Helping Doctors Make Better Decisions With Data: UC Berkeley’s Ziad Obermeyer

When Ziad Obermeyer was a resident in an emergency medication program, he discovered himself mendacity awake at night time worrying concerning the advanced parts of affected person diagnoses that physicians may miss. He subsequently discovered his method to information science and analysis and has since coauthored quite a few papers on algorithmic bias and the usage of AI and machine studying in predictive analytics in well being care.

Ziad joins Sam Ransbotham and Shervin Khodabandeh to speak about his profession trajectory and spotlight among the doubtlessly breakthrough analysis he has performed that’s aimed toward stopping loss of life from cardiac occasions, stopping Alzheimer’s illness, and treating different acute and continual situations.

Learn extra about our present and observe together with the collection at https://sloanreview.mit.edu/aipodcast.

Subscribe to Me, Myself, and AI on Apple Podcasts, Spotify, or Google Podcasts.

Give your feedback on this two-question survey.

Transcript

Sam Ransbotham: At present, machine studying researchers need to beg and plead for well being care information. This shortage essentially limits our progress. What may change after we get open, curated, fascinating information? Discover out on at the moment’s episode.

Ziad Obermeyer: I’m Ziad Obermeyer from Berkeley, and also you’re listening to Me, Myself, and AI.

Sam Ransbotham: Welcome to Me, Myself, and AI, a podcast on synthetic intelligence in enterprise. Every episode, we introduce you to somebody innovating with AI. I’m Sam Ransbotham, professor of analytics at Boston Faculty. I’m additionally the AI and enterprise technique visitor editor at MIT Sloan Administration Evaluation.

Shervin Khodabandeh: And I’m Shervin Khodabandeh, senior associate with BCG and one of many leaders of our AI enterprise. Collectively, MIT SMR and BCG have been researching and publishing on AI since 2017, interviewing tons of of practitioners and surveying 1000’s of corporations on what it takes to construct and to deploy and scale AI capabilities and actually rework the way in which organizations function.

Sam Ransbotham: As we speak Shervin and I are thrilled to have Ziad Obermeyer becoming a member of us. Ziad, thanks for being right here. Welcome.

Ziad Obermeyer: Thanks. It’s great to be right here.

Sam Ransbotham: I received to know Ziad on the NBER conference in Toronto, the place he was speaking about a few of his well being information platforms work, so perhaps let’s begin there. Ziad, inform us a little bit bit about what you’re doing, what this thrilling platform is about. Inform us about Nightingale.

Ziad Obermeyer: Completely. I can inform you perhaps a little bit bit concerning the backstory, which is that every one of my analysis is in a roundabout way or one other making use of machine studying or synthetic intelligence to well being care information. And regardless that I say I do analysis on this space, really what I spend a number of my time on is pleading for entry to the information that I want to try this analysis — pleading, wheeling and dealing, utilizing all the networks and contacts that I’ve amassed through the years. And it’s nonetheless simply extremely laborious and irritating.

And so, given how a lot time I used to be spending on that, one among my coauthors — Sendhil Mullainathan, who’s on the College of Chicago — and I made a decision that we had been most likely not alone on this ache. And so a couple of years in the past, because of the help from Schmidt Futures, Eric Schmidt’s basis, we had been in a position to launch a nonprofit referred to as Nightingale. Nightingale Open Science — that’s its full identify — is a nonprofit that makes use of philanthropic funding to construct out fascinating information units in partnership with well being programs.

We work with well being programs to know the issues which can be high-priority and fascinating issues for them to work on, and we construct information units that take large quantities of imaging — so chest X-rays, electrocardiogram waveforms, digital pathology, biopsy specimens — and we pair these pictures with fascinating outcomes from the digital well being file and typically from Social Safety information after we need mortality. And we create information units which can be aimed toward answering among the most fascinating and essential questions in well being and medication at the moment: Why do some cancers unfold and different cancers don’t? Why do some individuals get a runny nostril from COVID and different individuals find yourself within the ICU? So all of those questions are areas the place machine studying can actually assist — not simply assist medical doctors make higher choices, however assist drive ahead among the science. However these information units are in very brief provide.

We create these information units with well being programs, after which we de-identify them and we put them on our cloud platform, the place we make them accessible to researchers around the globe without cost. And I feel our inspiration for lots of that work was the large progress in different areas of machine studying, pushed by the provision of not simply information units however open, curated, fascinating information units that take goal at essential issues and which can be made accessible for individuals who need to drive efficiency ahead on a few of these duties. That’s one of many well being information platforms that I’ve been engaged on for the previous few years.

Sam Ransbotham: So give us some examples of that. What are some analogies? What are the opposite platforms you’re referring to?

Ziad Obermeyer: I feel essentially the most well-known one among these is known as ImageNet. This was put collectively a variety of years in the past by primarily getting a bunch of pictures from the web after which getting individuals to caption these pictures. So, , we get a photograph. It’s individuals enjoying Frisbee on the seashore. After which, as soon as we’ve received thousands and thousands and thousands and thousands of these pictures, we will prepare algorithms that map from the gathering of pixels in that picture to the caption {that a} human would assign that picture.

There are numerous information units like that: There’s a handwriting-recognition information set; there’s a facial-recognition information set. And people information units, as we’ve seen time and time once more, have simply been instrumental in driving ahead progress in machine studying. So individuals type groups. They collaborate; they compete with one another, all making an attempt to do the very best at these duties. And that’s simply been an enormous engine of progress that, together with computational energy and the {hardware} facet, has actually pushed the innovation ahead on the software program facet.

Shervin Khodabandeh: That is actually fascinating. Perhaps we take a couple of steps again. Ziad, inform us a bit about your analysis and what you’re aiming to do with machine studying [and] information. You’re a doctor by coaching. You are also a scientist and affiliate professor. So perhaps give us a little bit bit about your background, the way you ended up the place you might be — this wonderful unicorn of a number of disciplines and talent units.

Ziad Obermeyer: That’s a really type description. Thanks, Shervin. I studied historical past in college, after which I did a grasp’s in historical past and philosophy. And I used to be actually considering science and finding out how science received made: how new fields fashioned, how scientists sort of sorted into these factions and information was socially constructed.

After a short stint in administration consulting, I went to medical college, and I used to be really a analysis assistant for Chris Murray, who ran the International Burden of Illness undertaking and nonetheless does actually, actually wonderful work in that world of quantifying the burden of illness globally. I realized a number of the way to do analysis from Chris. And I feel that’s a recurring theme — that as a result of I’m extraordinarily fortunate and privileged, I managed to spend time round some actually, actually good individuals who invested a number of effort and time in instructing me stuff.

I skilled in emergency medication. Being within the ER is … an enchanting and fascinating and tense expertise, since you’re always confronted with the bounds of your personal skill to assume and perceive issues. So I did medical college and residency, after which I began working towards.

After I began working towards was once I began seeing all of this stuff about medication that had been so troublesome, and that may, , actually hold me up at night time. Like, I’d go house after a shift and I’d simply lie in mattress, and I’d be tremendous pressured about this one affected person that I’d despatched house, as a result of I’d keep in mind one thing about her or there was the check I ought to have ordered and I didn’t order it. So I turned a number of that stress into analysis, which — I don’t know what Dr. Freud would say about it, however there’s most likely some points I’ll must discover later with my therapist.

I feel that irrespective of how good you might be at that job, should you’re paying consideration, you’re all the time making errors. I feel that’s an expertise that nearly each physician has. And I feel the factor that I noticed sooner or later was that the sorts of errors that medical doctors are almost definitely to make are the sorts of issues that machine studying could be actually, actually good at. So one of many hardest issues that medical doctors must do — actually, a elementary exercise in medication — is prognosis.

So, what’s prognosis? Analysis is taking a look at a affected person, taking a look at all the check outcomes, the X-rays, the laboratory information, how they appear, all this stuff, and distilling that all the way down to a single variable. Like, does this particular person have pneumonia? Have they got congestive coronary heart failure or one thing like that? So mapping this very high-dimensional information set on the affected person stage to a chance of getting illness one, illness two, illness three — [it’s a] nice machine studying activity.

So medication is simply full of those issues that (a) medical doctors do very poorly on many alternative measures of efficiency and (b) if algorithms are constructed thoughtfully and punctiliously round these issues, they may actually enhance the standard of decision-making. And in order that’s the genesis of my complete analysis program: constructing algorithms that (a) assist medical doctors make higher choices and (b), optimistically, additionally attempt to push ahead the science underlying a few of these choices round who has sudden cardiac loss of life, who develops problems from COVID, who develops metastatic most cancers and who doesn’t.

Shervin Khodabandeh: I feel one other factor that presumably exists in medication versus different fields the place people and machines work collectively to make machines higher and make people higher might be the truth that the knowledgeable’s opinion that’s correcting the machine [has] most likely gone by much more diligence, as a result of there are playbooks and pointers and, by advantage of you changing into a doctor and licensed and having board certification, it’s unlikely that the extent of disagreement or variance between consultants in medication could be extra so than, let’s say, within the area of promoting or, , credit score underwriting, or … I might assume that that makes the coaching a part of the algorithm a bit extra standardized or much less topic to one knowledgeable’s opinion.

Ziad Obermeyer: It’s a very, actually fascinating set of questions. I knew rather a lot about medication. I knew some staple items about the way to do analysis. However I began working with a well being economist at Harvard, David Cutler, after which with one among his colleagues, Sendhil Mullainathan, and that’s the place I began investing a number of time in studying among the technical expertise that, together with these scientific expertise and scientific information, was the idea for the analysis that I’m doing at the moment.

Typically we attempt to clear up this downside by saying, “Effectively, we’re not simply going to have one radiologist. We’re going to have 5 radiologists. After which we’re going to take the bulk vote.” However are we actually working towards medication by majority vote? So it’s one among these fascinating locations the place doing machine studying in medication could be very, very totally different from different areas, as a result of we essentially have a extra difficult relationship with the bottom fact. And human opinion, as skilled as these consultants are and as a lot observe as they’ve gotten over years of residency and coaching — we will’t take into account that the reality.

I’ll inform you about one paper that we wrote a couple of years in the past. This was led by my colleague Emma Pierson, who’s a pc scientist at Cornell. And what we confirmed is that radiologists systematically miss issues on X-rays — on this case, knee X-rays — that disproportionately prompted ache in Black sufferers. So while you return to the historical past of what we learn about arthritis in medication, a number of the unique research that generated the scoring programs that medical doctors nonetheless use at the moment had been developed on coal miners in Lancashire, England, within the Forties and ’50s. So it’s in no way shocking that information constructed up in that very particular time and place wouldn’t essentially map onto the populations that medical doctors see of their workplaces at the moment in England or within the U.S.

And the way in which we confirmed that was really by coaching an algorithm not to do the factor that most individuals would have skilled an algorithm on, which is to go from the X-ray to what the radiologist would have mentioned concerning the X-ray. If we prepare an algorithm that simply encodes the radiologist’s information in an algorithm, we’re going to encode all the errors and biases that that radiologist has. So, what we did as a substitute is we skilled an algorithm to foretell not what the radiologist mentioned concerning the knee however what the affected person mentioned concerning the knee.

So we skilled the algorithm to mainly predict, is that this knee a painful knee or not? And that’s how we designed an algorithm that would expose that bias in — not the radiologist a lot as in medical information. And it actually supplied a path ahead … that algorithms, regardless that a number of my work has proven that they will reinforce and even scale up racial biases and different kinds of biases in medication, there’s additionally this pathway by which they will do issues that people can’t. They will discover sign in these advanced pictures and waveforms that people miss, and so they may also be forces for justice and fairness, simply as simply as they are often forces that reinforce all the ugly issues about our well being care system and society.

Shervin Khodabandeh: The issue of coaching is definitely a lot more durable since you don’t have floor fact. And also you’re not solely exposing … otherwise you’re not solely making an attempt to right the fashions’ biases or inaccuracies but in addition the physicians’ as a part of the coaching.

Ziad Obermeyer: Yeah, completely put. The problem isn’t simply constructing the algorithm. It’s not simply the identical problem we have now in some other area. The elemental problem in well being is creating the information set that speaks to that floor fact. The excellent news is that because of the large success of digital well being information, the flexibility to hyperlink information units from hospitals to state Social Safety information and different fascinating sources of fact from elsewhere, there’s a wealth of data that lets us sew collectively and triangulate that floor fact. Nevertheless it’s a really troublesome scientific downside, not only a machine studying downside. I feel that’s one of many large causes, in addition to the dearth of knowledge.

The opposite purpose that we haven’t seen machine studying rework the observe of medication in the identical manner that it’s reworked different industries is as a result of these issues require a sure bilingual talent set. It’s essential to perceive the way to do helpful issues with information. However you additionally want to essentially perceive the scientific medication facet of those issues to be efficient as a result of you’ll be able to’t simply swap within the radiologist’s judgment for the judgment of whether or not there’s a cat or not on this picture. It’s a a lot, a lot more durable downside.

Sam Ransbotham: That bilingual factor appears actually powerful, although. And it makes me take into consideration your background. You’re clearly able the place you’ve ended up with each of those languages. I’m additionally curious, do we have now to have that?

Ziad Obermeyer: With out minimizing the problem of this downside, let me level to an instance the place one thing like this has labored very well, which is a little bit bit extra in your world, Sam, than mine, however behavioral economics, I feel, is a very, actually nice instance of a essentially new area that requires precisely the identical sort of bilingualism. And so, what you wanted for behavioral economics as a area was, you wanted, to start with, economics to begin taking human habits critically, past only a easy operate of incentives. However you additionally wanted psychologists to make fairly large investments in studying the technical foundation for demonstrating what’s a bias and what’s an error, what’s not an error, issues like that. And so I feel there’s a very nice analogy to this world, the place the medical doctors are enjoying the function of the psychologists and the pc scientists, but in addition the economists are enjoying the function of the economists within the different world.

I feel that one of many causes that this stuff are laborious — to echo some causes for pessimism — is that we’re not superb inside academia, regardless of everybody saying how fascinating and great multidisciplinary work is. Not one of the incentives are actually set as much as promote that sort of work. And so should you’re a pc scientist and it’s essential to get your paper into some convention proceedings, whether or not you do an amazing, A+ job in attending to a floor fact label or whether or not you do a very unhealthy job doesn’t actually matter on your chance of getting that paper into your favourite convention proceedings. And I feel should you’re a physician, whether or not you’re making large investments in machine studying or not is just not actually going to have an effect on your chance of getting that NIH grant that you simply’re making use of for.

And so, to return to one of many causes Sendhil and I began Nightingale, I feel we want these sorts of establishments that assist construct that group of people who find themselves taking these sorts of issues critically. So should you’re a Ph.D. scholar, you want information to work within the area. I’ve a number of Ph.D. college students at Berkeley who come to me and say, “Oh, I’d actually love to use the factor that I’m good at in machine studying to well being.” And I say, “Nice, we’ll add you onto the Knowledge Use Settlement, and then you definately’ll need to do all the coaching, after which we’ll amend the IRB [institutional review board proposal]. …” And by the point that’s executed, they’ve already received a job at Fb or wherever, and it’s throughout. And so I feel that increase these public items on this space is a very good place to begin to construct the group of people that can do the work and be a set of collaborators and peer reviewers and issues like that. Nevertheless it’s a course of.

Shervin Khodabandeh: The opposite factor that your feedback lit up in my head is the likelihood that the experiment design that we might do at the moment, to do a prognosis of no matter it’s, on condition that information is a lot extra considerable now than it was perhaps 50, 30, 20 years in the past, the place information was actually scarce, I ponder if that really modifications even the rules for the way we take into consideration a optimistic prognosis, as a result of a few of these correlations that you simply’re speaking about could not have even been potential, so perhaps no person even thought of, “You for certain have this illness if these two issues occur.” Effectively, perhaps there’s a 3rd factor that may occur two months later that you simply don’t even know would occur, as a result of no person collected information on it or no person may correlate the information.

Ziad Obermeyer: Yeah, nice level, and I feel it actually highlights one of many large benefits of doing this work at the moment, when we have now longitudinal information from digital well being information that’s linkable to a number of different information from a number of different totally different locations. The worth of the information, because it expands in scale and scope and linkages, simply will increase exponentially. And precisely as you mentioned, it opens up a ton of recent potentialities to be taught that weren’t afforded to us beforehand.

Sam Ransbotham: You set us up with a Hobson’s alternative right here, like, “OK, we will both wait three days for the tradition check, or somebody has to make a name proper now.” And that is declaring simply how a lot better we will measure all kinds of issues and perhaps measure issues that we weren’t even occupied with measuring now … that may inform us what was going to occur in three days, and we don’t … such as you say, I feel we’re nonetheless very early in that course of.

You talked about among the issues, like detrimental information, that will poison the nicely. And once I take into consideration that … by analogy, I simply taught this week at school, the [module on] handwriting recognition. And at school, I’m in a position to take a bunch of scholars, and we’re in a position to carry out with algorithms what would have gained contests 15 years in the past. And we will do this at school on laptops.

Effectively, by analogy, with these information units you’re placing collectively proper now, what are these kinds of wins that we will anticipate? I imply, the way in which to offset the poisoned nicely is the miracle remedy. I don’t need to get too snake-oily right here, however what sorts of issues can we hope for? What sort of successes are you seeing thus far with making these information units accessible?

Ziad Obermeyer: I can inform you a little bit bit … once I take into consideration the output for a company like Nightingale Open Science, I feel the output there may be information and papers and computational strategies which can be developed on these information, however I feel there’s additionally one other manner to consider what the output is.

I can inform you a little bit bit about one other platform that I’ve been engaged on, which is known as Dandelion Well being. Dandelion is a for-profit firm, and what that firm does is, we first have agreements with a handful of very giant well being programs throughout the U.S., and thru these agreements, we get entry to all of their information. And once I say all of their information, I actually imply all of their information. So not simply the structured digital well being information, but in addition the electrocardiogram waveforms, the in-patient monitoring information when somebody’s within the hospital, the digital pathology, the sleep monitoring. Every part.

This firm is designed to assist clear up that bottleneck and assist individuals get these merchandise into the clinic quicker. And the way in which … , I grappled with this rather a lot, and I feel the way in which I give it some thought is that there are clearly downsides to utilizing well being information for product improvement, and I feel that there are actual dangers to privateness and a number of issues that individuals care about. And I feel these dangers are actual, and so they’re very salient to us. There’s one other set of dangers which can be simply as actual however rather a lot much less salient round not utilizing information.

I feel there are additionally a variety of functions to what individuals consider as life sciences and to scientific trials. There’s a complete set of situations at the moment, one thing like Alzheimer’s, and we [recently] noticed some unhappy information from yet one more promising Alzheimer’s drug. It’s been fairly unhappy information for many years on this space. And one of many causes is that this bizarre indisputable fact that I hadn’t considered till I began seeing a few of these functions, which is that if you wish to run a trial for a drug for Alzheimer’s, it’s a must to enroll individuals who have Alzheimer’s. However meaning the one medication which you can develop are those that … they mainly need to reverse the course of a illness that’s already set in.

So now think about you had an Alzheimer’s predictor, that with some lead time may discover people who find themselves at excessive threat of growing Alzheimer’s however don’t but have it. Now you’ll be able to take these individuals and enroll them in a scientific trial. And now you’ll be able to check a complete new sort of drug, a drug that would stop that illness, as a substitute of getting to reverse it or gradual it down. So, that’s, I feel, actually, actually thrilling too.

Sam Ransbotham: We’re most likely getting shut on time. Shervin, are you the five-question particular person or am I at the moment?

Shervin Khodabandeh: I can do it.

Sam Ransbotham: We’ll clarify this, Ziad. It’s not as onerous because it sounds. We’ve got a normal manner of closing out the episodes.

Shervin Khodabandeh: So, Ziad, we have now a phase the place we are going to ask you a collection of rapid-fire questions, and also you simply inform us no matter involves your thoughts.

Ziad Obermeyer: OK.

Shervin Khodabandeh: What’s your proudest AI or machine studying second?

Ziad Obermeyer: I’m engaged on a paper proper now that’s in collaboration with a heart specialist in Sweden, the place we’re linking all the electrocardiogram waveforms that had been ever executed in that area with loss of life certificates. And we’ve developed an algorithm that may really forecast with a shocking diploma of accuracy who’s going to drop lifeless from sudden cardiac loss of life within the 12 months after that ECG. And I feel, along with being tremendous fascinating scientifically, this opens up this large, large social worth of with the ability to discover individuals earlier than they drop lifeless to be able to examine them and even put in a defibrillator that would stop this catastrophic factor that occurs tons of of 1000’s of occasions yearly and that medical doctors don’t perceive and may’t predict.

Shervin Khodabandeh: That’s the sort of factor to be happy with. Wow. What worries you about AI?

Ziad Obermeyer: I feel the work that I’ve executed on algorithmic bias has made me replace negatively on how a lot hurt these algorithms can do. We studied one algorithm that’s sadly most likely nonetheless deployed in most of the greatest well being programs within the nation. By the corporate that makes its estimates, it’s getting used for 70 million individuals yearly to display them and provides them entry or to not further assist with their well being. And I feel that these sorts of merchandise, these aren’t theoretical dangers. These are actual merchandise which can be really deployed within the well being care system, affecting decision-making each day. They’re doing an unlimited quantity of hurt. In the long run, I fear that these sorts of issues are going to trigger reactions that may be fully justified in shutting down a variety of issues that that would in the end be very optimistic.

Shervin Khodabandeh: Your favourite exercise that includes no expertise?

Ziad Obermeyer: No expertise? Um, I’m going to interpret that liberally and assume {that a} surfboard doesn’t contain expertise, regardless that it takes a number of …

Shervin Khodabandeh: I understand, to an instructional, that’s a particularly ill-posed query.

Ziad Obermeyer: I actually like snowboarding, and my spouse, regardless of being from Sweden, doesn’t like snow or something to do with snow. And so our compromise was studying the way to surf collectively, and that’s turn into actually one among my favourite issues to do. And one of many issues I like about it, even over snowboarding, is how little expertise there may be. There’s no raise. There’s no boots. There’s no, like, all this stuff that you simply want for snowboarding. Browsing, you simply want one plank, and then you definately simply get on the market. And it’s great.

Shervin Khodabandeh: Sure, and thanks for difficult that query. I feel we have to rephrase that query. What’s the primary profession you wished? What did you need to be while you grew up?

Ziad Obermeyer: After I was in grade college and highschool was when there was this monumental explosion of curiosity and optimism round human genetics, and I used to be simply fascinated by that and by biology. And on reflection, I’m very glad that I didn’t do this, as a result of I feel that the stuff that I’m engaged on now — I’m going to make a prediction that many individuals will disagree with — however I feel that machine studying utilized to information has nothing to do with genetics. Like ECGs and pictures, it’s going to have a far bigger influence on well being and medication far earlier than human genetics.

Shervin Khodabandeh: And at last, what’s your biggest want for AI sooner or later?

Ziad Obermeyer: After I go searching at the moment on the makes use of to which AI is being put, I feel the proportion of issues which can be producing giant quantities of social worth is sadly pretty small. I feel there’s a number of ad-click optimization, and I don’t thoughts that. I imply, I profit rather a lot from it. I’m not somebody who opts out of all these issues; I would like personalised adverts. I purchase a number of issues which can be focused to me on Instagram. I feel it’s nice. I’m not … no critique of advert personalization, however the alternative prices of ad-click personalization, given the expertise and expertise and cash that’s being put into it, I feel is giant relative to different issues, like well being and medication and these different areas the place AI has large potential to enhance society as a complete. I hope that in 10 or 20 years, there’s going to be a a lot larger proportion of individuals engaged on these sorts of questions than there are at the moment.

Shervin Khodabandeh: Thanks for that.

Sam Ransbotham: You made some extent right here that actually resonated with me, and that’s the chance price of not doing issues with well being care information. And as you had been speaking about among the negatives of well being care, I used to be sort of shaking my head no, that I feel we have now a lot alternative on the market to … will I commerce a few of my information for 20 extra wholesome years, 30 extra wholesome years? Signal me up. And so I’m hoping that perhaps a few of our listeners — that resonates with [them]. Thanks for citing a few of these issues and elevating consciousness about an enchanting set of initiatives that you simply’re simply in every single place with. I feel we referred to as you bilingual, trilingual — I’m undecided what number of linguals we will go as much as. However thanks a lot for taking the time to speak with us at the moment.

Ziad Obermeyer: It was such a pleasure to speak to each of you.

Sam Ransbotham: Thanks for listening. Subsequent time, Shervin and I speak with Eric Boyd, AI platform lead at Microsoft. Speak to you then.

Allison Ryder: Thanks for listening to Me, Myself, and AI. We imagine, such as you, that the dialog about AI implementation doesn’t begin and cease with this podcast. That’s why we’ve created a bunch on LinkedIn particularly for listeners such as you. It’s referred to as AI for Leaders, and should you be a part of us, you’ll be able to chat with present creators and hosts, ask your personal questions, share your insights, and acquire entry to beneficial sources about AI implementation from MIT SMR and BCG. You possibly can entry it by visiting mitsmr.com/AIforLeaders. We’ll put that hyperlink within the present notes, and we hope to see you there.