pranit<>angie|soc_ent state action examples

    12:09AM Jul 27, 2024

    Speakers:

    Keywords:

    learning

    evolutionary

    people

    experiment

    worked

    situation

    bayesian

    behavior

    data

    information

    energy

    behavioral

    canvassing

    evolution

    boston

    economic impact

    point

    reason

    money

    competitor

    Do we have to pay for that?

    Yeah. But I think there is a free kind of amount of things. Okay. I just wanted to like use too much of one. Yeah. So why don't you walk me through what you wrote and then I can

    basically I will tell you did I send you the highlighted version?

    Yeah, mine's highlighted. Yeah. Like AI

    Okay,

    for the community engagements

    these two are I think a lot paradoxical. Set number, under set number of, they say, accurately measure which is economic impact state. This was slightly slightly arbitrary, because it's very hard to determine how much of an economic impact it has, besides on your energy bill, right? Because yes, okay, your energy bill videos, but what's like the tangible economic empire? Maybe you could like there's this like theories that you know, if you spend more money on like groceries besides, that goes a long way. Especially for low income. Environmental impact. This one's also a little like, like not super easily measurable, but there's ways like carbon credits, etc, etc. That kind of stuff. This one I wasn't super sure about it because operational efficiency right.

    I just realized I need coffee. So let's head to Starbucks.

    Yeah.

    This is way like a lot of information. Yeah, for sure. Sorry, but I was listening. No. No, definitely. No, I can't

    see. So yeah. For operational efficiency. I am not sure how you can like directly see that because one thing I have funny enough, I actually, they called me back to this and I went and talk through all the interns and all in energy. It was virtual.

    Oh, yeah. Yeah.

    But it was it was it was fun. But this was the one thing I realized was operational efficiency is dependent on the people themselves. So it's kind of hard to it goes up in reverse. This one is really as a tangible effect as I could do. A lot of our clients were low income. individuals of color. Primarily moved to Boston. Or like, you know, we're here but we're in like low income areas. And, yeah, feedback and adapt ation. Pretty straightforward. And then stakeholder. This was also an important one. That city partners

    the city parks

    city of Boston

    so yeah,

    these ones are like the real action but they're already doing. These are the actions. Yeah, the red one is the one that I'm not doing. Yeah. And I don't think it's prescribed

    but like something it's nice

    I tell you any candidate points. So

    when you're doing it a different kind of model and usually unit of observation and unit of information.

    So for me, state was like okay, what is presently happening? And like the constant action was okay, what did you do? What did you change? That was

    usually my time precedent, content coming

    in precedent,

    because

    actually, everything changed.

    Yeah,

    you can. Lm walls can change even though probably going to. Everything should be kind of you need to build a tree and you need to find my highest leverage point and then find that. Okay, so my assumption is when you were building the tables, building this table, cells is that if that isn't correct, it would need a lot of right. If you have, like, the theory is the amount of information that

    people have

    how much you committed under the experiment.

    Right, you're doing that's, that's why I did the highlights, right? Because those are the things that they have experimented and from an organization perspective, right is they have experimented and got to from like, you know, either pivoting or like, you know, donor expectations or things as such. But I would say the actions are definitely something they are committed to, through like trial and error and have achieved in to some capacity and there's like a measure or factor. Now stayed. You know, there is a time factor to it, I would certainly say but it's a very long time factor because even if it's a policy change, you're not going to see much. much change overnight. Same thing with like, measuring the energy efficiency, at least. All of it is in a cycle. At the least you're looking at a year.

    That makes sense

    what do you think?

    Should the next step I think, one time like I was

    the director, that's my question.

    How do we measure our success? One thing they preyed upon themselves is like we have reached like communities that not not many NGOs reach that have been like, sort of like not particularly record. It's like a weird issue. Because one on one hand, you are trying to advocate for racial justice and etc. through energy efficiency. So it's like, like goes around. Like, it's not there's not a direct connection. So I would want to know, what is my direct connection to the impact, right? Yes, I am claiming we're helping communities of color. We're having low income communities. But what is the direct impact on kids on the scariest?

    Your.

    Data for our learning happens in a sequence to the experiment

    is that the one here isn't gonna close this

    out auto auto

    auto bye a scientific approach in the primacy of the design of experiments, critical teams

    but more importantly, hypothesis is for impact the barriers to how much this experiment approximates that

    of course, better cheaper

    pretty challenging what is

    that before would be more

    interviewing people, so when they say that they are not in

    the US

    that's kind of your proximity to us if you happen to us up so you have people with the Boston from age approximately Okay. That's the hire like most cases if you increase the number to every people, fidelity the trade off because in that case, the opportunity cost Yeah

    Wait, let me let me get it for you. Oh,

    why? 530 All right,

    I'm ready. Okay, so let's experiment right

    so I'm trying to think what they have in mind.

    party formation that was a super one

    secondly, that was one experiment that was done this COVID Once COVID happened they had to pay with online video. And that kind of worked to their awareness

    to start educating people about privacy and respect especially because today we're talking about spam. So why

    Why was COVID

    Because a lot of people were at home. So then they realize we're spending so much money. Our energy bills

    increase because they have more time. Spent at home exactly.

    Because you

    and now, 14 was the switch to after that, and then you cannot afford it partnership that facilitates interaction to one

    resident themselves, which are

    clean energy facilities, and

    providers who provide this facility to contractors, and three was the actual savings from the government in the state of Massachusetts, that money which trickles down to city facilities where the money is actually coming from. So they became a platform for them and that sort of shifted them to much like a data company. So we were handling the data for the contractors. We're handling the data for the residents, landlords as well because when you were talking about residents part of

    the very interested in the money saving for housing Yeah.

    And the other thing is and contractors already advantage because they get more customers which they normally would not have

    because simply they don't have the capacity right because contractors they get what they take any chances experiment. I would like an external observer. What would this mean? For a policy implication? So if I, so currently, Massachusetts as a cleaner

    and one thing that could be experimented is if we were to increase the workforce of these organizations, by ExxonMobil. How does that impact the clean energy economy? And they have policies because Massachusetts, Massachusetts Clean Energy Control ship, and basically what they do is if you want, if you're a clean energy nonprofit, and you want to hire interns, they will pay the interns on your behalf. So it's a really amazing program. That's how I ended up and through threw back, but I don't think they have any sort of lag. They're just doing it. It's a great help. So I think one interesting experiment would be if you were to increase, or I asked me personally, if I increase my workforce by X amount, how many people do I impact or do a conversion from?

    Yeah. And that people include the intern.

    Yeah, I would say like workforce, in general, includes everyone, regardless of how they're getting paid. Because they are contributing to the economy. So if they're able to hire it, just like you know, in healthcare, you add one more doctor, you add one more nurse who can help but just adding that one

    Yeah, crazy. Yeah. So that's what

    is this an experiment that we can implement? I worked for

    but what were you saying about this?

    Learning methodologies. Yeah.

    There are three approaches and you're learning. Again, and the other is evolution. The other main differences there. Meaning in the graph that I shared with you, yeah. So for Bayesian learning the beliefs and probabilities like some coefficient that you just mentioned, right, how many how much people you impact over a certain amount of money. If that is a distribution, like from $1 you impact like one 0.1 people and that is not very, there is some uncertainty in that information. So if you do this, like $30 experiment and increase the precision of the information by certain amount, that's how you design the inference like the experiment whether this has cost effective or not. This increase of information is worth the money of your experiment because you don't want the situation where your experiment causes less than the information value. So that's usually how the Bayesian learning happens. So action is the experiment. So we were discussing mostly here but evolutionary learning is more like ask your organization organization matures from male to scale to salad face. Male scale sales, so that's a nail and scale and that's the title of the comic book my advisors writing. It's coming out in September.

    Wow. Yeah, it's nearly as like,

    nail in the meal. Yeah. So they need to very Be careful about so early stage startup. Yeah, they cannot afford a lot of different things. So they need to be all around. Yeah. So you need to find a value proposition that you nail that satisfies all the stakeholders. And once you do that, you need to get kind of 10 tools in order to scale up and you need, you need to scale with procedure. So you need to acculturate and processor fie and automate segments and professionalize for specify something like a day or two then choose to do that. So I'm not really sure how I'm so in here evolutionary state is population of strategies. So we can think of it as it's a little tricky, because in here, kind of the agency information Yeah, yeah. Because what we were focusing on was beta, right the coefficient, but I noticed that behavior learning that has a human agency, yes, like behavior, bounded rationality, or they perceive that as entrepreneurs is in the center and they meet, they have the goal and they have kind of human and entrepreneurs just assemble many different opportunities around them. Yeah, meet the goal. They don't really go deep into how they do that. That's just given. Yeah, opportunistic effectuation. It's what we say. So there is this is like biology agency and this agency. Evolutionary is a little kind of mix. Because they are saying the states are population of strategies, and I need to think more about it, but I can think of it as like you have three different parallel ways to develop this idea. And you experiment this and this and this a little bit and then as the information is collected, you kill one of the strategy and focusing on that's usually how the biotech platform companies do. So that's, I think what they mean by population strategies and genetic operations is like what I just said, they like select mutate, and like crossover Yeah, so well, that would be a good example of like, creating variation is like you choose, like, out of three, you killed this one. And then you somehow thought that like what are the different options that you were considering that all energy was considering for its business plan?

    Right, right. I can. Yeah, no, I so like I mentioned it earlier. When was the canvas that second? Try it was the partnerships that ended up working? And I mean, there were a couple in between one was like direct like, you know, landlord assistance, that also was similar to canvassing so instead of candidates, the residents there canvass landlords, but that also was not super successful. And

    even though canvassing failed that can be crossed over to other operations at some point. Yeah. So that's kind of biological. I heard that humans don't evolve in the optimal way. Like, at each step. Like for instance, we I heard that this under our nose this is kind of a mutation Yeah, not not mutation. We needed that long time ago where we had different organs. But now we don't need like, I don't exactly know which part of organs I don't need. The just evolution happens don't happen in individual perspectively optimized. So and that's because the environments and agents somehow we evolved. And here the accuracy, like the business evaluation manuals. So how this is just one table and another table compares how they differ in representation and evaluation and optimization. But I don't think that's super relevant. At this point of our example matching. So I want to make sure that this works. So can you look at this function I just asked me some questions.

    Yeah. Yeah.

    Know the behavior learning certainly works. And the reason why it works is I'll give you an example. This one canvas that we have, she was amazing. Now she works for the city of Boston. But she told me, you know, she would go canvassing, she would literally like she's talking to individuals who are. So she's from Guatemala. She speaks their language goes to households where, you know, they speak Spanish, because they're taking chic interest. But even then, people she said would literally yell at her. They thought she was like a telemarketer. Oh, you know, so like it was and they spent hours and hours trying to perfect pitch that they would get from like the ringing the bells. Right? Clearly, that didn't work because they were casting a very, very wide, too wide, right? Because me maybe energy bill is not my priority. And at that point, you're just like, you know, just like a random telemarketer for them. When they switch their strategy, what they did was they collected data from the city, right? collected data from the city. And then through that they found out okay, these are the folks whose energy really super high and in partnership with energy companies, and now they have the data so you know, that will only go into buyers sorry

    then that will go into your BIOS because now you have to do that right. So from that point all clean, okay. Now I need to contact these people. So now, based on that, you again switch back to your behavior. Now you change your pitch carefully. If it's someone who really needs to reduce their energy bill, you say, Hey, your energy bills high. These are the programs we have in partnership with the government. But if it's someone who's leaking money which is basically like they're using, like CFL bulbs instead of LED bulbs, and that's making their bill really hard. There's ways to find out I don't know the exact technicalities of it. There's experts out there who can determine that for them they can like pitch the Home Energy Assessment options, Conservation Officers, etc, etc. that I still don't think there's evolutionary learning in most of these. Maybe I'll have to look into it a little more. But I can definitely see the Biogen and Gabriel going back and forth and how that's changing. And again, you know, for market conditions, as well. Like you know, when they're casting a widest net

    so what hardware evolutionary comes in is kind of game theory situation. So allergy Oh, old energy, oil and energy has a competitor. Right? Yeah, kind of inverse plan. Yeah, you know, they are much post cost efficient and your then you would avoid their target market.

    Yeah, surprisingly, I don't think they have any competitors yet. At least. Not anyone who's doing the same thing. At first they did when they first were launching because the plan was that you know, they'll eventually like to take over like insulation and all that stuff. So there you could, there was a potential but they ended up just partnering with them. So

    they ended up partnering with a competitor.

    Yeah, I wouldn't call them competitor just but like from from a wider lens like you could consider them as competitors like people. I think the competitors would be in like a very like convoluted direction the energy providers because they are losing money, because the programs we're providing, right, but at the same time, they also want to accelerate clean energy because it's, you know, if a lot of more people are in solar, it reduces the power on the grid. So in that sense that could be a potential competitor in my opinion.

    Got it. So this is what GPP thought was for each state. I don't like the you. Yeah, tell me what you feel about this table, the state efficient learning strategy and the reason for

    Community Engagement behavior learning realize like, Yeah, I agree that energy Oh no, I hate this

    sorry.

    Pick up after your dogs

    I think I got rid of it.

    You

    know, it's it's quite old, but I don't wear it as much. So energy efficiency. This is by Asian uses probably. Yeah, yeah, I agree. Economic Impact is also by Asian Yeah. This one true, but like I said, there's not like a concrete measurement of it, right? Like, yes, you can say okay, this was money reduce. They can use it for something else, but it was there. Did that actually lead to an economic development for the family or no? That's still a question.

    Yeah, like, the reason I'm trying to do is I want to get a feel of when you use different learning strategies. Yeah. Because the YouTube lecture I shared with you was he was a writer, author of the book co Master Algorithm. What he was saying was just five different tribes of machine learning. Yeah, no converge. So he is proposing we need to build a hybrid learning algorithm, okay, which is like for instance, you'll find tribe is, of course, Bayesian evolution. Yeah. Symbolic. So, because if I wrote a paper that includes five of the category, it will be too overloading. Yeah, audio so I sub categorize them into Bayesian which includes symbolic which is a logic doing the deduction. Yeah. companion is doing the Bayesian inference. Yeah. And oh, by the way, each of them has their own master algorithm which is given the enough amount of data, you can figure out all the functions that produce the data. That was kind of the most fidelity situation, okay. And the second one was evolutionary and that is there. Is evolution of who you just start genetic program. Yeah. And the third one is behavioral and that includes both analyte, third, and fourth. Connection. So that's the neural network. Oh, so it kind of follows your approximation because when you think of behavior, yeah, we don't know why they do that, or we just observed they're doing that. Okay. So it's like instead of, like bottom up, meaning I don't know what's going on inside your body. They just react this way system behavior is from kind of the outer approximation of certain phenomena. And that is what kind of new are trying to do. Okay, because there is a input and there is some output. I don't care what this like the function approximation of given input and output. Yeah, so that's kind of I thought, What is behavioral decision making instruct to do? So, the reason I'm telling you this is I want to with example, get a feel when you would use certain algorithms. For instance, one may say, the more uncertain, you have no idea. There's too much things to optimize. And it's very hard to do the Bayesian inference or debate and learning so I agree with you, in your surprise for economic impacts, that we really have no clue, right? Yeah. So I would say maybe it's better to do behaviorally. Yeah, that is kind of the arguments or theorems. I want these cabinets. Okay. Tell me what you think about them. I think General license.

    I think normally speaking, yes. I agree with how it has categorized everything and reason for the categorization as well. Now, if you see this, this is what I'm saying. I couldn't think of like evolutionary impacts, evolutionary learning. What did what was the stage for evolution learn it says environmental impact and the reason why is you know, it optimizes renewable energy projects and test tracks environmental impact to adaptive methods. Is that really evolutionary? Sounds more by Asian to me, right? And not to like, go off shit on evolutionary learning, you know, but maybe there's some value to it that

    there is some thing that social reasoning. Yeah, it's more related to evolutionary because there are multi agents coming in. Yeah. So compared to like economic impact, which is just what is the revenue? Yeah, social impact, you need to think about many other people around. Exactly, exactly. Perhaps that may be why they're saying.

    And then the feedback and adaptation also goes into evolutionary continuously improves program based on feedback and this Yes,

    they have several evolutionary learning. I thought they had only

    one just to feedback on adaptation. But that also is one of those generic

    Yeah, I feel they have their core process very much high correlation between the application and environment. Yeah, so they just classify as environmental learning. I mean, the evolutionary learning. What I'm curious is, if you need if you're in a situation you need to pivot a lot. Yeah. Would you adapt for Bayesian learning or environment, the evolutionary learning?

    I personally think we're both. I do need both. But here's the situation right? Going back to that example of posturing, canvassing, that was, theoretically that was yes behavior, but also evolutionary because we decided, oh, like, this does not work after, you know, trying it out. Because I was I was the one who, like vendored you know, procured the canvassing software and everything. I taught all the people how to use it. So it was a great deal of effort on my part. I also had to like, No, we had to pay for it was achieved, and he said, all these things and now this is where it becomes interesting. If we had known we had that data beforehand, that data being the data about because remember, I said we're casting a wide net, if we have that data about, okay, these are the people you need to target. We probably wouldn't need that sort of like having software. We did need to train our canvassers as much as we didn't need to like tell them to do all these like extensive work, just to get a very small amount of results. But we didn't have that data is important. So we were sort of building this database ourselves until we got access to the data from the city. Right? The city wouldn't give us the data unless they would trust us. Right. So that's like,

    the, like kind of your ask for it. Yeah, I agree. Your effort to build a database even though you didn't end up using it? Yeah. They increased trust level for the government to give you the

    I think it did, because then maybe not because they didn't need the data itself to like know that. We were a trustworthy party. Did just figure it out through the programs that we were doing to the connections and the relationships that were made some dentists register more behavior, right. So it's I know who knows who knows who, that that becomes more important than just a symbol, or like, your database was half as accurate as ours. So we don't mind giving it back because that realistically did not happen. So that's what I'm thinking is. Yes, there is evolutionary learning, which I think is super important.

    Sometimes playing you like both of them, but you need to kill one of them. Because in order to, I guess, pruning, right, exactly, yeah. Even the both hair is very promising. You have to

    choose. Exactly.

    And I think in most cases, they probably chose evolutionary learning, just because of the fact that was easier. Now, I'll give you another example. When we were working with the City of Cambridge,

    Oh one more thing. Sorry. So uncertainty, I think is the most important thing. So here it's like the probability distribution like inherent uncertainty. Yeah. But evolution learning is population diversity. Yeah. And behavior learning is kind of implicit in rules. I'm not sure what that means. So I haven't seen many behavior learning, like explicitly dealing with the answer. Because here's the unit level is kind of biology the human Yeah. So it's kind of the uncertainty is inherent in this human right which is different from this where like, were modeling the information by itself. Yeah. So select finished rent, maximize utility, satisfying. Yeah. And also this is very subjective because it's depends on whether I think I'm satisfied

    or not. Now, here's where it gets. It gets a little interesting. I think the implicit in rules implies what a person's values are. And that becomes important when you're especially when you're working with city governments or state governments or any sort of like agency that has authority. Because it can entirely depend on what sort of relationship you have with with the person, right. I think one of the reasons why we were able to get access to a lot of the other cities because if I if you remember, I told you, the city of Boston was one of our last part. And the only trusted us once we had all the other partners. Yeah, right. Yeah. So someone in the city of Boston is, you know, Mayor's office or someone is thinking, Oh, no, that's just a small start off. We'll see what they do. Because they have their own biases, because they've worked with someone in the past. Maybe it didn't work out. Or you know, everyone has their own inherent biases for things as such. Now, in order to eliminate that bias, we have to prove ourselves, right. And that's where it becomes important is that you need to understand who's the sitting authority that makes decisions?

    Oh, yeah, yeah,

    that's that's how I that's interesting.

    They remember Robert, Robert Robinson, Robert Yeah. Saying that you need to know the people. Right. Exactly. And that's like, whether they satisfied about what you what you're trying to do yes or no, that matters, right. And what

    sort of policy dictated right because if, if, let's say, for instance, for Michelle, right now, if you see, her agenda is making hazard free, which is awesome. And that's high up on our agenda. So if you're someone who's working in that transit space, the chance of you getting funded from city of Boston is super high price, contrary to you know, Marty Walsh's term, he was not big on transit he was more on like eliminating homelessness, reducing the drug problem. So all the organization that worked in depth

    Oh, this is so far. Something is making sense. And the way that she presented as putting the pieces together, like we remember I said representation evolution optimization. Yeah, so it's kind of the mix of the three. And probabilistic logic is from Bayesian learning. And weighted formula is a little from the connectionist. The neural code people which is which I categorize as behavioral thing, right, because it's outer approximation. And evaluation is posterior probability. And user. This is why I'm saying this. This is what you just said, user define objective function. So we cannot model every utility of people out there at some point we need to just ask people, do you like this or not? Kind of Yeah, that's what an optimization is formula discovery through genetic programming. And this is like kind of You brute force every different things. Find, like, yeah, wait learning again through backprop. Yeah. So this is I haven't fully digested this. But yeah, this is what I hope to build. Like if you were if we were to build an AI that helps the operations little social entrepreneurs. Yeah, they need to be very flexible meaning sometimes you need to learn or design the best experiment using Bayesian learning. But if you're dealing with the organization like governments, you should somehow know who to reach out who's in charge and whether they're satisfied or not. And sometimes you need to operate many different strategies in parallel and kill some of them and like select mutate some. Yes. So I need some very concrete example of like, and I think maybe we're getting somewhere. So yeah, let's make a very clean case study with this. Does that make sense?

    No, that makes a lot of sense, actually. Yeah, and I'm thinking, you know, even for back propagation, that's kind of like evolutionary learning. It's something learn I went back to it.

    Yeah, some. There might be some connection. I haven't thought about it that way to be honest.

    Also, where did you park your bike?

    Oh, I live for under Memorial Drive. So

    I didn't bring bikes. Okay, that's cool. I can get him to sit on the way to my home so we don't have to go back. Oh, yeah.

    So could you share me more about behavioral learning like how your understanding is right now? Yeah.

    So one thing I'm thinking of is actual biases heuristics that people have. So for example, their story of our canvassers. The second thing I would say is trust, right, I would say. So, one of the important learnings for us was especially when we're working with a lot of city partners, and also like our like, vendors and contractors. The reason we were able to get them on board was one they knew the director of the organization they knew they had worked with him in the past, so they knew that he could deliver on something. So they have there's that one sort of behavioral bias that they have, okay. If you've worked with him, you might be able to like, get something in return, right. So that maybe is one behavioral

    aspect of it. Are you using bias in a kind of bad connotation just stuff?

    No biasing just like your pre existing information that you have right.

    So when you say behavioral bias in the situation if I work with pregnant I would somehow succeed that you're calling,

    right. So maybe a good thing that the other ones I would say, would anecdotes and testimonials right so one of the big things was, we like the neighborhood, in every situation, word of mouth from your neighbors is like, it's crazy, right? Like the way it spreads. And that's what worked for us. too, was like, Oh, we had this like one man or not forget her. Her like she had like 10 or 15 apartments. She rented it all out at a below average market rate to low income households, right? Very generous individual. So we reached out to her like okay, we can do this for you. ended up working. All the other landlords in that area started adopting. She's an influencer. Yeah, yeah. Influencer bias, right? Because they, the influencer does it. Then you're like, Okay, I worked for her. I'm gonna trust her instead of some random and typically, to be honest, people tend to really not like when things come from the government. That's just like the reality right? Even though you get a brand and a recognition. People are generally scared of government agencies.

    If you by any chance know what the difference between market condition and situation like I

    was thinking about it, what

    what is this?

    Medical? So sometimes when you say situation, it's more like cognition wise or psychology, or as condition is more statistical. You can list different things that for this sake like conditions, so my point is conditions is more information being situation is more like human beings. Okay.

    So what is that?

    Oh, do Where did you park your bike or where did you need to go?

    Oh, okay.

    Yeah, I'll walk you there. Thanks.

    So situation. Okay. In this case, what situation would be Oh, okay. So we also have the solar energy system, right, which is not super successful, but just just as an example, my situation would be everyone in my neighborhood is getting solar. Should I get solar? And I ended up getting solar is that market situation? me

    Are you taking the Tesla example what would you say is the situation

    behavioral thing behavior, market situation? market condition. Now when I hear the market condition, it feels like can be quantified. Like how fast like business cycle lengths for like stock price, their trend, something like that, okay? Whereas, like, would you just describe seems more like a situation because it's kind of situational. Yeah. I'll go there

    for like so, give you an example. Let's switch to induction induction stoves was only possible because you received access to electricity. And that became a market situation. Yeah, yeah. Or like electric cars was only possible because we get the Chargers.

    For some reason, like they're saying decision rules like rule based decisions. So compared to just Bayesian learning, heuristics, sight learning seems to be more like retrieving. Like if this condition than I was just kind of meaning it's last categorized. Like if you express in numbers, and it's continuous, so it's very unique to the procedure, it's much higher, whereas here, they are kind of lazy like for me, like whether I was gonna go to school tour or not kind of a very binary decision. And so that's what I mean by tree like situation, okay. Whereas this is more continuous situation.

    Also, what is that saying?

    There? I'm not sure it's true, but I heard that the call