Unlocking the Advantages of AI (With Pini Ben-Or)

Clay Hausmann:
Hi, I’m Clay Hausmann, CMO of Aktana, and the host of this podcast, Contextual Intelligence. One of the common myths about AI is the fear of a robot takeover or put differently, the fear that AI will replace humans in the workplace. And while AI does have some very real advantages, it also depends on human inputs to be effective. AI cannot create new knowledge on its own, or provide context to ensure accuracy or navigate a crisis.
It’s why commercial airlines depend on the presence of human pilots to fly planes. A machine can handle autopilot or flag a mechanical error assist with troubleshooting, but when an unexpected situation arises, human expertise and judgment are still required to help make those critical decisions. We all know from personal experience that AI is excellent at solving closed problems. Those are things like recommending content on Netflix or recognizing human faces in a crowd or translating languages.
But open problems tend to be far more complex than context dependent. Our guest today happens to be the perfect person to help us unpack this complexity, Pini Ben-Or, Chief Science Officer at Aktana. Pini is an expert in machine learning and specifically its application to business decision-making, business intelligence, data management, and optimizing operations and marketing channels. He has advanced degrees in physics, mathematics and philosophy where his research has focused on decision theory, AI and philosophy of physics. So Pini with that welcome, and thank you for joining us today.
Pini Ben-Or:
Thank you Clay.
Clay Hausmann:
So let me start with something very early, very basic. When did you first come across or how did you initially get interested in artificial intelligence?
Pini Ben-Or:
I think it was when I was an undergrad in the Hebrew University in Jerusalem on a beautiful building overlooking the desert and the Dead Sea. I took a course in cognitive psychology. And then I also took a bunch of courses in logic, mathematical logic, which made me think about algorithms and automation. You combine that with cognitive psychology, it makes you think about AI. That’s how it got me started.
Clay Hausmann:
And in those early days, what was your perspective as an undergrad? Did you think it was the beginning of something big and this would shape our way of life years into the future? Or were you not quite thinking that grandly?
Pini Ben-Or:
I was not thinking that grandly, but I started to think about it because it’s also connected with me. As a teenager, I used to read the science fiction, Isaac Asimov and The Three Laws of Robotics and The Psychology Of Robots, those kinds of things. And I started to see that these things are connected. Later on when I came to graduate school, this became more serious. So this is the early 1980s, it was a revolutionary era in AI with expert systems. Then I started doing real research in expert systems and statistical or inductive thinking in expert systems, so that was already real AI.
Clay Hausmann:
So let’s fast forward to where we are today. And instead of asking you to remember what it was like to be an undergrad looking forward, why don’t you look back from where we are today? And if you think about where we are in comparison to one or two decades ago, how do you see what we’re able to accomplish and what companies are able to do with AI in terms of the problems they’re able to solve compared to say 10 or 20 years ago?
Pini Ben-Or:
There has been a lot of advances in the last 10 years. The advances in terms of algorithms like deep learning, neural networks, primarily, but many other algorithms for machine learning. And also there has been a great advance in the availability of the technology, not just the algorithms like in papers, but also actual platforms that you can execute and use these algorithms. A lot of open source, many companies, small and large making their tools usable for many other people. So there is a lot of industry-wide and cross industry collaboration. The cost of developing complex applications has come down. So both in terms of the machine learning algorithms and costs of complexity, there has been great advances in obviously computing power. So these are great advances, but if you ask what real AI can we still do? I think that although a lot of progress has been made, there is still a long way to go.
A long way to go, if you want to think about AI applications performing with the level of intelligence that comes any close to humans. And even when it comes to interaction with humans, let’s say like in chatbots, today, you can have a conversation with a chatbot maybe question and answer, maybe a follow up on a question or an answer. And that’s it. You cannot really have a sequence of questions and answers. You can not have anything that’s like a conversation. So the distance between where we are and real complex AI like in contextual conversation is still large distance. But even though the progress is only partial, we are now able to solve more complex problems from a business perspective. So there has been a great advance, maybe a little bit of over promising with all that, but meaningful.
Clay Hausmann:
So I think you touched on a couple of these as you were talking there. But as we read articles now about where AI could be felt are some of the most exciting applications of it, we tend to get lowered to the very big, very difficult problems to solve, driverless cars and the like. What are some of the closer term, kind of the easier advancements do you think we’ll be able to pick off with regard to application of AI and the way that we’ll feel it either in our business lives or our personal lives in the next few years?
Pini Ben-Or:
So one area has to do with predicting behaviors, predicting events that have to do with human behaviors. So there’s definitely been a great advance in that. We can now have models predicting a large variety of behaviors. If you think about response to advertising on the web, that’s done now routinely on a very large scale in our world, in pharma behaviors of HCPs and reps, the engagement of a rep accepting a suggestion or not responding to an email or not these kind of events with sufficient data can be predicted somewhat reliably, sufficiently reliably to work with them in decision systems. So that area definitely is practical and enables usable AI. That doesn’t mean that it’s easy to solve more complex problems. Like the complex problem that we face of coordinating communications, including content in terms of who to talk to and when, across what channel on a daily basis, it’s still a very complex problem to solve.
Clay Hausmann:
Let’s talk a little bit about the market that Aktana serves, the life sciences market. In the current moment, there’s been sort of a growing chorus of voices of confusion, actually about AI amongst the life sciences CIOs. And I think personally that’s in part because they hear different perspectives, they even hear different definitions of the term. What do you think commercial leaders and CIOs need to understand about AI? Let’s start just foundationally in terms of how they might apply it.
Pini Ben-Or:
It’s important to make a distinction between AI and machine learning and within machine learning between any machine learning algorithm and deep learning specifically. The great advances recently have been in deep learning and more generally in machine learning algorithms. And many people confuse deep learning or machine learning with AI. The right way to think about it is that AI as a technology is a larger category that includes as components or enablers, machine learning and deep learning.
But the big question and the big confusion I think is about what is there in AI that’s not machine learning and how, whatever that is works together with machine learning. So that leads to a clarification and when you look at the actual applications of solving complex problems, you will see that AI consists in the combination of technologies and the trick of solving a problem is in specifically how you combine these technologies and they include machine learning, and rule-based systems and tools for experimentation, and business intelligence, and the explanatory dimension of AI, so that your application in some sense can explain its decisions. These are distinct technologies and the trick about solving complex problems with AI is combining them in the right way for the problem.
Clay Hausmann:
What do you think life sciences CIOs need to consider with regard to their own technical environments before making significant AI investments?
Pini Ben-Or:
They have to make sure they’re ready. And the readiness is as a technical dimension and an organizational dimension. On a technical dimension, the most critical element is data. The availability of data and the comprehensiveness of the data, the ability to put all the data that’s required and updated on a frequent schedule and put it all in one place, it’s available to AI applications. It’s a big undertaking and many organizations are not really ready. They may rely on CRM vendors or on their vendors for part of the problem, but they are not ready overall in terms of being able to bring all the data sources together on a regular basis, in a comprehensive and well organized way.
So that’s the key part of the technical readiness. There is also organizational readiness because the problem that we are dealing with crosses multiple organizational functions in terms of its impact and multiple need group areas in the organization need to be involved in solving it, marketing sales, operations, maybe finance, maybe vendors that execute some of the marketing campaigns and so on. So bring to these organizations, so they work together well on an ongoing basis is a challenge. So you have both of these readiness challenges, the organizational and the technical.
Clay Hausmann:
So we talk often obviously on this podcast about context, the name of the podcast is Contextual Intelligence, of course. And so as you think about how it’s applied or coupled with intelligence, it’s not a new concept. It’s existed in virtually every industry in some form, but why is it so critical in the commercial process for life sciences companies?
Pini Ben-Or:
Maybe we should talk a little bit about what context means. A simple definition is everything that’s relevant to the conversation at hand or to the problem at hand. All the information that’s relevant. That’s a strong and clear definition, but it’s very broad. But even with that, it’s clear that when you look at the pharma commercial world and the problem of coordinating communications across many channels, the context is large and complex. So one dimension of it, or one aspect of it is the customers, who are the customers here. You want to know about the customers, the knowledge about the customers is relevant to any decision you make with respect to how to communicate to them. But in pharma, we have multiple customers. There is the physician, there is the patient and there is the marketing manager or vendor. So you have multiple customers and you need to know the relevant information about all of them at any given time.
That’s one dimension of the complexity and with that comes the data complexity and fact that the data is distributed. Not all of it exists within the purview of anyone partying in the ecosystem. So one reason why it’s especially complex is that. The other dimension is the decision itself, the decision space. If you play a video game, you can move up or down or left and right, and fire or jump or something like that. So how many decisions do you have? Five, six, seven at any moment. Even with the autonomous driving, the number of decisions, the range of decisions is not very large, accelerates, decelerate, stop, turn right, turn left, things of this nature. But if you think of again about the multi-channel coordination problem, many channels, many forms of communications, many different kinds of customers. The content is a very complex dimension. You can talk about many different things in many different sequences. So all of these dimensions make it a more complex problem.
Clay Hausmann:
And I’ve heard you talk often about the concept of relevance and the importance of it, when thinking about communication with HCPs. Can you expand upon that a bit? Explain what you mean by relevance.
Pini Ben-Or:
That’s a hard question. So here’s one definition, a piece of information is relevant if having it has an impact, if it having it makes a difference. So we, humans are very, very good at identifying very quickly what’s relevant. We have enormous memory. We know a lot. Okay, it’s a doubtless, at least. We know enormous amount about many, many things, but when we solve daily problems in any interaction we have with the world or with others, we are very good at only calling in the relevant, the impactful pieces of information from our memory and our knowledge. Machines are very bad at that.
So when you do a search, if the search engine to you, the more we knew about you and about the context of your problem, or the reason why you’re searching for something, the narrower would the search be. So luckily the search engine is very powerful and can do things very fast, so it may not rely on knowing you much. So it does not have to know much about relevance, but if you want to be efficient, you need strong ability of narrowing down quickly on those pieces of information that matter. And that’s a difficult task for let’s say algorithms. I would say that the problem of focusing on the relevant background knowledge, it’s one way to think about the problem of context. That is the hardest problem in AI.
Clay Hausmann:
So the life sciences industry often trails other industries in adoption of new technology in large part because the ecosystem is so much more complex. And there’s so much more at stake with regard to patient health and wellbeing, as opposed to a pair of shoes or gaming experience, as you mentioned. What do you think the life sciences industry can learn from both the successes and failures that companies and other industries have already experienced?
Pini Ben-Or:
That’s a very interesting question. There are some key lessons one can gather. One is readiness. Readiness, again, in the sense of technical readiness, data integration and in our sense of organizational readiness. And one particular element of that is the technical readiness that comes from being able to deploy machine learning models on a large scale efficiently and reducing the learning cycle time associated with these models. So the learning cycle time is how long does it take from the time that there is something new in a data to the time that your AI has learned from that new information generalized it maybe, and is using it or applying it in production in the world.
One way to think about the learning cycle time is in terms of how long does it take to develop and deploy a model, a new model. So readiness in terms of the ability to build and deploy models and modify them and adjust them rapidly, that’s a key capability. And in some industries that was learned early, like advertising, targeting advertisements on the web, for example, and now there are very large scale recommendation systems that do that. So one of the things that was learned there is the ability to update models in production very rapidly. That’s a key area where the pharma companies have a lot to work on let’s say, or use vendors like us.
Clay Hausmann:
As chief science officer, most of your work obviously involves technology and quite often in its most complicated form. But we’ve talked about the importance of that pairing between technology and human intelligence. What do you think about that human component and the role that people will play and do play in making an AI investment successful?
Pini Ben-Or:
Okay. I would like to answer that in two ways. One is the, let’s say the people dimension in making technology work well. And the other is in the human dimension of working with the technology once it’s done. So these are pretty different things. I start with the second. The second is about how AI is used. And today the successful applications of AI to complex problems are not autonomous AI. They are what we may call, collaborative AI, collaborating AI is where the human works with an application gets advice from it, gets recommendations from it, interacts with it, and in some sense, solve the problem together. So now in our world of pharma recommendation system, let’s say next best action is like that. The AI in our world does not make all decisions on its own. It makes recommendations. There are others who make final decisions, be it marketing managers, or sales reps, and also the AI itself, like I said earlier is a combination of technologies one of which is rule-based systems.
Rule base systems express human expertise in the form of rules and it compliments and envelopes in some sense or constraints or overlaps with the pure machine learning type engines. So that’s another form of collaboration between humans and AI in the sense of how AI is applied. So again, AI, in this case is not just a machine learning model that makes all decisions on its own completely autonomously. It collaborates in various forms with humans. So that’s one dimension to the answer. The other is, in our world of the software company that develops software and deploys with clients, and generally even on the client side, the pharma company may have a data science team that develops some models and implement them internally.
What makes that successful? What makes that successful is not just technology, it’s not just algorithms and databases, there is a human dimension, which is very important. And that has to do with the fact that AI and data science is a very interdisciplinary world. It’s not a single discipline and to develop applications of AI, you have to have people across multiple disciplines working very well together. That is difficult to accomplish. So one challenge of anyone who manages data science teams or AI technology teams is to get people to work well people with different experiences, different disciplines, different languages, different conceptual frameworks. So success in the AI world depends partly on that.
Clay Hausmann:
That’s really interesting. I obviously think mostly about the former, the first category that you were referring to, but there was a lot that makes a ton of sense about how to manage the people who are involved from the technology standpoint, and how they think about that. In terms of that, actually the last question I have for you kind of relates to people and how they work and their behaviors and how quickly they change. So obviously we’re in a very unique time right now, all working from our homes, doing this podcast from our homes.
And that is no different for the impact in terms of how physicians see patients, the way that life science companies communicate with those physicians and others. What predictions or perspectives, let’s say perspectives, it’s a less intimidating word than predictions, about how you think things will evolve, will we see accelerated change because of what this has forced in terms of telehealth or other related aspects of the life sciences market? Or do you think we’ll revert back to old methods once we have the freedom to travel around more, once again?
Pini Ben-Or:
Well, I’m not going to try to make a prediction about how the world generally will change. My guess is that there are going to be many changes in the work patterns, but I’d like to focus on the AI side. There are multiple dimensions here. Again, one is the speed with which AI will evolve. I don’t think this pandemic is going to make a difference to that. Certainly not directly. One thing is clear is that strategically pharma companies are going to accelerate the transition to more digital strategy, to more emphasis on many alternative channels, other than the sales face-to-face visit channel. That is started in different ways and in different degrees of acceleration, let’s say globally, but it’s going to be accelerated further as a result of the current situation. And that change is going to be permanent. And that kind of a change that kind of disruption, I guess, puts pressure on problem solvers to move faster and to innovate.
And in that sense, the effect may be faster innovation in AI, in the pharma world and maybe more broadly. There is an interesting aspect of the recent disruption, which has to do with the effectiveness of current AI and machine learning models. When the data varies very significantly, and the data about behaviors of reps and doctors certainly is different now in the last few months than it was, the model’s that are based on past history are affected and they may be affected pretty negatively in the sense that the quality of decisions that these systems are making is reduced. So everyone who’s using this kind of platforms that depend on data needs to assess the impact and come up with mitigation. Definitely also accelerated innovation. Maybe I can comment a little bit about an analogy I’ve been around and very much in the middle of the financial crisis, 2008 and the years that led to it.
And that’s an example of an environment where the data and the change in the data caused serious problems to the system during the economic crisis and in the time that led to it. Everyone in the financial world used risk models and mortgage pricing models that were based on data from, let’s say 2000, 2005, very different environment and environment with subprime pricing, let’s say in 2006 and seven, and the models they used could not really illuminate the risks in a different environment. Everyone who’s used these models. And that means very large range of agencies and institutions in the financial world, they were blind to the risks. That’s not the only problem that caused the financial crisis in 2008, but it certainly shows how disruption in data or changes in the environment that are reflected in changes in data can create major risks to models that drive decisions. And in some sense, we are in an environment like that now. So we all have to think about the impact and think about how to mitigate it.
Clay Hausmann:
All right. So we always bring on interesting people and we want to get their business perspectives because that is very interesting, but we also want to understand them a bit deeper. We call it in context. So I have a couple of questions here for you. They’re going to give us a little bit of a different view into what has shaped and molded you as a professional or a person. My first question is, who has been an influence on your business career that might surprise us?
Pini Ben-Or:
Okay, I know this will surprise you. So I have a principle in my life that comes from the Talmud, which is, make yourself a rabbi and add a few other things. So everyone should make themselves a rabbi or more. I have a number of teachers that really influenced me and I stuck with them and continue to work with them and think and reflect about my relationship with them over my careers. One of my advisors in graduate school, Isaac Levi, his field was decision theory and he practiced it in a very theoretical way, even though he wrote once about assessing the risk of failure of nuclear plants. But generally he was a theoretician. But I learned a lot from him. And he was a very practical person in many ways. And throughout my career in the business world, after I left graduate school, I keep going back to things I learned from him and thinking about how applicable they are.
Clay Hausmann:
Okay. So if money was not a factor, what profession would you most want to pursue and it can’t be Chief Science Officer at Aktana?
Pini Ben-Or:
Well, there is something I would like to do, but my talents don’t afford it, and that’s be an opera singer. But what I could do, and I would want to do is teach philosophy.
Clay Hausmann:
I have to say Pini, I had good money in my mind on you being like a national park ranger, something similar, given your love for hiking.
Pini Ben-Or:
That’s a hobby, not a profession, I guess.
Clay Hausmann:
Okay. Okay. So what profession would you most not want to pursue no matter what it paid?
Pini Ben-Or:
One thing that I think is I like about my life is that I’ve done many different things. I was in the army. I had many, many different jobs when I was undergrad and also graduate student. I’ve done so many things and some of them involved really, I know dirty things. Okay. Clean up bathrooms, anything like that. Nothing, nothing deters me. So yeah, I don’t want to do that kind of work, but I’m not deterred by anything. So it’s hard for me to say what I really would not want to pursue. It’s a lot easier to say what I want to pursue.
Clay Hausmann:
That’s good. Okay. What is the best book you’ve read recently, and why?
Pini Ben-Or:
I’m reading now a book about William James, the American philosopher. I can’t remember the title, but it’s something about the soul. It’s a bit of a biography and intellectual biography of William James. I’m learning a lot from it about how he developed his ideas and the personal struggles that he went through. So I’ll give one example of something I learned about it from this book, which to me is very striking. He had the crisis in his life, crisis of fatal depression. Some of it was intellectually motivated because determinism was in the air in the middle of the 19th century. And the idea that we are machines, the physical world causes everything that we do, so we would have no freedom, that depressed him a lot. And then he read some work by a French philosopher that taught him that the core of freedom is our freedom to focus our attention and control what we attend to. That is one thing I learned from this book that I think is of great value.
Clay Hausmann:
Pini, it is not surprising what a deep thinker you are. Very encouraging, little intimidating, but not surprising. So now I’m going to challenge your marketing skills here, so you’re at a family gathering and your eight year old nephew asks you what you do for a living, what do you tell him?
Pini Ben-Or:
I tell him that I work with data and I teach robots how to behave.
Clay Hausmann:
That’s good. That’s good. I like it. You might get a couple follow-up questions on that, but that is a good, concise description. All right, last question. We are scheduling your ultimate dinner party for four, who is attending this dinner party, and what is being served?
Pini Ben-Or:
You know, I read this question and I was thinking about what is served, but I was not thinking about who attends. I’m vegetarian. I like Korean vegetarian or vegan food. So it’s going to be someplace like a Korean vegetarian restaurant. There aren’t many of those, but some of them are very good. Who attends? A few of my best friends that I never brought together. Mike Fraily and Frank Ruana or Peter Lupu, very good friends from different areas of my life in my career. That I think shares some interests, but have never met each other.
Clay Hausmann:
I love that. I love that because the tendency is always for people to think about famous philosophers or celebrities or others. I love an answer of three good friends who’ve never met each other, but might learn a lot from one another. So Pini, thank you so much for joining us as always, it’s very enlightening. It was a great conversation and we really appreciate you taking the time.
Pini Ben-Or:
My pleasure. Thank you.
Clay Hausmann:
That’s it for this episode of Contextual Intelligence. I’m your host, Clay Hausmann. And we’ll be back in two weeks with a new episode. In the meantime, you can find all our episodes on Apple podcasts, Stitcher, Spotify, or wherever you get your podcasts. And please leave us a review or a comment or a question or all the above, so we can make sure that this podcast brings the proper context to your work. Thanks everybody for joining us.