So you know AI in health care has many headwinds to it. First of all, if you compare it with the automotive industry or retail, all the legal frameworks that govern health care are built to slow down innovation. And always prepare for a worst-case scenario. You’d rather have from a legal standpoint, you have– you’d rather have people receive less care, and people receive more care, but some of that care will be erroneous. The issue that comes up even more into play when you go into a hospital setting, which is typically a high-stress environment in which health care providers, doctors, surgeons, nursing staff, et cetera, need to make decisions which are always under a shroud of mist.
We don’t actually know what’s going to happen in many cases, and in fact, much of the biologic mechanisms that govern our health were not elucidated today. These things mean that the appetite for experimentation and the appetite for scaling out processes and actually moving them away from the human is fairly diminished across the system. The way that we deploy AI now in our environment starts, from first of all, and I think you said this really well. And both of you, to be honest, talk with the people and identify the problems.
The best solutions to date haven’t been the most sophisticated AI algorithms [INAUDIBLE], which [? all ?] seem as transformative throughout the entire US and beyond. But have been things that have been developed by researchers based in the institutes for very practical problems, and deployed in a clinical setting in which in one service, and then maybe two services, maybe three services. Now from the point of innovation and especially in terms of our why, and actually investing in innovation, that’s a very challenging story to tell investors, right? Because you need to do a lot of homegrown solutions, and some will stick, and some scale and some will evolve.
Investors like really upstream RND like much of medical research in general, which on the one hand, and on both hands really makes the hospital into the ideal environment to develop these solutions from the people who would actually use that. We saw Dr. Constance Lehman and Dr. Regina Barzilay before that on the film here. Dr. Lehman is [? the ?][? best ?] radiologist, women’s health advocate. She’s not an AI developer, and yet she collaborated with a Ph.D. in computer science from MIT and developed a product that has the potential to change women’s health screening in the US and beyond that.
I think it’s a great story. I think we were a bit– I think we’re in a very early stage of that disruption, and I’m not sure if it’s going to be a disruption as much as it could be in other fields given the large level of heterogeneity, and differences between cultural tendencies, medical practices, human anthropometrics, and human disease across different geographies and systems.
But even incremental changes can have a big impact. If you look at the world through the work that people do, and through the problems, and say how do we solve these problems rather than saying how do I apply this technology in this environment. We could go a long way in all of these industries, it seems.
Right. I absolutely agree with you, and I think the unique factor here is that in order to maintain AI, which will continue being [? performance ?] and maintain some robustness to change or adaptation to changes, it’s an ongoing [? employee ?] involved process. [? So and ?] different from maybe in some places we can say I’ve really built a system that knows how to heal itself and knows how to grow, and I can deploy that. And hopefully, I won’t create a fascist chatterbox as we often hear about. And if it is [? then ?] you can, you know, stop the [? bot ?] at least at this point in history.
In health care, you’ll need to continue being involved with the upkeep, understanding the results. And when we actually deploy AI solutions in the hospital, one of the key characteristics is that the way to monitor them, and continual reviewing of the results is often developed by the staff, actually does that. Because every person would have a different dashboard, and one size will never fit all.
One of the things that someone told us when we started this task force on the work of the future about a year and a half ago was everybody is going to focus on robots and AI. But it’s really about the computers. And we were talking a little bit before we came on stage here about many applications where you have a simple iPad or a simple phone-based– that all it does is present a document or a form in a place or in a way that’s more convenient than a piece of paper at a desk. And you can say that simple. You can also say that’s ten years ago technology. Maybe the tablet is introduced. And it takes that long for these devices to really proliferate at scale.
And I wonder if each of you may have some stories about just the slowness of proliferation, and at the same time once it really happens how widespread and profound it can become.
Yeah, I can give an example of the technology. I don’t know why it took so long to do this. Actually, in this case, I think it was the CEO who had a bad customer experience, and that’s why the company ended up fixing it. But since I gave a Costco example, I’ll give a [? Sam’s ?][? Club?] example not to be neutral in the industry. So [? Sam’s ?][? Club?] recently implemented this thing called Sam’s garage. And then you take your car to the warehouse club. You have your tired that needs to be changed. And previously the front line associates will get lots of information about you, about your car, and then figure out what tires would go with your car. But they would put in the information in one system. They had manuals they had lots of documents to look to see what tires will actually fit with your car, and then another system to look at the prices.
This whole process went anywhere from 30 minutes to 45 minutes, just to figure out what tire fits your car, right. And they realized we can totaly redesign this, used very simple technologies, computers then and put all those manuals together, and now that process takes 2.2 minutes. They had over 50% increase in the throughput time. They had an increase in their member satisfaction and also an increase in how associates feel about their jobs because now their job is so much easier. But now they can be advocates for the customer. They can help them compare instead of wasting their time trying to figure out the different tires. They can see all these different options and talk about which one would work best, and how they could pay for it et cetera.
So this is a great example of one that is solving a real problem, improving the experience of the associate as well as the customer, but it’s very simple technology, and again, we have an opportunity to look at the [INAUDIBLE] through the lens [? or ?] work and see how can you redesign it in a way that’s a lot more efficient and better for everyone involved. So that’s one example.
David, do you have examples from Nissan?
Yeah, I’ll give you an example of our frontline supervisors. Every automotive manufacturer has some level of production system. We have the [INAUDIBLE] production [? way ?] within Nissan, which requires front line supervisors to do certain audits in their particular zone on the manufacturing line. These were traditionally done as paper audits, and then they were transferred into a computer database once the paper audit was complete. And it [? tallied ?] up a tremendous amount and the supervisor’s time. And a group of folks got together within our in-house [? APW ?] group and said, you know, we can take iPads tablets, and we can do all of this work on the tablet have it seamlessly updated into the database, and it’s going to free up the supervisor’s time.
So we got the group [? of ?] the [? APW ?] team together along with the manufacturing systems team, which is an in-house IT team if you will. And they worked together to develop the supervisor tablet system. This tablet system, although it’s very simple. It’s basically the same audit that was being done before, but the time that it saves is now given back to the supervisor to have more dialogue with the workers online, develop better relationships, look for solutions to either ergonomics or quality issues that may be happening.
And, you know, it’s a simple technology that we use in our personal lives every day, but it had never been deployed into that environment because of the most dangerous words within manufacturing. We had always done it the other way. We’d always use the paper copies, so that’s what we’re going to continue to do. Well, innovative people got together and said no, there’s a better, way let’s change it. And you know the pace of adoption was slow, we had pilots that would use and feedback. And now, you know, it’s completely deployed within the manufacturing environment, and paying great dividends.
Ittai do you have a favorite example within health care?
I think revolution which is happening right now but is yet to seize a lot of publicity is the [INAUDIBLE] interaction. Today the way that [INAUDIBLE] try to control risk and also to control utilization is a very labor-intensive process, it typically required at least in many cases until recently [? people ?] forms, and filling out stuff, and sending things and fax machines et cetera. Most of that has been digitized, and [INAUDIBLE] to a small extent.
But to the next leap which is happening right now also with companies that spun out of large providers such as ourselves is about applying AI and more machine learning methods into these interactions, and that could actually really move the needle in terms of [INAUDIBLE] cost base with some [INAUDIBLE] savings to the provider side.
So how in each of these cases or in the industries that you work in how are jobs changing, whether the high-level skill of the doctor or the retail worker? And how are organizations, if at all responding to the need for different skills, maybe lower skills, different tasks? Do you have a good example or maybe a bad example of how each work? Maybe starting with the Zeynep.
I don’t have either good or bad examples. I will say that I mean in terms of number of jobs we haven’t seen much change in retail. Ten years ago there were 4.4 million retail salespeople, now there are four– 4.5 million retail salespeople, the same thing with cashiers. There could be some wholesale change, perhaps in the cashier roles, if we have scans and go technologies. In terms of skills, someone who used a computer now is going to use an iPad app. There is not much change in the skills that I see at least from outside, but it could be very different in different industries.
Does the rise in the presence and the pressure of e-commerce change the way that the in-store retail workers need to work?
And that has– I think that is one of the reasons why we are still seeing some slight improvement in the number of jobs. So while some of the technologies have made things more efficient, the e-commerce means that now there is more work to be done inside the stores. So previously we would go to the stores just to buy our merchandise, now the story employees they assemble our order for us, and we either pick it up at the store, or we pick it up at the curbside. So technology has added more work to the retail store employees, and it has also increased the importance of competence in the operations in people’s side.
So if you’re relying on technology to pick up your orders, curbside delivery, or check inventory at the stores, you better make sure that those data are correct data from which you’re promising customers of delivery. And many retailers tend to have tremendous inaccurate inventory data, point of sales data. So these technologies are, I think, increasing the importance of paying attention to your store processes, store operations, and also the competence of your employees. So if you operate with 100 percent turnover, it becomes very difficult to get all these things done because you’re teaching somebody to do something and then two months later some other employee comes in, and these technologies are much harder to adopt with high turnover.
David, how does Nissan see scaling across the workforce?
It’s a very interesting question because to understand the skills and the need for new skills within automotive. You got to understand the automotive industry on the manufacturing side is largely defined by a lot of legacy systems. I mean, we have very large plants, huge capital assets that evolve over time, and that evolution usually takes place whenever new products are introduced, and with legacy systems, and legacy processes also come legacy people, legacy ways of thinking.
So it’s how to bring the new skill sets and the new technology in without losing your best know-how, which is housed in your legacy employees. They’re the ones that actually know how the business works, and how to bring the manufacturing systems together to produce a final product. I’ll give you an example of how we’re doing it at Nissan, and since we’re at MIT, I’m going to use an MIT student in the example.
In the body shop, we have a vision for predictive quality, and in body, this is where all the metal components come together to build the foundation of the car, very complex, and it requires a lot of technologies coming together to be able to predict the outcome of the quality. Well, one of the elements that we wanted to look at is fit and finish, and surface quality so we paired a 20-year veteran in the body shop with an MIT student. And we said, guys, we want to know how we can do this autonomously and feed the data into big data, do some analytics, and predict what the outcome is going to be. And I have a lot of meetings in my world, and the meeting that I look forward to the most was the meeting between these two individuals because it was absolutely amazing.
You had this 20-year Nissan veteran that didn’t know anything about advanced vision, and data analytics, that now talks like he invented this stuff. He is absolutely immersed in the technology. He understands it, and that understanding now gives him the ability through his 20 years of previous experience to look at other areas within the body shop, and how to deploy. And then on the flip side, I’ve got this MIT student who had never been inside of a body shop before, that is talking in detail how the body comes together, how the processes work, how the new technology fits into the process.
So it’s not just developing the skill for a new employee coming in on the legacy– on how to do the legacy work of the company, but you know it’s also those new employees teaching the veteran employees about new technologies. And only through that synergy, when you purposely pair these people with a common goal, can you really drive the skills forward where they need to be for mass new technology adoption and deployment.
That’s a great example because we hear a lot about companies who try to mix the foundational skills of older workers with the kind of data and technical skills of younger workers. And we actually have a cognitive scientist on our task force who’s taught us that a lot of those skills just decline linearly over one’s lifetime. And but other skills, relational skills, social relationships, more foundational skills actually increase pretty linearly. Unfortunately, they tend to cross right about my age. So it’s no wonder that they don’t let me write code anymore.
And I wonder Ittai if you see any of that kind of interaction in health care or you– how the partner system deals with changes in skills especially around information technologies across different types of providers and age groups.
Yeah. So first of all, in our case that [? we ?] see a lot of interest from fairly senior and experienced physicians in AI and in new technologies. So I definitely see there is a lot of appetite for adoption. I think there’s also a certain level of perhaps a narcissistic trait here because a lot of the idea behind building AI products is to scale up cognitive processes. And if you believe that your cognitive processes are worth scaling up, you’d be interested in contributing to that product.
No one at MIT feels that way. I’m sure our [INAUDIBLE]
There’s also another piece I’d say that we’ll now have to be– the wave of medical informatics in which we had a lot of, kind of post-residency programs in which doctors actually learned how to operate computers, understand more basic stuff like ICD codes, basic data analytics tool. And now we’ll– there’s a huge interest now in the next wave of more advanced analytics as well as AI. We run a fellowship program at the Center for Clinical Data Science, of which I’m the executive director of. And we get hordes of candidates of post-residency, board-certified, MD, PhDs who want to stop everything for a year and or two years, and only learn about AI, develop AI products, contribute to publications in the field et cetera.
And I actually asked some of them, you as a neuroradiologist could in some cases may be making [INAUDIBLE] to half a million dollars a year, what drives you to actually come back and do something which is similar to a [? postdoc? ?] And many people would tell you, it’s clear to me that the way that my job isn’t at risk, and [? might– ?] the field they work in is not at risk, but the way I conduct my affairs is going to change dramatically over the next few years. I’d say that’s definitely an example of that.
Interesting. Such a wealth of experience on the panel here I’m sure we’re going to have a lot of questions. So maybe we don’t have the questions yet on the monitor here, but we have a few emails in here.
Yeah. I’ll read the first one. Technology always strives towards doing more with less or enabling that could not have been possible before. So how should externalities be managed?
Good question.
Go ahead.
A little unclear how we think about externalities. Again, one of the things I always teach is that what you define as internal or what you find as external as a huge value judgment around how you draw the boundaries of the system. Maybe another question here.
It’s Paul [? Barter ?] from Toronto Management Consultant and an author. Largely you’ve had a conversation about what we would call competition within existing firms here. But we’ve got to connect two big elephants in the room in competition these days. We’ve got the global giants, the Google, Amazon et cetera, that are entering many adjacent businesses. And we’ve got startup. So you have, in the case of Google, they’re entering both automotive and health care. And we’ve got startups Tesla, and automotive, and a myriad of health care startups. Maybe speak about how you perceive competition from those, you know, existential threats if you like?
I’m happy to take the retail perspective. The big threat in the retail world is Amazon, which accounts for almost half of all the e-commerce sales. And I think for many companies, one of the issues have been that they focus too much on Amazon, and not enough on their customers. So retail is still a big space. There’s still a lot of opportunity to compete. Price and convenience are not the only thing that matter to customers.
So I think the way to address that threat in at least my world of retail is to reoffer your customers a compelling reason to buy from you. And that compelling reason can’t just be more products, and more convenience, it has to be something else. And for that to be goods, you have to invest in your store processes, and you have to invest in your store employees so that customers have a fantastic experience.
- Hi everyone. My name is Anthony [INAUDIBLE]. I am pharmaceutical chemist, and I work at Johnson and Johnson in Peru. So my question is for Ittai, maybe if we talk about sales, marketing, [? and ?][? other. ?] We need to communicate at human level. But if we are talking about a treatment– a medical treatment, maybe we need a treatment at human level, but also at the molecular level. So how are you using these AI to maybe know about the DNA of the patient, and design specific treatment for that patient? Thank you.
That’s an excellent question. First of all, I’d say that one of the main limitations in implementing AI in health care, as I said earlier, is lack of understanding of biologic systems. And in that aspect AI has a huge potential in order to make these connections and really converge different data sources, which is something that us as humans do a very poor job at usually. I’d say that in terms of– if you look at the hospital [? tumor ?] board that’s always one of the most confusing of meetings one could attend. Because each specialty brings his own angle of view towards something, and based on their experience, and based on the data sources you need to have a geneticist based on genomics, you need to have the surgeon based on morphology et cetera.
And in that sense, there’s a lot of work, including in our system of how to fuse these data types together. The current generation was– could use different [? neural ?][? nets ?], but eventually, we had to merge them together, the different data sources you’d still go back to classic methods, and in that sense, you were limited to some extent. But the frontier of science and computational science is now progressing beyond that. I would assume that the area of big advancement in that field is still not the hospital environment, but however, in pharmaceutical RND.
And I am looking forward to seeing more biosensors and more ways to efficiently capture endpoints and patient biomarkers in order to really implement these in clinical care, which today I find fairly limited.
I have a question from the online slide, which has a lot of votes. What has been the reaction of the employees who see parts of their job being automated, and how do you communicate upskilling opportunities to your workforce?
I’ll take that one because I think I’ll probably get more on the automation side than my other two panelists. So most of the time, people are excited to see their job automated because we’re automating jobs that are traditionally very difficult jobs to do, maybe an ergonomically challenging job or have some other constraints that’s around it. Now, as far as the redeployment of that workforce, there is all kinds of opportunities that are communicated within these, and I’ll take the engineering group as an example.
Within engineering, we have a group of frontline technicians that we brought online and work very closely with our engineers as they know the production processes better than anyone. So they work with frontline technicians, back with the engineers to go out and solve problems. So those were opportunities that we create as we automate jobs, and repurpose people to a high-value use. Similar to the example that I used with the maintenance technicians and the predictive analytics around robotics.
One more online, and then one more for the audience. [INAUDIBLE].
Yeah, absolutely. So there’s a shortage of labor around technical skills in the US like welding and nursing. So why do you think that students are not enrolling in vocational schools to fill those gaps?
Probably take that one as well. Yeah, I’ve made the comments locally within Tennessee. A lot of that I think is a societal thing, because we say go to school, get a four-year degree, get a good job. And we don’t talk about the vocational opportunities that are out there. That welders, and pipefitters, and the skilled trades. You can make a very good living doing those. And it’s one of the things that I’m proud to say in Tennessee that has a lot of push behind it, because those skilled trades are in short supply, and they are very good-paying jobs. And we have many initiatives to help support students that want to take that direction.
But it’s– that’s definitely something that as a country we have to continue to promote and celebrate whenever people make the decision to go and get a skilled trades vocational degree.
One more final question from the audience here.
Hi. [INAUDIBLE] fellow from MIT. Last month publication from researchers at UC Berkeley, UC Chicago, and [INAUDIBLE] health care found that several artificially intelligent algorithms handling the health care of over 200 million people here in the US were infected with the same racism and racial bias that human decision-makers have. So basically, the algorithms where denying critical medical interventions to black patients that were sicker than none black patients. So the question that I have probably for Ittai and David as well. So how– what you as a senior leader is actually specifically doing in order to identify and tackle the automation of discriminatory practices?
- Another excellent question. I’d say that each type of use case has a potential for different [? risks ?] and different biases. The main two that I’m the most concerned about these days is the agnostics in terms of generalizability across different populations, and being able to provide quality of care for a diverse set of– an equitable care for diversified patient population. We are managing that by reaching out to additional data providers I’d say, and identifying different strategies, and technologies, such as federated learning. And ways in which you can improve the robustness of your product without actually jeopardizing the safety and privacy of a patient data.
There’s another aspect which is more hospital operations management in which you could make optimization decisions that would be unequitable to different socio-economic strata of society. In that aspect, I can say that part of it is about the thinking, about how you minimize your cost function, and which techniques and which data points you would do or wouldn’t do. And where you’d maybe take from the computational sciences, you’d stay at the local minimum rather than the global minimum if it does, in fact, create inequality.
And I’d also say that the positive, optimistic note on that is that all of our commercial sponsors we are also involved in collaborating within developing these solutions are actually extremely attentive to that issue, and are actually interested in resolving that ahead of time and making sure that models can’t be actually used on to [INAUDIBLE] purposes. Some of it is because of reputational risk, and some of it is because I think the real understanding that we are in it in order to improve throughput in hospital operations, but also maintain a very high level of respect towards human dignity and cultural diversity.
Great. Great answer to a very important question. And on that note, please join me in thanking our panel for a wonderful conversation.
[CLAPPING]