phase of gen AI, you could have used chat GPT, you could have issued a prompt,
14:20a.k.a. ask a question, and it would tell you something, give you information, give
14:24you insights, satisfy your curiosity, whatever.
14:29Now it's about actually executing a task, executing a workflow, executing an interaction.
14:38It is doing something for you.
14:40It is not just telling you something or giving you information, which means it's suddenly
14:45become a lot more useful, because it is actually offloading some of the work that you would
14:52have done.
14:53It is offloading some of the activity that you would have actually kind of undertaken.
14:58And it is truly becoming a co-worker or a e-buddy or essentially undertaking tasks for
15:07you.
15:08So the big challenge there is of hallucination.
15:10And you'll see this tomorrow when the chiefs are here, some of the datasets that we've
15:15created have been used, have been created using the deep research feature on chat GPT.
15:20And I saw just in a few minutes, it was able to pull out what would take a serious military
15:25researcher several months to do, which is phenomenal.
15:29But the fear I had was, what if some element of this is hallucination?
15:34What if it gets fact-checked?
15:35And what if this is wrong?
15:36To what extent?
15:37And we've seen a lot of crazy hallucinations, some of it is being addressed.
15:40How much of a concern is it still?
15:42And to what extent do you think it's been dealt with?
15:45How accurate are humans?
15:48Are we kind of free of hallucination?
15:51We essentially are holding AI to a standard of 100% accuracy, as opposed to the relative
15:59standard of our essentially ability to be accurate all the time, or perhaps kind of
16:05not, and our ability to always be 100% reliable.
16:10AI is a system that learns.
16:13It learns essentially with every piece of data, every interaction, every experience,
16:18and practically learns every day, the more that you use it.
16:22The more that you use it, the more data that you feed it, it becomes better and better
16:26and better, more accurate, more accurate, and consequently, less hallucinations.
16:32So we have to basically give it time and runway for it to essentially surpass what we assume
16:40to be the human level of accuracy.
16:42Okay.
16:43By the way, we have references now, right?
16:44So I was telling you this story outside, which is before coming for this panel, I went to
16:46my co-pilot and said, I'm meeting Rahul and Nitin for this panel, Man versus Machine.
16:50By the way, the title I don't think is right.
16:51I think it should be Man and more Humans plus Machines, Man or Woman.
16:54And I said, listen, what questions should I expect?
16:57And then I said, tell me something funny or peculiar about Rahul, and tell me something
17:01interesting about Nitin.
17:02And it gave me some stories about you, about horse riding, and your first interview when
17:06you were an intern with the Prime Minister, and it gave me references.
17:10So I actually clicked on the reference to make sure, I hope I don't pick up something
17:13about you which is not true.
17:14So it's getting fact-checked and validated already now, if you're using the right tools.
17:18So let's spend some more time, Puneet, on your suggestions to everyone sitting here
17:23and watching, on what you think are the most practical, immediately doable uses of agentic
17:30AI, which you think, hey, here are some really cool things.
17:32I don't know if you know, but you better do, because this is how it can change your life.
17:36I think the first is productivity.
17:40And I'll tell you a quick story.
17:42When I joined the workforce 25 years ago, I had a dream of doing fascinating work.
17:47But if you look at our work, whatever job you do today, the drudgery of the tasks that
17:51you have to do, the grunt work, like the email inbox, it's supposed to help you, but it's
17:58become a burden now.
17:59These documents, everything that you work on.
18:01So the first liberating fact or the liberating dream that I have out of AI is, it's going
18:06to take this drudgery out of your work.
18:08I was telling Mr. Puri, two-third of my emails, I don't type anymore.
18:11I literally don't type those emails, I get the right draft.
18:15Every morning when I get up, my inbox is full of hundreds of emails, I say, listen, tell
18:17me the priority ones, tell me the important ones, tell me the urgent ones.
18:21So it's taking the drudgery out of my work away, it's bringing joy back.
18:26And I'm saving roughly 30 minutes per day on an average, I think I should be saving
18:29more as I use more of this.
18:31But imagine if every one of you gets 30 minutes back, literally 30 minutes back every day.
18:36So that's a massive productivity uplift and more than productivity, it takes the drudgery
18:40out of the bad stuff and grunt work and brings joy back at work.
18:44Second for your business is whatever you're doing, whichever company you work for, right?
18:47I think there are two sets of use cases that are emerging, Rahul.
18:49One is what I call numerator, second is denominator.
18:52Numerator is top line, right?
18:54Which is in terms of your business, what more can you do for your customers?
18:57How do you improve customer experience?
18:58How do you get more things to them and sell more and deliver more and get them a better experience?
19:05And then the denominator, how do you drive your productivity across your employee workforce?
19:08I'll give you like as Microsoft, roughly a third of our code in the last year for all
19:13our new products have been written by Generative AI by GitHub Copilot.
19:16In the last one year, we shipped more products than the last five combined.
19:20Last week I was with Cognizant Ravi and he spoke about 20% of the code that Cognizant
19:24is writing today is written by GitHub Copilots.
19:27Infosys, seven million lines of code written by machines now.
19:31So I think, so there is enough and more, right?
19:33I think the days of skepticism are gone.
19:35I think this hype cycle is gone.
19:37There's real pragmatic value out there as an individual, as a business, as an enterprise.
19:41And my only suggestion to all of you is don't wait till conventional wisdom catches up.
19:45Just go, lean in now and get started today.
19:49Let me get you to answer the same question.
19:51Your most practical tips on this is what you should be doing now because if you aren't,
19:55you're really missing out.
19:58Every single use case where you need abundance.
20:02And let me actually explain that.
20:04When I say every single use case where you need abundance.
20:08If you're a pharmaceutical company and you've come out with kind of a new therapy and as
20:14part of that, you're trying to essentially have an outreach program to the patients in
20:19terms of the right dosage, adhering to the protocol, taking kind of the medicine in a
20:25timely manner, et cetera.
20:27You essentially have a nurse practitioner, as an example, reach out.
20:32A nurse practitioner could reach out to a few of the patients.
20:36You got basically 10 nurse practitioners, a few hundred.
20:40But now if you've got an agentic based digital nurse that has essentially been trained in
20:48reaching out, you have suddenly basically kind of liberated from that equation of a
20:53single nurse practitioner reaching out to 10, 20, or 30 patients to the digital nurse
20:59reaching out to thousands.
21:01Now you have an abundant workforce.
21:03You have an abundant agentic based digital workforce that can reach out to hundreds of
21:10thousands of patients and help those patients in terms of right dosage, adherence to kind
21:15of the protocol, taking medicine in a timely manner, and you've suddenly basically improved
21:20lives and essentially kind of the clinical experience that patient is having.
21:25Every single use case where it lends itself to abundance.
21:29But let's then come to the gorilla in the room, which frankly is jobs.
21:32Because if Puneet can get half the work on emails done through an agent, he doesn't need
21:38an EA.
21:39If all these IT companies can write the code through co-pilot, you need fewer coders.
21:45And I am 100% convinced that the likes, I don't want to name anyone, but the big consultancies,
21:50the big tech companies don't tell us in public the true impact on jobs, that the true impact
21:56is far deeper than the likes of, Nitin would admit, in public.
22:01How do you respond to that?
22:03So here's what I would say.
22:04I'll give our kind of personal experience.
22:07When we undertake particularly what is called software engineering, aka coding work for
22:13our clients, we inevitably use Gen AI.
22:17Now we all have kind of different experiences in terms of what's the productivity and the
22:21efficiency of using those Gen AI tools.
22:23Puneet kind of went there a bit.
22:25We see anywhere from 30 to 50%.
22:28Has that reduced our workforce by 30 to 50%?
22:32Absolutely not.
22:33In fact, it's actually increasing our workforce.
22:35Because what's actually happening is there's a lot many more clients that we're serving.
22:40We're not starting.
22:42Our clients are growing.
22:43We're figuring out new ways of actually kind of introducing new products, new ways of penetrating
22:49the marketplace, new ways of actually going to the top line.
22:51But it's only a matter of time.
22:52If a lot of the code is written by agents and through AI, over time you need fewer coders.
22:59And similarly with nurses, for example, if some of the work is outsourced to these agentic
23:03nurses, you need fewer physical people.
23:06I mean, that's just reality.
23:07So Rahul, let me kind of paint this picture.
23:10Here's the other reality, which we should perhaps appreciate.
23:13PC revolution, 80s and 90s.
23:17Every job associated with a typist, a clerk, a bookkeeper, vanished, gone.
23:24Did that reduce the workforce in the economy?
23:26Absolutely not.
23:27What happened to them?
23:28They became software engineers, network administrators, data scientists, data analysts, et cetera.
23:35New jobs, reimagining jobs, new ways of actually economic productivity.
23:39That's exactly what's going to happen with today's world.
23:41But Puneet, it's not that simple.
23:42Because there are people, say, in the middle of their careers, who are used to a certain
23:46kind of work.
23:47And then suddenly, like, middle America got hollowed out.
23:49We've got a lot of people, because manufacturing moved internationally, who simply then don't
23:52have work.
23:53And that, in the context of a country like India, with what is supposed to be a demographic
23:57boom, runs the prospect of turning into a massive demographic disaster.
24:03I don't think I'll go to it.
24:04First, I would just say our jobs are already changing on us while we're not changing jobs.
24:09And I'll give you a quick example.
24:10I looked at some LinkedIn data.
24:11It said, for the same job today, 25% of the skills required for that job have changed
24:15in the last couple of years.
24:17So the job I was hired for in the last couple of years, 25% of the skills required have
24:21already changed.
24:22And potentially, I'm outdated.
24:23I don't know.
24:24But in the next few years, it'll be 65%.
24:25So our jobs are already changing on us, even if we're not changing jobs.
24:27So that's the first thing.
24:28Second, you're right, Rahul.
24:29I think jobs will change.
24:30A new set of jobs will be created.
24:33And there's fascinating work happening in India, CARIA, which is a foundation in Mumbai
24:36I recently met.
24:37They're creating data sets for large language models and going to Dharavi and villages and
24:41getting women and villagers to work for us and create some fascinating new jobs.
24:45So new jobs will get created.
24:47Our jobs will shift.
24:49I think we all have to realize that our jobs will shift.
24:52And again, I always tell my teams, you will not lose your job to an AI, but you will lose
24:57your job to somebody who can use AI really well.
25:00And I think the imperative for India, Rahul, is skilling, which is, and I spoke about 10
25:05million people at Microsoft to be trained on AI in the next few years, each one of us
25:09needs to learn.
25:10Like, I spend an hour every day learning about this, and I'm way behind.
25:13So if you're not learning about this, if you're not adopting this technology, if you're waiting
25:17for conventional wisdom to catch up, you will fall behind.
25:20So new jobs to be created, jobs will shift.
25:23But I think the call of the hour is skilling and learning.
25:25You know, so rebrain or rot, right?
25:27So we have a couple of minutes left.
25:29I want to spend this on the third dimension, which is edge AI or physical AI, which is
25:34basically machines with some kind of artificial intelligence doing the work that workers would
25:40have done, for example, in robotic manufacturing.
25:43Now there again, it comes back to the question, which I think is being underplayed about the
25:48impact and joy of a lot of vehicle manufacturing can happen.
25:50I was listening to Howard Lutnick, the US Commerce Secretary, speak, and he'll be here
25:55in a short while from now, talk about how you can come back and invest in India, in
26:00America, where labor costs are higher because you now have robots that can do the work which
26:05expensive American labor can't.
26:06So that's the idea, which essentially comes back to how many people you need to work on
26:11the shop floor.
26:13Who's manufacturing the robots?
26:15Who's actually kind of building the IT systems for the robots?
26:18Who's integrating those robots into the actual warehouse?
26:22And who is monitoring those robots for the quality of work and the accuracy of work?
26:27New jobs, new economic basically streams, and new ways of actually kind of being a productor.
26:34Yes, the job in the actual warehouse where the robot is applied may actually diminish
26:41or disappear, sure.
26:43And a warehouse that may have had a hundred basically kind of humans in there may only
26:47have 10, but the other 90 are not basically sitting at home idle.
26:52They're going into different fields.
26:54What will kill jobs is complacency.
26:57If we actually adapt and we focus on what is required to manufacture those robots, make
27:02them essentially accurate, test them, integrate them, and assure the quality and govern it,
27:08that's where the jobs are.
27:09So we're out of time, but I have to ask you this question about general intelligence machines
27:14developing a mind of their own and potentially going out of control.
27:20And now that's the doomsday scenario, right?
27:21And I want Nitin, you to come in first on this.
27:24Why I'm not asking him the tough questions is because he has to go back to work on Monday,
27:27right?
27:28We don't want to land him in any trouble.
27:29Nitin is the consultant at Deloitte.
27:30Keep it going that side.
27:31Keep it going that side.
27:32He can say what he can wax on everything.
27:33So talk about this fear that people have of machines going out of control and the risks
27:39that poses.
27:40How real is that fear?
27:41A lot of that fear gets fed by Hollywood and Bollywood movies.
27:46As a student of history, I would say that as humanity, we have probably kind of created
27:49more doomsday scenarios and doomsday events than we could even sort of contemplate.
27:54Yes, there's always the possibility, but I would kind of place my faith in humanity,
28:00which is we always find a way to manage, govern, control, improve our lives, and we will adapt
28:08and make sure that things don't go out of way.
28:10Given the rapid pace of change, are you concerned that these machines become so intelligent
28:13by talking to each other, by making each other more intelligent, by routing data off each
28:18other's servers, that you don't really know what's going on?
28:21Maybe they're not in some time in Satya or your control.
28:23I don't think…
28:24Listen, I think this lump of labor fallacy that we spoke about, this man versus machine
28:28fallacy we're speaking about, these machines are supposed to amplify human ambition.
28:33They're supposed to empower us.
28:35Everything that we've spoken about, Rahul, is all about getting energy back, getting
28:38joy back, getting time back to do more work.
28:41I think that's what I'm excited about.
28:42I don't worry about that personally.
28:43You know, this conversation about AI is so deeply ingrained in all of what we do.
28:48I can go on for hours and it would still be as fascinating.
28:51But we've got a very packed schedule and I deeply thank Nitin and Puneet for joining
28:55us here.
28:56They're still here for a while.
28:57So, if you have any questions…
28:58Nitin charges a lot for consultancy, but he's here.
29:01So, if you have anything about AI that you're worried about, you can go catch him during
29:05the course of this evening.
29:06And that's the magic of the Conclave.
29:08Thank you very much for joining us.
29:09Absolute pleasure.
29:11Thank you, Rahul.