La curiosidad de Joelle Pineau la llevó a realizar un doctorado en ingeniería con especialización en robótica, que ella describe como su "puerta de entrada a la IA".
Como vicepresidenta de investigación de IA en Meta, lidera un equipo comprometido con la apertura al servicio de la investigación de alta calidad, el desarrollo responsable de la IA y la contribución de la comunidad.
Como vicepresidenta de investigación de IA en Meta, lidera un equipo comprometido con la apertura al servicio de la investigación de alta calidad, el desarrollo responsable de la IA y la contribución de la comunidad.
Category
📚
AprendizajeTranscripción
00:00Hi, Sam here to tell you how you can unlock the transformative power of generative AI
00:07with a new online course from MIT Sloan Executive Education.
00:12You may be wondering what GenAI is and why it's relevant for your business.
00:16In this on-demand online course, led by experts from MIT Sloan and the Schwarzman College
00:21of Computing, you'll explore GenAI's promises and limitations, investigate its applications
00:27for your business, and learn how you can implement a strategy for it.
00:31Visit executive.mit.edu slash smrai to sign up for the course and gain the MIT edge today.
00:40That's executive.mit.edu slash smrai.
00:58Why is being open about your company's AI research a benefit more than a risk?
01:04Find out on today's episode.
01:06I'm Joelle Pinot from Meta and you're listening to Me, Myself & AI.
01:12Welcome to Me, Myself & AI, a podcast on artificial intelligence and business.
01:17Each episode, we introduce you to someone innovating with AI.
01:21I'm Sam Ransbotham, professor of analytics at Boston College.
01:25I'm also the AI and business strategy guest editor at MIT Sloan Management Review.
01:31And I'm Shervin Korubande, senior partner with BCG and one of the leaders of our AI business.
01:37Together, MIT SMR and BCG have been researching and publishing on AI since 2017, interviewing
01:44hundreds of practitioners and surveying thousands of companies on what it takes to build and
01:50to deploy and scale AI capabilities and really transform the way organizations operate.
01:57Hi everyone.
01:58Today, Sam and I are happy to be joined by Joelle Pinot, vice president of AI research
02:04at Meta.
02:05Joelle, thanks for speaking with us today.
02:08Hello.
02:09Okay, let's jump in.
02:11A good place to start might be for you to tell us about Meta.
02:14A lot of people know what Meta is, but maybe you could describe it in your own words and
02:19also your role in the company.
02:22Well, as many people know, Meta is in the business of offering various ways for people
02:28to connect and build community, build connections, whether it's through Facebook, WhatsApp, Instagram,
02:34Messenger.
02:35We have billions of people around the world using our products.
02:38I've been at Meta for about seven years now, leading AI research teams.
02:43I'm based in Montreal, Canada, and now I lead FAIR, which is a fundamental AI research team
02:49across our labs in the US, in Europe.
02:51The role of our group is actually to build next generation AI systems and models, discover
02:57the new technology that will eventually make the products better, more engaging, safer
03:03as well.
03:04That's a great overview.
03:05Can you give us a sense of what some of those projects are that you're excited about or
03:09you're working on?
03:10You don't have to give us any secrets, of course, but what are some fun things you're
03:13excited about?
03:14Well, I hope we have a chance to talk about it, but there's not a ton of secrets because,
03:19in fact, most of the work that we do is all out in the open.
03:22We adhere strongly to open science principles.
03:25We publish our work.
03:26We share models, code libraries, and so on and so forth.
03:29Our teams cover the full spectrum of AI open problems.
03:34So I have some teams who are working on understanding images and videos, building foundation models,
03:40so core models that represent visual information.
03:44I have some teams that are working on language models, so understanding text, written, spoken
03:50language as well.
03:51I have some teams doing robotics, so understanding how AI systems move in the physical world,
03:57how they understand objects, people, interactions, and a big team of people who are working on
04:04core principles of intelligence.
04:07So how do we form memories?
04:09How do we actually build relationship between different concepts and ontology of knowledge
04:15and so on and so forth?
04:17It seems like there's almost nothing within artificial intelligence you're not working
04:20on there.
04:21Tell us a bit about why you think open is important.
04:25So FAIR has been committed to open research for 10 years now, since day one.
04:30We've really pushed on this because whenever you start a project from the point of view
04:36of making it open, it really puts a very high bar in terms of the quality of the work
04:41as well as in terms of the responsibility of the work.
04:44And so when we decide what algorithms to build, what data sets to use, how to evaluate our
04:51data, how to evaluate the performance of our model through benchmarks, when we know that
04:56all of that work is going to be open for the world to scrutinize, it really pushes us to
05:01put a very, very high bar on the quality of that work, on the rigor of that work,
05:05and also on the aspects of responsibility, safety, privacy, and so on.
05:11The other reason I think that open is really helpful is a lot of researchers come from
05:15a tradition of science where you're always building on the work of others.
05:21Science does not happen in a silo.
05:24And so if you're building on the work of others, there's also a desire to contribute back to
05:27that community.
05:28And so researchers are incredibly interested in having that kind of a culture.
05:32So it helps us recruit the best researchers, keep them.
05:35It is quite different from how other industry labs operate.
05:40And so from that point of view, I think it's definitely a big advantage.
05:44What's interesting is from the point of view of meta, there's no concern in terms of keeping
05:49some of that research internal in the sense that it doesn't in any way stop us from using
05:54this research into our products.
05:56Not because we've published the results that we can't use it into our product.
06:00Really the power of the product comes from all the people using it.
06:03It doesn't come from having a secret sauce about AI.
06:06And we know how fast AI is moving.
06:08A lot of that knowledge is disseminated across the community very, very quickly.
06:13And we are happy to contribute to that.
06:15That makes a lot of sense.
06:16A lot of my background is in computer security.
06:19And so I think openness is a great segue there from security because of both of those points.
06:25First, in security, anybody can design something that they can't break.
06:29But the question is, can someone else break it?
06:31And I think that's always a more interesting and more difficult problem.
06:35But then there's also the idea of building on others' work.
06:39I think that's huge.
06:40If you think about what's happened in research from the history of all of mankind, historically
06:46research happened in academia.
06:48And then eventually more basic research became more applied within industry.
06:54But it seems like with artificial intelligence, that a lot of this has shifted to industry first.
07:00In fact, what you described to me sounds very much like an academic lab.
07:05So is that a problem that we're moving basic science from academia?
07:09Or are we?
07:10Maybe I'm begging the question.
07:11Is this a move that's happening?
07:12Is this a problem?
07:13What do you think?
07:14Well, I think both academia and industry have advantages when it comes to AI research.
07:20And I'll maybe not speak broadly across all areas of research.
07:24But for AI, in today's context, I do think both have significant advantage.
07:31On the one hand, on the industry side, we do have access to vast resources, in particular
07:36with respect to compute.
07:38And so when it comes to scaling some of the large language models, you need access to
07:43thousands of GPUs, which is very expensive.
07:46It takes a strong engineering team to keep these running.
07:49And so it's a lot more feasible to do this with a large industry lab.
07:54On the academic side, it's harder to have the resources and the personnel to be successful
07:59in terms of scaling large models.
08:01Now, on the academic side, there are advantages.
08:04And I do have a position in academia, I have many colleagues, I have some grad students.
08:09So I think you have to be very smart about what research questions you ask, depending
08:13on your setting.
08:14On the academic side, we have the privilege of often working in highly multidisciplinary
08:19teams.
08:20I work with people who come from philosophy, cognitive science, linguistics, and so on
08:25and so forth.
08:26We ask much broader questions.
08:28And as a result, we come up with different answers.
08:33One of the places where I sort of track the different contributions is looking at some
08:37of the very top conferences in the field and seeing like, where do the outstanding paper
08:41awards go?
08:42Do they go to academia?
08:43Do they go to industry?
08:46And in many cases, we see a mix of both.
08:48There's some really seminal work coming out, both of industry and academia, that is completely
08:55changing the field that is bringing forth some breakthroughs in AI.
08:58So I'm quite optimistic about the ability for researchers across different types of
09:05organizations to contribute.
09:07And beyond that, we haven't talked about startups, but there's a number of small startups that
09:11are doing some really phenomenal work in this space as well.
09:15And so overall, having a thriving ecosystem is in everyone's advantage.
09:20I think I'm more interested in a lot of our work in looking in ways that we can work together.
09:26Because in general, I strongly believe that having more diverse teams helps you ask different
09:32questions.
09:33So a lot of the intent behind our work on open sourcing is actually to make it easier
09:37for more diverse set of people to contribute.
09:40You made the analogy with the security community really relying on open protocols.
09:44I think there's a lot of that in how we tackle this work from the sense of like, I have amazing
09:50researchers who are putting their best every day into building models.
09:53But I do believe by exposing these models to a broader community, we will learn a lot.
09:59So when I make the models available, you know, researchers in academia and startups take
10:04these models, in some cases, find flaws with them, give some quick feedback.
10:09In many cases, we see derivatives of the model that have incredible value.
10:13One of the big launches we had in last year is our Lama model, Lama 1, Lama 2, Lama 3.
10:20Thousands of people have built derivative models from these, many of them in academic
10:24lab, fine tuning models, for example, to new languages, to open up the technology to different
10:30groups.
10:31And to me, that's where a lot of the value of having different players really comes from.
10:37I think we certainly see the value in, let's say, collaborating and iterating and keeping
10:42things open, but that's not always guaranteed to happen.
10:46What kind of incentives are there for us all to work together like this?
10:51It's always hard to predict the future, and in particular with AI and how fast things
10:55are moving.
10:56And so I completely agree with you, you know, what I will say is, as you mentioned, there's
11:00a strong culture towards open protocols at Meta that predates the AI team, the basic
11:06stack, the basic software stack is also based on many open protocols.
11:11And so that culture is there to this day, that culture continues, it goes all the way
11:15to the top of the leadership, and that commitment to open sourcing the models is strongly supported
11:21by Mark Zuckerberg and his leadership team.
11:23So I don't see that this is going to stop very soon.
11:27What is going to be important is that we continue to release models in a way that is safe.
11:32And that's a broader conversation than just one company.
11:35The governments have several points of views of how should we think about mitigating risks
11:42for this model.
11:43There's also a lot of discussions about how to deal in particular with very frontier models,
11:48the largest, most capable models.
11:51And so we're going to have to have these conversations as a society, beyond just the
11:55labs themselves.
11:57You raise the specter of risks, you know, the worry out there is that, oh, my gosh,
12:01these models are going to take over everything, and our world is going to collapse, and this
12:06is an existential threat.
12:08I'm kind of setting you up with that straw man, but do you buy that?
12:12I don't really spend a lot of time planning for the existential threat in the sense that
12:18many of these scenarios are very abstract.
12:21They're excellent, you know, stories in terms of science fiction.
12:26But in terms of actually taking a scientific and rigorous approach to that, it's not necessarily
12:32the existential risks that take most of my attention.
12:36I will say with the current generation of models, there are several potential harms
12:41to different populations.
12:42You know, algorithms have been known to have biases towards underrepresented groups, for
12:47example, in facial detection system, as well as being on the language side, very angle-centric.
12:55And so I do look quite carefully at the current set of risks and try to measure them as much
13:02as possible in a rigorous way.
13:03We build mitigations whenever we can.
13:06We've invented new techniques for doing watermarking to make sure that false information can't circulate.
13:12We've done a lot of work on bias assessment so that we can actually measure the fairness
13:17performance of our algorithms.
13:19So I look a lot more at current risks rather than these really long-term ones, just because
13:25I feel we can have a handle on it that is based on a rigorous approach, based on metrics,
13:30based on really analyzing what the harms is and how to mitigate them.
13:34The very far-fetched scenarios, it's really hypothetical.
13:37It's hard to build good systems.
13:40It's hard to do good science.
13:42It's also hard to do good policy.
13:44Yeah, I think your point's well taken about bias and metrics that you mentioned, for example,
13:50these models that have biases built in, but I mean, my gosh, they're built off training
13:54data that has massive bias built in.
13:56I find it hard to attribute that to the model itself and more to the training data.
14:01And your point there is that you can build in bias mitigation there.
14:04What kinds of things have you done towards that?
14:06Yeah, in fact, on the question of bias, it's a little bit of both.
14:10There's no doubt that many of our data sets are biased.
14:14The data sets are a reflection of our society, and unfortunately, a large amount of unfairness
14:20remains discrimination as well as having underrepresented groups in our society.
14:25So there's no question that the data sets themselves don't start off soft on a very
14:29good foot.
14:30However, the model themselves also tend to enhance these biases in that most of the machine
14:36learning techniques we have today, they're very good at interpolating the data.
14:41So you sort of take data distributed in a certain way, and the models will really push
14:46towards the norm of that data.
14:49The models tend to be very poor at extrapolating.
14:51So making predictions outside of the data set, they tend to have a larger error.
14:56So if anything, when we train the models, and we try to sort of minimize the error,
14:59you do well by predicting more towards the norm versus towards the sides of that distribution.
15:06And so the data is responsible, the models are also responsible for doing that.
15:11And then there's the way in which we deploy the models.
15:14We tend to often look at aggregate statistics.
15:17So we'll look at the overall performance of the model.
15:20And based on the overall performance, we'll say, great, we've got 95% performance on this
15:24model, it's ready to be deployed.
15:27But we don't take the time to look at a more stratified analysis of results.
15:32What is the performance with respect to different groups?
15:36And how are these groups differentially impacted with respect to how the system is deployed
15:41in a bigger system?
15:43I think there's different points where we can be much more rigorous and thoughtful to
15:48make sure that we don't enhance biases.
15:52And ideally, that we actually use AI towards a more fair and equitable society.
15:57Yeah, I think that point of averaging is huge, that we've got so much.
16:04Models feel right when they give us the answer we're expecting.
16:08The image generation feels right when it gives us the image that fits our stereotypes.
16:14And fighting that seems like it's a quite difficult problem.
16:18But on the other hand, I feel like these models can try to solve it in a way that we're not
16:22going to convince everyone in the world to suddenly stop being biased tomorrow or suddenly
16:26not have a stereotype tomorrow.
16:28But we could convince an algorithm not to have a stereotype tomorrow by tweaking some
16:33weights and changing things.
16:34And so that gives me a little more hope to manage the risks.
16:37Perhaps it's not the existential threat we're getting there yet, but it seems more plausible
16:42to me that way.
16:43I think one of the challenges is determining what we want out of these models, right?
16:48We've seen some pretty egregious examples recently of groups, which I assume is well-meaning
16:54intent to rebalance datasets, especially with representation of, for example, different
17:00racial groups in images.
17:01You know, of course, if someone asks for like an image of an engineer, you don't want only
17:05men to show up.
17:06You would hope to have a few women show up.
17:09And there's ways to rebalance the data.
17:11There's ways to sort of recompensate at the algorithmic level.
17:16But sometimes you end up with very unusual results.
17:21And so it's also a question of what are the distribution of results that we expect and
17:27that we tolerate as a society.
17:29And in some cases, that's not very well defined, especially when the representation is biased
17:35within the real world as well.
17:38That seems incredibly hard because the problem switches from being an engineering problem
17:42and engineering problems you can typically solve with enough pizza and caffeine.
17:48And when you get to these more difficult problems, then they tend to be trade-offs and they tend
17:52to be choices.
17:53And these choices are very difficult.
17:56They're not improving an algorithm, which is the kind of thing that we can get into.
18:00But knowing what it should do seems like a much harder problem.
18:03And again, that seems much worse, too, as these technologies become so pervasive.
18:08If, for example, Meta does make these algorithms available to people as part of the open source
18:13process, by definition, more people have access to them and then more people have to make
18:18these difficult decisions.
18:20That seems much harder to scale than algorithms.
18:24I agree.
18:25I think in many ways, deciding as a society what we want these models to optimize for
18:32and how we want to use them is a very complicated question.
18:36That's also the reason why at Meta we often open source the research models.
18:40We don't necessarily open source the models that are running into production.
18:44That would open us up, I think, to undue attacks.
18:47And it's something we have to be careful about.
18:49But we often open our research models.
18:51And so that means very early on, if there are major opportunity to improve them, we
18:56learn much faster.
18:58And so that gives us a way to essentially make sure that by the time a model makes
19:03it into product, it's actually much better than the very first version.
19:08And we will release multiple versions as the research evolves, as we've seen, for example,
19:12with the Lama language models I mentioned earlier.
19:14We released Lama 1, Lama 2, Lama 3, and so on.
19:18And every generation gets significantly better.
19:21Some of that is, of course, the work of our own fabulous research teams.
19:25But some of that is also the contributions from the broader community.
19:29And these contributions come in different forms.
19:31You know, there's people who have better ways of mitigating, for example, safety risks.
19:36There are people who bring new data set that are allowing us to evaluate new capabilities.
19:42And there's actually some very nice optimization tricks that allow us to train the models faster.
19:48And so all of that sort of converges to help make the models better over time.
19:53I think the analogy that sticks with me is how image processing improved from the 2012
20:00and the ImageNet competition that, you know, again, that came out of originally academia,
20:05Toronto, but then exploded as everyone could see what everyone else was doing.
20:11Everyone brought something better, a faster implementation, a smaller implementation,
20:14a bigger one.
20:16And the accuracy just over the very short time got really truly phenomenal.
20:21Yeah.
20:22Let's shift gears a little bit.
20:24Joelle, you're an AI researcher and also a professor.
20:28How did you find yourself in this line of work?
20:30I'm very driven by curiosity, I have to say.
20:34I first got into robotics, that was sort of my gateway into AI.
20:40I was doing an undergrad degree in engineering at the University of Waterloo.
20:44And near the end of that, I had the chance to work on a robotics project, building a
20:47six-legged walking robot, and in particular, the sensor system for that robot.
20:53So we had some sonars and had to process the information and from that decide sort of where
20:57were the obstacles in the environment.
20:59And so that led me to doing graduate studies, master's, PhD at Carnegie Mellon University
21:04in Pittsburgh, which is a phenomenal place to study robotics.
21:08And from there, I really got into machine learning.
21:11I found that for the robot to have relevant, timely information and to be able to take
21:17decisions, you needed to have a strong model.
21:20So my thesis work was in planning under uncertainty, the ability to take decision when there's
21:25some uncertainty about the information and developing algorithms for doing that.
21:30And from then on, I took on an academic career at McGill University in Montreal, where I'm
21:34still based, and pursuing work across areas of machine learning.
21:39A lot of applications of machine learning in healthcare.
21:42We have a fabulous faculty of medicine here at McGill, and so I had many very exciting
21:48partnerships there.
21:50And also a lot of work on building dialogue systems, which today, you know, we recognize
21:55as language models and chatbots, but I was building some of the very preliminary version
22:01of this work in the early 2000s.
22:03And so, because I do use curiosity as my main motor, it has allowed me to work across
22:09several subfields of AI, robotics, language, perception, and applications.
22:15And so that gave me a pretty good set of knowledge and experience to then come into a place like
22:21meta where the teams that I work with do fundamental research, but we work closely with product
22:28teams and try to both push the frontier in terms of the science, but also push the frontier
22:33in terms of new products, new experiences.
22:37So clearly there's lots that meta is doing around the core meta products, but there's
22:42the general scientific discovery that meta research is working on.
22:46What are some examples of projects that are in progress there?
22:50This is such an interesting area.
22:52I think there's enormous potential to use AI to accelerate the scientific discovery
22:58process.
22:59When we think about how it works, often, you know, let's say you're trying to discover
23:02a new molecule or discover a new material.
23:05There's a very large space of solutions, often combinatorially large, and the traditional
23:11methods have us looking through the space of molecules one by one, and we take them
23:16into the wet lab and we test them out for the properties that we want, whether it's
23:19to develop a new medication or develop a new material.
23:24And so we've had a few projects over the years that look at this problem.
23:28More recently, we have a project that's looking specifically at direct air carbon capture,
23:34really the desire to build new materials that could capture carbon in a way, of course,
23:40to address our environmental crisis.
23:42Now when you do this work, there's many steps.
23:44One of them is even just building up the data set for doing that.
23:48So we've built up a data set, synthesizing many different possibilities for this problem.
23:55And out of that, we often partner with external teams to try to validate which of these solutions
24:01may bring the most value.
24:03We've done previous work also in the area of protein synthesis that had a similar flavor,
24:08though the specifications of the protein was a little bit different.
24:12But at a core fundamental way, the problem looks very similar.
24:16So I'm really excited to see what comes of this.
24:19I've had some cases where partner teams came to me and said, in the space of about a year
24:24of working with AI, they were able to cut down the scientific process in terms of experiments
24:31that would have taken them like 25 years if they were going through the search space
24:37with more traditional methods.
24:39And I think that's something that we're seeing from other people we've talked to.
24:41We talked to, for example, Moderna, talking about their vaccine development and how it
24:46helped explore that space.
24:47And we talked about Pirelli and how they use it for tire components.
24:52So I think this idea of exploring a combinatorically large space is really pretty fascinating.
24:58It's not something that I would have expected MEDA to be involved with it at first blush.
25:04I can see, for example, the carbon dioxide from the air problem.
25:07That's probably just something you're facing in data centers, but I wouldn't have expected that.
25:11Yeah, I think, I mean, you bring up the case of data centers, I would say that's a prime application for this.
25:16We are building several data centers and it's in everyone's interest for those to be very
25:21energy efficient. We also have some strong commitments in terms of using renewable energy.
25:26And so there is a strong motivation in that space.
25:29And not to be forgotten, we also have all of the work that's happening on our work towards the
25:34metaverse, the reality lab side of MEDA, which is really the longer term vision of building
25:40AR and VR devices.
25:42And when it comes to that type of hardware design, there's a lot of really hard problems,
25:47whether it's in the space of optics or other components where AI guided design can actually be
25:53very useful to accelerate that work.
25:56Yeah, that's pretty interesting.
25:57We actually just talked with Ty Sheridan, who is the star of the Ready Player One movie.
26:01And so that's a perfect segue from the metaverse to there.
26:05We have a segment where we ask you a little rapid fire question.
26:08So just first thing that comes to your mind, what's the biggest opportunity for artificial intelligence right now?
26:15I do think that the ability to open up, to connect people across languages is huge.
26:22We've had systems where we're building up machine translation to go up to 200 different languages.
26:28But there are many more languages that are spoken only.
26:31And so we're really having the ability to build technology for anyone to understand anyone else across the planet.
26:38I think that's going to be really crucial for us to figure out how to all live together on this earth.
26:44So what's the biggest misconception that people have about AI?
26:49I don't know if it's the biggest, but one that really gets to me is thinking of AI as a black box.
26:54People think, you know, information goes in, something happens and then something comes out.
26:58I think in many ways, from where we stand today, the human brain is a lot more of a black box than AI.
27:04When I have an AI system, I can trace down with a lot of precision how information circulates,
27:09how it's calculated and how we got to the output.
27:12I cannot do that with a human brain in the same way.
27:15So, yeah, whenever someone says AI is a black box, I sort of frown a little bit and feel like, no, it's a complicated box.
27:22But we have a lot of understanding of what goes on inside there.
27:26Yeah, other people's brains make no sense to me.
27:28Mine makes perfect sense, but everyone else doesn't.
27:30What was the first career that you wanted?
27:33Oh, early on, I wanted to be a librarian.
27:37I loved reading books.
27:38I still do.
27:39I still read quite a bit.
27:40And I thought, you know, having a job where you can just sit in a space filled with books and read all day sounded delightful.
27:47When do we have too much artificial intelligence?
27:50When are we trying to put that square peg in a round hole?
27:54I don't think of it as like one day we have enough and one day we have too much.
27:58I think it's really about being smart about where you bring in AI into a system.
28:05So already there are places where AI shouldn't go.
28:08And there are places, or at least the version of the models we have today,
28:12and there are places where we could bring in AI much more aggressively, I think.
28:15So I think what's really important is figuring out how to bring it in in a way that it brings real value,
28:22economic value, of course, but real social value as well, and being thoughtful about that.
28:27Yeah, that ties to your previous answer about the difficult parts of using the technology or not the technology itself.
28:34So what's one thing that you wish that artificial intelligence could do now that it can't do currently?
28:40I wish that AI systems could understand each other.
28:44Right now we're building a lot of AI systems that are individual.
28:48They're all fine-tuned for an individual performance.
28:51But once we start deploying many of these AI agents together in the same place,
28:56our methods for understanding the dynamics between several agents are very primitive.
29:03And I think there's a ton of work to do.
29:05You know, if we look to humans as the society of agents that is most evolved today,
29:10we derive a lot of our longevity, our robustness, our success through our social ties.
29:17And AI systems today have no idea how to build social ties.
29:21That's interesting, because I think we spend so much time thinking about the human-computer interface
29:25and the computer-human interface and not as much about the computer-computer interface.
29:31This has been a fascinating discussion.
29:32I really kind of opened my eyes to all the things that Meta is doing that's beyond just that sort of surface research
29:38that's more obvious in the newspapers and media reports.
29:42Thanks for taking the time to talk with us today.
29:44Yes, very inspiring conversation. Thank you.
29:47My pleasure. Happy to be with you.
29:50Thanks for listening to Season 9 of Me, Myself and AI.
29:54Our show will be back in September with more episodes.
29:58In the meantime, if you missed any of the earlier episodes on Responsible AI,
30:03please go back and have a listen.
30:05We talk with Amnesty International, Airbnb and Salesforce.
30:10Thank you for listening and also for reviewing our show.
30:14Talk to you soon.