Causal inference

With all the LLM hype, it’s worth remembering that enterprise stakeholders want answers to “why” questions. Enter causal inference. Paul Hünermund has been doing research and writing on this topic for some time and joins us to introduce the topic. He also shares some relevant trends and some tips for getting started with methods including double machine learning, experimentation, difference-in-difference, and more. Sponsors: Fastly https://fastly.com/?utm_source=changelog – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.com https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorshipFly.io https://fly.io/changelog – The home of Changelog.com — Deploy your apps and databases close to your users. In minutes you can run your Ruby, Go, Node, Deno, Python, or Elixir app (and databases!) all over the world. No ops required. Learn more at fly.io/changelog https://fly.io/changelog and check out the speedrun in their docs https://fly.io/docs/speedrun/. Changelog News https://changelog.com/news – A podcast+newsletter combo that’s brief, entertaining & always on-point. Subscribe today https://changelog.com/news. Featuring: Paul Hünermund – Website https://p-hunermund.com, LinkedIn https://www.linkedin.com/in/paul-hunermund, X https://x.com/PHuenermundChris Benson – Website https://chrisbenson.com, GitHub https://github.com/chrisbenson, LinkedIn https://www.linkedin.com/in/chrisbenson, X https://x.com/chrisbensonDaniel Whitenack – Website https://www.datadan.io/, GitHub https://github.com/dwhitena, X https://x.com/dwhitenaShow Notes: How Can Causal Machine Learning Improve Business Decisions? https://www.causalscience.org/blog/how-can-causal-machine-learning-improve-business-decisions/Causal Inference is More than Fitting the Data Well https://www.causalscience.org/blog/causal-inference-is-more-than-fitting-the-data-well/Causal Data Science in Practice https://www.causalscience.org/blog/causal-data-science-in-practice/Causal Discovery https://blog.ml.cmu.edu/2020/08/31/7-causality/DoWhy Github https://github.com/py-why/dowhyThe Book of Why https://www.penguin.co.uk/books/289825/the-book-of-why-by-judea-pearl-and-dana-mackenzie/9780141982410Causal Data Science Meeting https://www.causalscience.org/Paul’s study on causal ML adoption in industry (incl. an overview of useful software packages in Table 3) https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3867326Causal Data Science MOOC on Udemy https://www.udemy.com/course/causal-data-science/Upcoming Events: Register for upcoming webinars here https://practicalai.fm/webinars!

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:07Speaker 00

  Welcome to Practical AI. If you work in artificial intelligence, aspire to, or are curious how AI-related technologies are changing the world, this is the show for you. Thank you to our partners at Fastly for shipping all of our pods super fast to wherever you listen. Check them out at fastly.com. And to our friends at Fly, deploy your app servers and database close to your users. No ops required. Learn more at fly.io.

• 0:43Daniel Whitenack?

  Welcome to another episode of Practical AI. This is Daniel Whitenack. I'm a data scientist building a tool called Prediction Guard. And I'm joined as always by my co-host Chris Benson, who is a tech strategist at Lockheed Martin. How are you doing, Chris?

• 1:00Chris Benson?

  I am doing very well. I've been watching you building Prediction Guard from afar and looking forward to hearing more about it in the days ahead.

• 1:07Daniel Whitenack?

  It's a fun one. We'll talk about it in more detail soon. And the causal reasons why I ended up doing those things that I'm doing.

• 1:19Chris Benson?

  Could that be a transition?

• 1:20Daniel Whitenack?

  Yes. Speaking of cause and effect and causal things, we're really privileged today to have with us Paul Hunermont, who's an assistant professor at Copenhagen Business School. Welcome, Paul.

• 1:35Speaker 03

  Hi, Daniel. Hi, Chris. Thanks for having me.

• 1:38Daniel Whitenack?

  Yeah, yeah, it's great to have you here. And I think also, this is so cool because I think the topic that we're going to talk about is so very practical and important because I've tried to do, as many of our listeners know, my wife owns a business and I've tried to do a bunch of like analytics or like predictive things for her over the years, just out of either need or fun. And often the question is like, why is the prediction that? Or what, like for a business person? And like they're wanting to know what is the attribution? What is the behavior behind this thing that I'm seeing? So you're an expert in causal AI, causal machine learning,

• 2:20Daniel Whitenack?

  and have been doing research in this area and are very well versed in it. And I'm wondering if you can, as you start out here, just like give a brief understanding to everyone about like, what do you mean when you say causal AI or causal machine learning? And how maybe is that differentiated from what people might commonly think of when they think of AI or machine learning?

• 2:46Speaker 03

  So I mean, then many names causal AI, causal machine learning, causal inference, I think, is the more traditional term. But I think the basic idea is pretty intuitive for everyone who works with data is that if we look at correlations and patterns in the data, sometimes they can produce quite surprising and probably nonsense results. I mean, we all know the story about ice cream sales and shark attacks that are highly correlated over the course of the year. Chocolate consumption and Nobel Prize winners in the country, right? Probably driven by Switzerland predominantly. Stocks and the babies, right? Like the stock population and fertility rates are correlated.

• 3:31Speaker 03

  With these examples, we usually use them in the classroom as sort of like a caveat, right? Wait a minute, a correlation is not causation. People have heard this term. But then causal inference and causal machine learning is really this idea of taking causality seriously and trying to build tools, algorithms, that allow you to draw causal inference from data to distinguish cause and effect and weed out the kind of nonsense correlations. And yeah, that comes with a different tool set. So I mean, you can approach this from a purely algorithmic point of view and you would probably apply different tools than than standard machine learning.

• 4:14Speaker 03

  It goes even deeper. There's a whole epistemological point about it. Well, if you want to do causal inference, you cannot do this in a purely model three way. You actually need background knowledge, expert domain knowledge in order to do this, in order to, for example, distinguish between possible alternative explanations. And that is a whole paradigm shift in terms of how we approach data and how we approach machine learning. Well, and then my last word on this is what's the difference to standard machine learning is that, well, standard machine learning, the bulk of it is really correlation based. I mean, all the tools that we're having, deep learning, support vector machines and so on.

• 4:57Speaker 03

  These are predictive tools, prediction means correlation, finding, detecting patterns in data. And so they're not suitable for it. I mean, there's a branch of AI that is reinforcement learning. Maybe we could talk about this later, that goes more into the direction of actually intervening yourself. The learner intervenes itself in the environment. So that goes in the right direction, but is not getting there full way. This is the main difference to standard machine learning.

• 5:27Daniel Whitenack?

  That kind of gets to the what, I guess, of what is causal AI, causal machine learning, causal inference? I guess the next question that's probably a good foundational one is the why. And maybe this is even exacerbated in recent times. I don't know if you've seen this with all of this sort of hype around large language models that are incredibly non-interpretable or produce, you know, things that are very factually incorrect or unexplainable. But how should a data scientist working in an enterprise, let's say, like, why should they care about causal inference rather than just making good predictions, let's say?

• 6:09Speaker 03

  Yeah, so maybe it helps if I've defined first what I mean exactly with causal influence, because there's actually neat definition by James Woodward, which is a philosopher of science, who said that causal influence or causal machine learning is a special kind of prediction problem. So in a sense, it is a prediction problem. But here we are predicting the likely impact of an action, intervention or manipulation. Really this idea of I do something like I increase the chocolate consumption in a country, will that produce more Nobel Prize winners, right, in this context? I think if we approach from that perspective, you immediately see the value for business. Because in business, we're always asking these kind of questions.

• 6:54Speaker 03

  What if we do X? What if we implement this new HR policy? What if we enter a new market? Should we invest in this product or another product? So business always involves actions, interventions, and we want to forecast, predict the likely outcomes of this. That would be, well, what causal influence people call the interventional level. Then one level above that is the counterfactual level. So counterfactual meaning we're reasoning about Tuesdays of the world. Had I not taken the aspirin this morning, would my headache be worse today still? These kind of questions.

• 7:37Speaker 03

  And they are also very relevant in sort of hindsight retrospective. Like, was it again HR policy that we implemented that improved employee satisfaction and so forth? So immediately relevant in all sorts of domains in the business world. Specifically in AI, we're talking about fundamental problems in AI, which is fairness, robustness, explainability. And I believe that causal AI has to say something in all of these domains. So there's also an immediate practical value in this.

• 8:11Chris Benson?

  Let me ask you a question on, I want to throw another term that we haven't used yet that sometimes gets thrown in in casual causal conversations, say that 10 times fast, determinism versus non-determinism, because we have a habit of applying that at a high level with AI models and say, ah, they're non-deterministic. And, and there's a certain expectation over the years in training AI models that are non-deterministic in terms of, you do have that disconnect in terms of understanding that causality from beginning to end. Can you kind of distinguish a little bit between the two terms in the sense of if someone is just kind of getting into this, they're early in data science, and they're trying to go, wait a minute, I thought AI was non-deterministic, but yet causality

• 8:55Chris Benson?

  is explainable. How do those fit together? How do those as slices of perspective on an AI model, how do they work together where you have determinism, non-determinism, and causality or not? And what are the implications of those?

• 9:08Speaker 03

  When I put out the definition, I used the term likely impact. And that already hints that that also causal inference, the approaches that we have in our probabilistic framework. So there's no determinism in this. What we're interested in is still a probability or contrast of two probabilities, right? The probability of a certain outcome that I care about if I had taken the specific action or if I hadn't. So these are factual question, but it's still, there's no determinism in the sense that if I implement something, it will always work, or we will always have success with this product and so forth. There's an interesting, I think, sort of intellectual history, because the frameworks

• 9:50Speaker 03

  that we're having in causal inference directed acyclic graphs, developed by Julia Pearl and these kind of people, they built on actually earlier work in AI, like Bayesian nets, for example, that were at that point, still a purely predictive tool, but a way or tool to deal with complexity in terms of probabilities, because their expert systems were too rigid. We figured that out in the 70s and 80s. And building on that, then people immediately, once you reason probabilistically- People made the shortcut, the mental shortcut in reasoning in terms of cause and effect, probably because it's so intuitive to us, but the tools were actually not ready

• 10:32Speaker 03

  for it yet. So that was the intellectual history, how we moved from probabilistic AI frameworks to causal inference. And again, I think people immediately started to think that way because causality is such a fundamental concept for human thinking, right? We learn it very early in our development. Babies can't think causally. There's some psychology work that we picked up at the age of two or so. Pets sometimes can't think causally probably. So it's a very fundamental concept.

• 11:00Daniel Whitenack?

  Have you found like in practically interacting, because I know you're involved sort of in the data science community as well and have helped run events and other things related to this topic. Have you found data scientists are sort of, because I could see some data scientists were so in the mindset of like, we're making a prediction, we probably understand that we're thinking about correlations in many cases, I think. But then it's sort of scary for us to think about like, well, I don't know if I want to like put out a, if I tell my executive, like this is the reason that something happened, like I can see the

• 11:41Daniel Whitenack?

  value of it, but how confident can I be in that? And that also gets to maybe people's, I think people during COVID and other times realized maybe how rusty they were on like basic statistical and probable probabilistic type of concepts where, you know, everyone was all of a sudden thinking about medical trials and such. Have you found this sort of hesitation amongst data scientists as you've interacted with them? And what maybe is some steps that data scientists can take to gain confidence in initial thinking and education around this topic?

• 12:18Speaker 03

  Yeah, so we talked with a lot of data scientists from industry practitioners and I don't think there's hesitation. It's actually the opposite. There's lots of interest. Of course, this is sort of a new topic. You need to tool up in a different area. So I mean, that's a step that you need to take. But many people are very curious and we just simply wanted to understand sort of where are we right now? And we had a hunch that this is the toolbox that we know predictive analytics, correlational AI. And then based on that, what kind of questions practically do you address? Also, maybe in the interplay with the broader organization, right? What is it

• 13:01Speaker 03

  that the executives want to know? What do they approach you with? And is there sort of a mismatch between the methods that you're working with and the questions that are asked? And in our interviews, and we did some qualitative analysis on this too, we could clearly see this kind of mismatch between many questions that are asked, do we actually have this causal component to it? Because actions forecasting interventions is so wicked. The standard tools, right, are not up to the task for this. And that creates actually this this interest in approaching causal inference and looking beyond what we currently do. So one interviewer is stuck to my mind was a IT consultant that was working a lot in in the data science

• 13:45Speaker 03

  field. And he said, like, yeah, most of the questions that our clients asked are causal questions in the end. But what we do in the end with them is always some form of predictive analytics, deep learning and so forth. That always created this kind of tension in the projects that he was working in. So that was very eye opening for us.

• 14:11Speaker 00

  It is now time for a changelog news break. The team at Suno AI is helping change the game in text to speech realism by releasing Bark, a transformer based text to audio model that can generate highly realistic multilingual speech as well as other audio, including music, background noise, and simple sound effects. It can also laugh, sigh, cry, and make other nonword sounds that people make. Crazy, right? Here's an example that includes sad and sighs meta tags.

• 14:45Speaker 04

  My friend's bakery burned down last night.

• 14:50Speaker 04

  Now, his business is toast.

• 14:53Speaker 00

  And here's one more with laughter.

• 14:55Speaker 05

  I don't like PyTorch, Kubernetes, or Schnitzel. And xylophones flummox me.

• 15:04Speaker 00

  You can still hear some digital artifacts and blips here and there, but we're getting closer to synthesized audio that's indistinguishable from the real thing. And that's cool slash scary. You just heard one of our five top stories from Monday's Change Log News. Subscribe to the podcast to get all of the week's top stories and pop your email address in at changelog.com/news to also receive our free companion email with even more developer news worth your attention. Once again, that's changelog.com/news.

• 15:53Daniel Whitenack?

  Well, Paul, you've described really well how to think about generally this sort of causal inference, causal AI, causal machine learning, the importance of it. And you mentioned that in doing causal inference, you have sort of a different toolset or maybe different algorithms that are applied. I know that one thing that of course I've done before and know about from various data science positions is like experimentation or hypothesis testing, like A/B testing. I know that only scratches the surface. You were talking about, you know, directed acyclic graphs and other things. So could you give us like a broad

• 16:35Daniel Whitenack?

  sketch of like currently, what are the like main categories of approaches within causal inference and how can we think about those?

• 16:46Speaker 03

  Like from really broad categorization. Traditionally, people divided the field into experimental and observational methods. And experimental would be the A B testing that you're talking about. One of our interviewers even called it the big hammer that tech companies swing around, A B testing, and it's applied a lot, sometimes together with some form of multi-armed bandit reinforcement learning type of approaches, but often just in a plain vanilla way. And that's great because, well, experiments are easy in many domains to set up, easy to understand, and you don't need a lot of background knowledge for it.

• 17:30Speaker 03

  You'd simply try out different things, shades of a button on a website, classic example. But in other domains, it's really not that simple because, well, experiments can be very costly, they can be unethical in many questions. I think you mentioned the COVID pandemic earlier. That was an interesting example to observe because when we tested the vaccines, of course, we did the standard clinical trials, which is an experimental method, and A B testing, if you want, right? That costs a lot of money, but we have these procedures for it and we need to approve drugs in that way. But then after we rolled out the vaccines, immediately there were follow-up questions, like, for example, where's the vaccine more effective?

• 18:13Speaker 03

  Is it for older population or a younger population? Or in which way do we need to roll out SARS vaccines and so forth? These kind of questions, they were not included in the control trial. We didn't have experimental evidence for it. We needed to answer this based on ex post data, so people picking up vaccines and then seeing where they're most effective. That was interesting to see because many of the questions that we ask in practice do involve this observational causal inference, and with observational causal inference, I mean, we don't actively intervene ourselves, but we passively observe the data and still want to get cause and effect out of it, although we haven't designed the experiment ourselves.

• 18:56Speaker 03

  So in a sense, we're then trying to mimic a thought experiment, if you want, with observational data. And that creates all sorts of problems because, um, well, those people that picked up the vaccine earlier probably are those that thought they had the most to gain from it, for example. So there's this sort of self-selection bias or confounding bias in this. And we need to address all of these things. These are the two main categories. And then within those categories, we have all sorts of different techniques, algorithms, like experimental design is an entire course catalog at our university. In the observational field, so for example, I originally come from an econometrics background.

• 19:39Speaker 03

  And in econometrics or in economics, we ask a lot of causal questions and then we have tools like regression discontinuity design, difference in differences, nearest neighbor matching and so forth. The new kid on the block are the computer scientists and they catch up fast in causal inference and they develop these techniques like directed acyclic graphs, causal reinforcement learning. So all sorts of exciting streams of literature coming up these days.

• 20:09Chris Benson?

  I'm really trying to absorb what you're saying and it's very interesting. I'm kind of wondering if I have a problem today, uh, like before our conversation and I want to go through your kind of the typical data prep and, you know, model training and model testing to deploy. But now I've listened to you and I want to start implementing causal approaches into my workflow. How does my workflow change? What does it look like with typical tools now? And where might gaps be in the typical tool chain that we currently have? How do we make it practical and go do it after the show?

• 20:41Speaker 03

  Starting from the epistemological challenge is that we cannot do causal inference in a purely data-driven way. We cannot just optimize a target function and or look at our confusion matrix, loss function. In that sense, we need to complement this with background knowledge that, well, in the simple examples, right? It's not just ice cream sales and shark attacks. There's a third variable lurking, right? Uh, which we need to consider, which is weather, probably or sunshine. And so this is in the simple case, but uh, now imagine a problem that you approach for the first time, right? You do exploratory research, so you don't have this good theory. So we need to do something about this.

• 21:24Speaker 03

  Then a lot of the standard challenge that we're having, right? Collecting good data, maybe designing an A B test are the same. But this additional step of bringing in background knowledge. And there it depends, I guess, right? Like how also the data science team is structured in an organization? Do we need to bring in outside stakeholders? Do we maybe need to talk with the marketing people or the logistics people depending on the project? Often at the moment, data science teams are almost this kind of in-house consulting type, right? And there, for example, not that many mixed teams that could bring in this background expert domain knowledge.

• 22:03Speaker 03

  Practically speaking, I mean, there are all sorts of tools out there in the standard software languages. Like it's a little bit scattered probably the landscape. So you really need to know what kind of libraries are out there. For example, in Python, the do Y package by Microsoft really became a standard industry wide. Because they also have this kind of causal inference pipeline implemented in the package, which starts from sort of modeling a specific domain or phenomenon to applying the causal inference algorithms, getting causal effects out, and then also refuting the model or challenging the model. So you have this kind of step-by-step procedure

• 22:45Speaker 03

  that that can really help you in getting started and getting results quickly.

• 22:50Chris Benson?

  One of the things that you said that I wanted to ask kind of a clarifying question was, you kind of talked about, you know, going to that kind of external source, the extra authority, if you will. A lot of practitioners, you know, these days are kind of starting on their own. If they're not on a big data science team, like, you know, I, I work in a big company and we have tons of data scientists. So this probably doesn't apply to me in that capacity. But a lot of people in startups are out there trying to kind of delve into new businesses and stuff. And they may not have access to kind of outside the data expertise to apply. Do you have any tips or guidance on like, if you're that practitioner and you're trying to solve a problem for which you don't have that external expertise? How would you go about tackling that?

• 23:34Chris Benson?

  How would you go about saying, it's me, myself and I, you know, joking around and I'm gonna, this is a way I can apply causal approaches when I don't have a lot of resources available to me?

• 23:44Speaker 03

  First of all, I would say you're never really alone, right?

• 23:47Daniel Whitenack?

  True.

• 23:48Speaker 03

  So I think outside of the box a little bit. I mean, often it doesn't take that much. You can just approach people and maybe talk with them for an hour and get the insights out that you need. Consulting, probably the scientific literature on a certain topic can help too in sort of figuring out alternative explanations that you can then bring to the data and test to the data. We're also not completely helpless in a sense that everything has to come from data. There are data-driven approaches, so that would be the area of causal discovery that we can apply to get closer to kind of a causal model based on the relationships that we find in the data. We know that that never gets us 100% all the way.

• 24:32Speaker 03

  So we will always need to complement it with some form of background knowledge, but it can already help. And then I would say, I mean, talking also to practitioners, I think sort of the AG20 rule, I think it's called, right? Applies, I mean, already getting closer to something causal is often good enough. And we should get away from this idea that it's zero or one, right? That either it's causal or it's not. Often we get closer to the truth. And if not, we can do, for example, there are whole tools on sensitivity analysis that we can challenge our assumptions and see how robust they are. And I think in practice, this already helps tremendously.

• 25:10Daniel Whitenack?

  You mentioned kind of reaching out to practitioners and the community around this. Could you describe a little bit? I know, like I mentioned earlier, you've, there are some resources that you've kind of helped cofound and run over time related to this. Could you mention those so that people could find those as they're looking into the topic?

• 25:29Speaker 03

  Based on what we identified in where the field is, we actually saw the need for more exchange between different academic fields because causal inference is such a general purpose technology almost. It's applied in various different fields. I had mentioned economics, computer science, epidemiology, health sciences, but then also practitioners. So it's really like a mixed group. So we set up the annual causal data science meeting. We started in 2020, so had to do it online because of the COVID pandemic and then realized that this really An easy way to get people into one, well, virtual room in this case, and there was lots of interest from practitioners. And we're gonna have the third iteration of this this year in November.

• 26:14Speaker 03

  So still some time, but hopefully listeners will make a mental note. Well, there are also good teaching tutorials out there, many blog posts, online courses that you can sign up to. Books like The Book of Why by Judea Pearl, maybe not really a textbook, but really drives home the idea of why causal influence is so important. It has really nice historical anecdotes because Judas is really a giant in this field. Causal Inference, The Mixtape by Scott Cunningham. If you have more of an econ background, perhaps. The Effect by Nick Huntington Klein. So these are all beginner-friendly textbooks that you can pick up. And then trying out the different packages like Dewey and Python, for example.

• 26:58Speaker 03

  There's this startup called Geminos that is developing causal influence software. And they are having free trial versions where you can start out drawing your directed acyclic graphs and see how answers change if you change assumptions, for example. So I think that is usually the best way to learn and pick this up.

• 27:35Daniel Whitenack?

  Well, Paul, I'm selfishly going to present you maybe with a scenario and do some sort of on-the-fly problem solving. I figure you're probably good at that, being a professor and always solving problems with students and others and colleagues. So I mentioned my wife runs a business. It's a candle manufacturing business. And there's actually this sort of like why question that we've been talking about a little bit. So last year, to give context, I just logged into Shopify. So last year, they had eighty-seven thousand eight hundred and thirty-seven orders.

• 28:13Daniel Whitenack?

  Each of those orders, or at least most of them, when they ship them, included a free sample two-ounce candle. It's like a freebie add-on. And over time, the assumption has always been, oh, people really like that. And it's sort of like part of the package that they get. It increases reorder value, right? Like, they see the package and they're like, oh, cool. I got like a free candle and now I'm gonna, I love these people forever and I'm gonna reorder, right? Well, the question has come up, obviously, at this scale of orders, that's a lot of free two-ounces. And even just the savings of those would be huge. So how might you, as a practitioner

• 28:58Daniel Whitenack?

  or someone thinking about this problem, both in terms of the experimental or the observational approaches, what might be some ways to dig into this? Obviously, it's very expensive to send out like, get different packaging and do like a experiment at that scale. So it'd be nice to know without doing like a large-scale shift in packaging and that sort of thing. Any tips for me?

• 29:22Speaker 03

  In this case, the big advantage is that you or your wife, you are actually controlling this process, right? So you decide in which packaging to put the free add-on. And in that sense, you immediately understand the selection process or the treatment assignment, how we would call that here. And in that situation, I think an experimental approach would be the way to go. Then it becomes more of a statistical question, like how much your sample or how large your sample needs to be in order to draw robust conclusions, right? And if it's just the yes/no questions about a free sample or not, the experiment can probably be quite small.

• 30:06Speaker 03

  That relates a little bit to the COVID example that I discussed earlier. Probably you want to broaden that up and think about, for example, heterogeneous treatment effects in the sense that, well, is it high-volume customers that like this free add-on most? Or is it... More like the casual shoppers, right? And suddenly you have four groups that you're catering to because well, high volume, low volume and treatment and control. These are problems that always come up in causal inference. So the tools like causal random forest developed exactly for that problem. How do you efficiently partition the the population

• 30:46Speaker 03

  in order to reduce costs associated with an experiment, right? You want to be as cost efficient as possible, but also still get robust conclusions out. A similar problem arises then of, I mentioned earlier, robustness of findings, right? So transfer learning is a big topic in AI. So maybe let's assume you've done this experiment at this point in time and you found that robust treatment effect, right? Like, so people react to it, it increases reorder value. The question is then, in six months from now, will the world have changed or will these results still be valid, right?

• 31:26Speaker 03

  And maybe you're thinking about that in six months, not so much has changed of the business, right? But for example, platforms like booking.com, right? Hotel bookings for them is very relevant because you have people that book hotels in summer for leisure travel are very different than business travel, for example. So you have this kind of problem. Can you transfer causal knowledge that you've obtained to a different domain? Because that would save you on experimentation costs, for example. So, yeah, all sorts of interesting question in that domain. But the big advantage is you you can actually run experiments here. Like in other domains, we are reliant on data where people self-select into something like, for example, the standard

• 32:11Speaker 03

  question about in economics about, for example, the returns to college education, right? There we could not randomly assign people to colleges, right? Or whether they can go to college or not. There we have to rely on them self-selecting into a sort of treatment and control group. And then it's always the question whether we we have really an apples to apples comparison or is it perhaps an apples to oranges?

• 32:35Chris Benson?

  I just wanted to say, I think you're lucky that you got Daniel's example question because coming from my industry, I would have had to ask like, I don't know, hypersonic missile design or something. And I don't think we want to go there. So

• 32:47Daniel Whitenack?

  This is a great thing about the podcast, right? We get to have like the expert on and I get to selfishly ask the question that helps me in my day, day to day. So

• 32:57Chris Benson?

  Excellent way of getting some free consulting in there.

• 32:59Daniel Whitenack?

  Yeah.

• 33:01Chris Benson?

  So wanted to actually take you back to something that you mentioned a little while ago. We were kind of talking about the benefits of causal inference and you brought up reinforcement learning, but we were generally talking about kind of fairness, bias, robustness, the impact of causal on those. Could you kind of go back to that point and kind of talk a little bit about what that means? You know, these are huge topics that are in all of the different branches of AI right now. And it's on everyone's mind, especially with all the advances this year. How does causal affect that worldview of doing these amazing things in these different branches of AI, but doing it without bias, doing it fairly, such as that?

• 33:42Speaker 03

  I'll start with fairness, because that's actually the very first example that I use in my own course, causality, causal inference course here at Copenhagen Business School. It's a case taken from Google, actually, so a while ago, I think in 2019. Well, already earlier, the story goes longer, but they have been accused of underpaying women in their organization, right? So there we have a classic example of like a protected attribute like gender, race, and so forth. And we want to prevent bias in some form of automated or semi automated decision making, right? And that comes up all the time. I mean, in loan acceptance models, for example, we want to remove bias and so forth. So to make the story

• 34:24Speaker 03

  quick is they have been accused of underpaying women in our organization. Then they did a fairly sophisticated analysis, published a white paper, and the result was of that analysis that they found that they're actually underpaying men, at least they thought so. And not only men, but actually high level software engineers, so high seniority software engineers. At Google. And then because they're committed to fairness in their organization, they actually raised salary levels for these high level software engineers. Based on analysis, so it also had a practical component to it or like a policy implication. We cannot analyze this case here in detail, but if you do that analysis is very likely that

• 35:06Speaker 03

  they actually did a sort of fairly common causal inference mistakes. So they conditioned on some variables that are downstream of so that are affected by gender like occupation, for example, right? And then if you have discrimination already at that stage, that for example, women don't have it so easy to get into high level positions for various reasons that we know of. Then that will be a classic mistake and you can produce this kind of again nonsensical correlations in the end, like the sharks and the ice cream. That's one example that can actually easily transport to other kinds of questions like I mentioned, algorithmic bias.

• 35:45Speaker 03

  And that's a causal question because if you don't understand how variables in your model causally interact and relate to each other. You cannot answer this question, you cannot decide how to correctly analyze the data. Robustness, I mentioned, so the the transportability, transfer learning kind of aspect of experimental knowledge and their causal inference techniques have been developed. Also dealing with selection bias in data, so data set that might not be a representative sample of the population that you care about, but is measured with some form of selection bias because only happy customers answer your consumer survey. Or unhappy customers, but no one in between, right, for these questions.

• 36:28Speaker 03

  And then lastly, explainability, I think explainability almost comes for free with causal inference. I mean, don't get me wrong, causal inference is a hard task, but once you solve it, explainability almost comes for free because, well, I mentioned the book of wine, right? So causal questions are always related to why questions, counterfactual as well, right? Like, why did my headache go away? Was it because I took the aspirin this morning? Uh, I mentioned this example, this is the way we reason, this is the way we explain, for example, things to other humans. And so there's an immediate connection to explainability.

• 37:05Daniel Whitenack?

  That's a really great way to think about this and it gets me thinking like what will be the impacts of these two fields as they interact more over the coming years and I'm wondering from your perspective because you're so plugged into the research that's going on in this area but also the practical side of this and how data scientists are beginning to use these techniques. What as you look forward to the next, let's say, year or whatever time period you want to have there, like what gets you excited or what trends would you like to highlight that maybe people should be thinking about in this field or maybe it's just things that you're excited about in terms of new, new opportunities or new methods or whatever it might be?

• 37:50Speaker 03

  Yeah, I just attended a conference last week in Tübingen, Germany, the causal learning and reasoning conference and it was just exciting to see how many young minds there were attending the conference. So it was a very young audience, a lot of grad students in computer science specifically there and a bunch of seniors as well, but that really showed me that this seems to be the next big thing in AI and people have confirmed that to us, that there's more and more interest in the academic side, but we also see that in practice in the industry. Yeah, so I'm excited about, well, experimental design,

• 38:31Speaker 03

  what I mentioned earlier, right? Heterogeneous treatment effects, for example, not only being satisfied with having one average treatment effect or average causal effect, right? One number and this is what we're expecting, but actually making this more fine-grained and opening up, answering questions like, is it old people, young people that benefit more from vaccines? In which way do we need to roll that out in the most effective way? Then on the observational side, I think causal discovery is really promising and this is really the idea of how far can we go with just simply trying to get out causality from observational data. And we will never get 100%, I mentioned that, but how far can we go?

• 39:15Speaker 03

  So one big challenge in that area is, for example, to have good benchmarking data sets, right? In machine learning, that's usually easy. You divide a sample up into a training and benchmarking data set. With causality, that's not so easy. You often need an experimental benchmark, for example. A lot of work has been done in genomic research where you can knock off genes, for example, in an experimental way. So that is really exciting. There's new work on causal root cause analysis by Amazon, for example, so figuring out what are the causes of actually outliers, even in an engineering system. So you mentioned, Chris, that you're working in that area. So I've seen, for example, companies in the defense industry thinking about this problem of root cause analysis.

• 40:00Speaker 03

  Lastly, perhaps, because originally, before I came to causal inference, I was actually trained in economics, like I mentioned, but specifically innovation economics. So this idea of how do we produce knowledge as a society? How does knowledge spread across a society? And so in causal inference, there's new lines of work thinking about actually interactions between treatments. So not just the idea that I take a pill and I get an outcome from that, but it's like you take a pill and that reduces the viral load in our community and that's why I actually have a lower likelihood to get sick, for example. So these kind of interactions between people are really important, I think, in many domains and specifically also in the way knowledge spreads across networks.

• 40:45Speaker 03

  So that is something I'm really excited about.

• 40:48Daniel Whitenack?

  Awesome. Well, I am really, really happy that we got to have this conversation on the podcast because I think it highlights something that's a real compliment to many of the things that people are exploring around deep learning and large language models and other things. This is a really important piece of the sort of practical side of what data scientists are doing in the enterprise. So yeah, thank you so much and thank you for your research on the topic and also engaging the community around this. It's really great and really happy to have had you on the podcast.

• 41:22Speaker 03

  Thank you. I really enjoyed the conversation.

• 41:33Speaker 00

  Thank you for listening to Practical AI. Your next step is to subscribe now if you haven't already. And if you're a longtime listener of the show, help us reach more people by sharing Practical AI with your friends and colleagues. Thanks once again to Fastly and Fly for partnering with us to bring you all Chang Talk podcasts. Check out what they're up to at fastly.com and fly.io. And to our beat freak in residence, Breakmaster Cylinder, for continuously cranking out the best beats in the biz. That's all for now. We'll talk to you again next time.