- 2 days ago
In a wide-ranging discussion on the race for superintelligence, Professor P. Murali Doraiswamy stated that in the future, AI won’t replace doctors, but doctors who use AI will replace doctors who don’t use it. The session at India Today Conclave Mumbai 2025 explored the rapid advancements in artificial intelligence, from AlphaGo defeating human champions to AI now composing poetry and writing articles.
Category
🗞
NewsTranscript
00:00Good afternoon and welcome to this session, which will be, I think, one of the most urgent
00:06conversations of our age, something we all should be caring about, and of course the
00:12session is titled Human Mind vs. Machine Mind, the Race for Superintelligence.
00:19The human brain is seen and is regarded as the crown jewel of our evolution.
00:27Our brain, they say, is an organ of 86 billion neurons that help us fuse our feelings, perception,
00:37memory, imagination, and morality into what they call a stream of consciousness that,
00:45for centuries, has shaped our culture and civilization.
00:50In doing so, we have become the dominant species of our planet.
00:54Yet, and yet, today our own machines are beginning to challenge human supremacy.
01:04Artificial intelligence, or AI, can already outpace us in language, science, and problem solving.
01:12So, let's pose the big questions.
01:14Just how close are we to truly understanding the mind and replicating the brain functions?
01:23And will this race for superintelligence, or the moment, that big moment, when machines outthink humans,
01:31will it redefine consciousness and the very meaning of being human in the 21st century?
01:38To help us untangle these complexities, I am delighted to welcome Professor P. Murali Dharaiswamy,
01:46one of the world's leading researchers at Duke University.
01:50Let's give him a very warm round of applause.
01:53His pioneering work bridges neuroscience, brain health, and human cognition.
01:58Over to you, Professor Dharaiswamy.
02:14Thank you, Raj.
02:16It's a great honor and pleasure to be here, and good afternoon, ladies and gentlemen.
02:20I think Raj very eloquently stated how important the brain is.
02:30I think the picture here shows you some of the greatest inventions, some of the greatest artistic creations that the human brain has produced.
02:41You know, we can contemplate the stars, we can compose ragas, we can invent things from the light bulb to the internet to the smartphone.
02:51But the fundamental question that is being posed today is, is this all a product of our genes, of biology, and all of our social interactions and environment?
03:02Or is this simply code?
03:05And can it be replicated by code?
03:08And that, I think, is the mirror that is in front of us that we are facing today.
03:13In fact, Thomas Edison, who is the founder of the Second Industrial Revolution, famously quoted,
03:19the main purpose of the human body is to carry the brain around.
03:23In fact, today, in the fifth industrial revolution, we can rephrase it to say the main purpose of the human body may be to carry an AI around.
03:32So this, as the previous panel discussed, this graph charts the explosive rise of AI over the last 75 years.
03:41So AI really is 75 years old.
03:44There have been many waves, about five, six waves.
03:46This is the biggest wave yet, because trillion-plus dollars are being invested.
03:52By comparison, the human IQ has only risen by about 10 to 12 points over the last 100 years,
03:59and the prediction is that it will rise only about four points over the next 30 to 40 years.
04:06And some of the major benchmarks that AI has beaten, I think, started first with AlphaGo, as you can see here.
04:14AlphaGo was considered one of the most difficult board games,
04:19and the prediction at that time was it would take a computer 15 years to beat it,
04:23until a spin-off in London called DeepMind developed an AI program based on the brain called Neural Networks and Deep Learning.
04:33And they were able to not only beat AlphaGo, the number one reigning champion of AlphaGo,
04:38but the subsequent iterations of AlphaGo did not even need to be told the rules of the game.
04:45It could learn the rules of the game by itself, and it beat the champion 100 times to zero,
04:50to the point where the champion just retired and said,
04:52there is no hope for a professional player like me.
04:55I will never beat a machine again.
04:57So the next landmark occurred with the development of something called Transformers.
05:02I won't get into the detail.
05:03And then DeepMind again produced an iteration of AlphaGo called AlphaFold,
05:09perhaps the greatest application of AI to human health ever.
05:15The human body has maybe 100 million proteins or more, which occur in three-dimensional shapes.
05:22And understanding how these proteins fold or misfold in diseases is very crucial to finding new treatments
05:30for the hundreds and hundreds of diseases that impact human suffering.
05:35It was predicted that humans in our lifetime would never be able to uncover the folding of these proteins,
05:43because a typical PhD student doing a PhD for five years would maybe uncover the structure of a single protein.
05:48AlphaFold, within a couple of years, came up with a program that was able to predict the 3D structure of proteins
05:56to the point where the newest iteration of AlphaFold has been used in 90 countries
06:02by hundreds of thousands of researchers to uncover the structures of 200 million proteins.
06:08And many, many drugs are already in late-stage development based on those structures.
06:14And now the newest iteration of AlphaFold is publicly available,
06:18and you can use it to predict the structures of not only proteins,
06:21but a variety of other cellular structures in the human body.
06:26The next major breakthrough was in language, which was GPT and chat GPT.
06:32So with chat GPT for the first time, AI started feeling very, very human.
06:37Chat GPT can now compose poetry that brings us to tears.
06:40I recently was asked to write a magazine for an article for Time magazine on AI,
06:46and my co-author first generated a version of the article using chat GPT,
06:50and then I wrote an article, and we sent both to the editor.
06:54The editor actually preferred the one that was written by chat GPT without their knowing it.
06:58So now it can write articles better than the best trained human.
07:03And then, so where are we headed next?
07:09With this massive investment, AI is going to be everywhere.
07:14AI is going to be in every machine.
07:16It's going to be in our body, and it possibly will eventually also be in our brains.
07:21The next big benchmark that everybody thinks will achieve within three years or so,
07:27according to Silicon Valley, is called artificial general intelligence.
07:31Artificial general intelligence is when a computer is good, not just in a narrow field,
07:37like playing chess or playing a board game, or like answering some, you know,
07:41interesting question or drawing a cat face, but it will surpass humans across multiple abilities
07:47and to the point where you will not even be able to tell whether it was done by a human or a computer.
07:54The so-called Turing test.
07:56The Turing test is when you cannot tell the difference between a computer and a human in its output.
08:02In fact, the prediction is that when we reach AGI, the computers will be so good,
08:07it will be very easy to tell that it was done by a human because, by a computer,
08:11because no human probably could have ever done it so well.
08:14And then the prediction is in the next 15 to 20 years, maybe by 2050,
08:19we will reach a state called artificial superintelligence, where AI will surpass humans in every single field,
08:27whether it be physics or biology or poetry or music or filmmaking or journal editing,
08:34to the point where it will be able to instantly solve every unsolved problem in humanity.
08:40Now, of course, the biggest threat is that once we get to that point, we may not be able to control it.
08:46Rogue actors could weaponize it or we would become slaves or we would become so dependent on it
08:53that it would be called cognitive offloading.
08:55We would completely become dumb.
08:57So those are some of the biggest.
08:59Those are still theoretical risks that we should guard against.
09:02Now, in my own field, I'm a doctor.
09:07Most of my time is spent with patients, but even I, I'm starting to think of AI almost every day.
09:13Not just front patient facing, but behind the scenes in terms of charting, in terms of drug discovery,
09:22in terms of better prediction of disease.
09:24And so these are some of the major milestones that have been achieved in medical AI.
09:30At least in the U.S. now, there are at least 1,500 AI-based clinical decision support tools
09:37that have been approved by the FDA.
09:39And a number of tools are already cleared in India as well,
09:42and Apollo Hospitals is an example of one of the hospitals.
09:45You now have on its website something called Ask Apollo, which is an AI-powered chatbot,
09:51works in about a dozen languages.
09:54It can not only field your questions.
09:56It's back clinical intelligence engine will automatically assign a differential diagnosis.
10:01It will tell you which is the best consultant and what kind of tests are recommended for you.
10:07And this will continue to get better and better and better.
10:10So Microsoft just unveiled something last month, a multi-agent tool.
10:16So it's not just a narrow AI.
10:17It's an AI that works across multiple areas called Medical Orchestrator.
10:22So it's like a Zubin Mehta that coordinates a full orchestra.
10:26They gave it 320 of the most difficult medical diagnoses where a doctor only has a 20% accuracy rate.
10:35And this particular AI agent was able to diagnose those cases with an 80% accuracy rate,
10:41so four times better than the best consultants in the field.
10:45And this tool is expected to be rolled out pretty soon,
10:48at least initially across the U.S. and then the rest of the world.
10:52There's another paper that just came out in a journal called Nature,
10:55where a group of researchers developed an AI algorithm that can predict your risk for 1,000 diseases.
11:03You can take a 25-year-old, give the person the history, the family history,
11:08a little bit of the person's background.
11:10It can predict that risk.
11:12It's unbelievable what is happening.
11:14Just imagine the power if you have this tool in your smartphone,
11:18what kinds of health behavior changes you can make to reduce your risk for future disease.
11:24Now, of course, AI can be super smart, but AI can also be super dumb.
11:29It can hallucinate.
11:31It can make catastrophic errors.
11:33It can have biases.
11:35It can, of course, disclose your information and give you stupid information,
11:38like the chat user with mental illness who was asked to commit suicide that many of you may have read about.
11:44So clearly there are many, many areas where the human brain is still far better than AI,
11:50but AI is catching up very soon.
11:52And this slide lists some of the areas,
11:55and certainly as AI begins to automate workflow and job losses occur,
12:00these are some of the skill sets that are going to be in demand.
12:03And ultimately, it's not going to be about AI replacing humans,
12:07but about the two, AI and human, working together, keeping human in the loop,
12:12something we call as intelligence augmentation.
12:15So coming to my area of work in my lab,
12:20I'm focusing on how can we enhance the longevity of the 8 billion computers inside our brains.
12:26And it's very interesting because the biggest threat to this cognitive revolution
12:31is neurological and psychiatric diseases.
12:34And the exact cost of these illnesses to society in the next few years is predicted to be $1 trillion,
12:40the same amount we're spending on AI.
12:46So one of the biggest threats to our brain health is a condition called dementia.
12:51There's something like 50 million cases of dementia worldwide,
12:55and about 150 to 200 million people are at risk.
12:58As this slide points out, there's a case of dementia that occurs every three seconds.
13:03And the two biggest causes of dementia are called vascular dementia and Alzheimer's dementia.
13:10So this is a brain, and many of you may be shocked to know this is the vasculature of the brain.
13:18This is a little bit of a graphic image.
13:19We have hundreds and hundreds of miles of blood vessels,
13:22including very, very tiny end arterioles that have no collaterals.
13:26So if there's a blockage in those blood vessels, there's a critical area of the brain that's damaged.
13:30One millimeter of the brain has something like 10,000 to 30,000 nerve cells and millions of connections.
13:37So it's crucial to keep your blood vessels patent and the blood flowing.
13:41So any kind of blockages from hypertension, high cholesterol level, lack of exercise,
13:46eating too much fatty food, it's obviously bad.
13:49And then silent strokes and blockages start building up in your brain at the age of 45 or so.
13:54You throw about 200 clots into the brain on an average day, a lot more if you have irregular heart rate.
14:02Now the Apple Watch, even the Apple earbuds now can detect irregular heart rate,
14:07and that's something very crucial, a condition called atrial fibrillation
14:10that throws a lot of blood clots into the brain and is a leading cause of strokes.
14:15So you can see some of the white areas are called patchy white areas.
14:19These are silent strokes that you're having in the brain without your even knowing you're having it,
14:23but it just reduces your memory and your speed of processing by 5-10%.
14:28So this is Alzheimer's disease.
14:32The left side is a normal brain.
14:34The right side is an Alzheimer's brain.
14:36You can see there is shrinkage and loss of brain tissue.
14:39That's why it's called a neurodegenerative disease.
14:42Now Alzheimer's disease starts something like 20 to 25 years before you start having memory problems.
14:51It's just like heart attack.
14:52It doesn't start the day that you go into the emergency room.
14:56Your lifestyle, junk building up in your arteries, that's what leads to a heart attack.
14:59The same thing with Alzheimer's.
15:01We now know that it's a gradual process that can start in your late 40s and 50s.
15:05So this picture depicts normal nerve cells, and you can see one of the abnormalities that builds up is called amyloid.
15:14It's a protein junk that builds up.
15:16It clogs your nerve cells, and it's a misfolded protein.
15:20Again, using AI, we can now decipher the structures of these kinds of proteins and develop exact targets and drugs.
15:27The second abnormality that builds up in the Alzheimer's brain is a protein called tau, and you can see tau is the green stuff that accumulates inside the nerve cells.
15:37It collapses the nerve cells, and it causes damage from the inside.
15:41So you have two forces, one a plaque accumulating outside, causing damage, and two tangles inside the nerve cells, causing damage.
15:51Now sleep, it turns out, is very crucial.
15:53Now what you're seeing here on the left side is an asleep brain, and on the right side is an awake brain.
16:00You may be wondering, did I mix the two slides up?
16:03No.
16:04It turns out that when you're asleep, there is a new system that comes alive in your brain that was only discovered about 10 years ago.
16:12And this system is called the glymphatic system.
16:15It's like the brain's recycling system.
16:17It takes out all the trash, it takes out all the toxic products that have accumulated in your brain, and it gets rid of them through your lymphatic system and your blood supply.
16:25So adequate sleep is crucial.
16:28If you have sleep apnea, which can now also be detected by a watch, by a variety of sensors, about 30-40% of people probably in this room have mild sleep apnea, and they don't know it.
16:38If you have a slight punch, there's a very high likelihood you have sleep apnea.
16:43So again, make sure that if you have sleep apnea, it's detected, because you want to have a restful sleep to clear toxins out.
16:51So we now have amazing tools to map the changes that are happening in the brain.
16:58There is a tool called PET scans, which we were part of developing at Duke, called amyloid PET scan, that can map the buildup of these plaques in the brain.
17:07We also have another tool called tau PET scans.
17:11So these are like an MRI scan, but it's like a scan that involves injecting a radioactive dye into the brain, and it can map that.
17:19We now also have blood tests.
17:21Just like in cancer, you have something called as a liquid biopsy.
17:25We now have simple blood tests that can measure how much plaque buildup there is in the brain.
17:31It can also measure how healthy your nerve cells are.
17:34So these tests are now going to be crucial for monitoring you during your annual physicals, especially if you have any risk factors such as high blood pressure, cholesterol, family history, etc.
17:49There are now a variety of drugs in development and vaccines.
17:52There are two drugs that are approved to clear these plaques in the brain.
17:56These are the first generation drugs, and there are newer drugs that are going to be coming out very soon.
18:01So the field looks very promising, and I think we can expect to have some very, very rich therapeutic inventory very soon.
18:13What about lifestyle?
18:15Dementia is about 40% lifestyle determined.
18:19It's about 60% genetic.
18:21So some of the lifestyle factors that you can are listed in this slide.
18:26You can see sleep, as I mentioned.
18:29There's a U-shaped curve again.
18:31As you sleep optimally between six to eight hours, you have peak cognitive performance.
18:36If you sleep too much, your cognitive performance goes down.
18:40Again, aerobic exercise.
18:41You can see, you know, no exercise is bad.
18:44You can exercise three to four days.
18:46It's really good.
18:46If you exercise too much, especially if you're someone running marathons, ultramarathons, that's bad for the brain again.
18:55Alcohol, you can see the very steep curve.
18:58A couple of drinks is okay.
19:00Too much deteriorates, so watch out tonight.
19:02So with AI tools and using your risk factors, we can now compute a brain age for each individual person and compare it to your chronological age.
19:16And we can see the gap, and we can give you a brain number that hopefully motivates you to lead a better lifestyle.
19:21And there are a variety of multi-domain interventions we can use to actually bring your brain age back by five years or so.
19:29So that is the future of brain health.
19:34And one of my goals in my research is to democratize brain health.
19:38We don't want to just study white brains.
19:40We want to make brain research universal.
19:44And now with Internet, we can do that.
19:46So we are creating a global neural lab.
19:49And we already have 51 million people from 160 countries contributing voluntarily data through this portal.
19:57And so we are creating the world's largest living lab of brain health, through which we hope to gather new insights to help the next generation and the present generation.
20:07So what's coming in the future?
20:10You know, Silicon Valley is thinking way upstream.
20:13Of course, the first we are going to start with non-invasive devices.
20:17Mark Zuckerberg is thinking about meta, the glasses.
20:20So these are one way.
20:21They provide, you know, information from the outside world to the brain.
20:26The next generation of devices is called neuromodulation.
20:30Could we selectively, non-invasively stimulate different parts of the brain through lights, through very mild currents, which are not dangerous, that can stimulate specific regions of the brain and enhance our cognitive abilities.
20:44So the military will likely use this.
20:46Patients with diseases will likely use this.
20:49But eventually it may come for available to the general population.
20:52Ultimately, of course, you know what Elon Musk and some of the other pioneers are thinking about is a chip in the brain that does not damage the brain, that can read and write bi-directionally, can receive input, close loop, sort of give the right signals back to the brain and enhance the brain when it's needed or rest the brain when enhancement is not needed.
21:17And again, that's also in early clinical trials.
21:19There are many risks and many hurdles to be overcome.
21:22But that's where we're headed.
21:24So I think I want to stop with this.
21:26And really, when you think about it in the AI revolution, the brain is our most important asset.
21:32It's important for us to live by design.
21:35It's important also for us to design AI that enhances human capabilities and works well with human wisdom.
21:41So I just want to thank you.
21:52Let's really give him a very, very warm round of applause for that.
21:56Brilliant presentation.
21:58And we just got a couple of minutes and we'll some ABC questions while we're putting up this, which is I think everyone's asking this, right?
22:06And those who have teenagers or, you know, maybe 20s, will AI replace human friends?
22:15I've seen that happening among the youngsters.
22:17Will that happen?
22:18And I will ask you the most, I think, provocative question in some senses is, will it replace human sexual relationships?
22:28I think AI is already replacing human friendships because, for starters, one, youngsters spend a lot of time.
22:39You know, there are many, many people who are lonely, especially elderly, and loneliness is a huge medical condition.
22:46You know, loneliness leads to depression, leads to emergency room visits, leads to all kinds of issues.
22:51And so we've been testing a companion robot, and we actually developed for the first time a rating scale called the Cobot Scale, companion bot impact scale, which measures the impact of a companion robot on nine dimensions of well-being, including happiness, including, you know, depression, sleep, etc.
23:11So we have had some users that have had, you know, 3,000 cups of coffee, and they just chat with this robot as though it's normal because they have nobody else to chat to.
23:21So it could be, you know, a widow or a widow or living alone at home, somewhere isolated.
23:25So for those kinds of people, I think it's amazing.
23:28Now, one of the things I'm concerned about is OpenAI has just announced an alliance with Mattel, the toy maker, to introduce large language models in toys for children.
23:38And so we don't know the impact that it's going to have on the developing brain, which is when attachments are formed.
23:44So that's the area that I'm a little bit uncertain about.
23:46And sexual relationships?
23:48Sexual relationships?
23:50You know, I haven't studied that area, so I can't, I'm not an expert on it.
23:54You're ducking this question.
23:58I won't let you go.
24:00You have to answer it.
24:01Go ahead.
24:01Try it.
24:04Look, I think it's coming.
24:08The sexual industry has always been at the forefront of the technology revolution, whether we like it or not.
24:15And I don't know the way it's going to progress, but there already are, obviously, you know, virtual replicants.
24:25You can have virtual girlfriends.
24:27You know, the most interesting quote I heard was from the CEO of Bumble, a dating agency.
24:33And she said, the future of dating is my AI will date your AI, and then we'll decide whether we actually go out for dinner.
24:40Okay, let's come to the other big question.
24:46How much of job loss would happen as a result of AI?
24:50I mean, that's one of the big concerns.
24:52And for instance, will it replace doctors?
24:54You talked a lot about how medical science has been progressing through AI.
24:58Would you lose your job?
25:00So I think the answer to that is this well-known quote everyone has had.
25:04You know, AI won't replace doctors, but doctors who use AI will replace doctors who don't use AI.
25:09Ouch.
25:10Okay.
25:11So it's just like anything else.
25:12You know, if a doctor doesn't have an x-ray or a doctor doesn't have a stethoscope, you would rather go to the one who has the latest tools.
25:18And to be honest, the AI is far more sensitive.
25:22The AI is going to pick up a lot of subtle things a doctor may miss, so I would feel much more confident.
25:27So we have done studies showing if you take an average MRI scan and put it through an AI, the AI picks up far more lesions, subtle lesions, than a doctor's human eye can.
25:37So ultimately, it's going to be human plus AI, which will provide the best diagnosis and care plan.
25:43Okay.
25:44Your job is safe or not?
25:45I think my job is safe because, first of all, I'm also involved in the research.
25:49So it's like that old saying, right?
25:50If you're not on the table, then you're on the menu.
25:52So I'm at the table.
25:55Great one.
25:57And just a little tip for us, right?
26:00What can we do to keep our brains in top shape so that we can compete with the AI and be ahead of AI?
26:08Well, I think it goes back to this first, my original quote.
26:11I think I also laid out a slide with all of the critical abilities, you know, whether it be reasoning, whether it be leadership, whether it be empathy, soft skills, all of those are going to be critical in the new era.
26:25Of course, working with AI is going to be crucial, AI literacy, whether it be coding, whether it be learning how to protect biases, whether it be asking the right question of AI, or whether it be, you know, building guardrails.
26:41So those kinds of skill sets are very important, but I firmly believe that the users should be the ones at the front line leading the AI revolution rather than the coders, because otherwise we're going to be the ones coded.
26:55Okay, I'm going quickly because we've got very little time, but one of the things that defines human beings is feeling, right?
27:03And consciousness.
27:04So I don't have an answer to that, so I'll flip the question.
27:12When does a baby become conscious and how does a baby become conscious?
27:16I've been wondering this.
27:17We know sometime during pregnancy the baby becomes conscious.
27:20We still have disagreement of when and how.
27:23Is there a critical brain mass that's needed?
27:25Scientists think there's a projection going from the thalamus to the cortex that determines consciousness, etc., etc.
27:30Neuroscientists can't agree when someone is brain dead or whether someone is still alive.
27:36So I think it's possible a machine might become conscious, but if it does, it will be accidental.
27:41I don't think we have the exact formula for creating consciousness yet.
27:45And finally, there's 55 seconds left, which is a huge question to answer, but atrophy.
27:52All of AI is what happened to humans in the past.
27:55It picks up that, it's able to then act on that.
27:58But as we know, our lives are unpredictable, and you cannot really predict the future.
28:04So where does AI come into the picture then?
28:06Because it can build models of the past and tell you what's happened and all these lesions that you talked about,
28:11but life is uncertain.
28:14And that's the surprise of humans.
28:18True, but just like humans can imagine the future,
28:21I think AI models can be built to imagine the future and predict the future.
28:25So, for example, AI models are predicting the trajectory of my life.
28:31You know, I put myself through the AI, it has projected my medical history over the next 20 years.
28:35So AI is capable of predicting the future and projecting what happens in the future.
28:40Is that what you're asking?
28:41In some senses, hallucinating, right?
28:43We use the word hallucinate for AI, but in a worse sense because there is no future that, you know, we can predict ourselves.
28:54Yeah, yeah, no, the AI will make it up simply.
28:55So we have to make sure it's not making up stuff.
28:58And so when does that point reach when you think AI will equal the human brain and actually cross it,
29:04and then when do we become master and slave?
29:07So AI has already crossed the human brain in multiple areas.
29:09Like I said, alpha, fold, all these areas, right?
29:12So the question is, can it master the human brain with zero-shot learning?
29:17Like when a baby bumps into a table, it instantly learns,
29:20ouch, it's causing pain and I need to avoid these kinds of objects.
29:23And it generalizes it to other objects called transfer learning.
29:26I don't think we're there yet.
29:27That zero-shot learning and transfer learning, once it hits that, then we need to be scared.
29:31We have plenty of questions.
29:32Our time is up.
29:33Thank you very much, Murali Durarswamy.
29:35Let's give him a very, very warm round of applause for this.
Recommended
0:55
|
Up next
10:21
8:09
0:56
9:00
1:48
9:41
9:39
1:16
10:01
1:04
Be the first to comment