Loading...


for sharing. Bob. So I think the work here is somewhere between Bayesian and evolution and JB and I had discussion last time about how to understand the adaptation, co adaptation of agent and environment from a agent's perspective. So this is one attempt for it, and the product in this page is mostly based on Bayesian agent. So you can see what I mean by Bayesian is they have some belief about the rate of rate, ROI return per investment, and they are learning depending on what environment they are in. So the easiest way to think about is, if you I think this graph, so I observe that in a digital product area versus industry versus physical industry, there seems to be different ratio of time invested in the marketing versus the operations. And this model tries to explain there is that is like different behaviors observed in different environments is not irrational. It is because that environment has higher return of investment in so in digital it is better to invest more like proportional to the underlying ground truth of that environment. So, yeah, long story short, what this model I built is trying to explain is 

investment as a vector. And by vector, I mean it has both direction. So this is what direction would be about, and also the speed, the magnitude. And what determines the magnitude is how fast, like thinker versus doer, and in most cases, the faster clock speed industry, like it, has more execution because the delayed cost time, expected utility, utility gain per time if you are delayed in the fast moving industry is much larger, so it is logical to act more compared to think more. So that's what I have right now, and what I wanted your input was this part. So since, like, considering JB would be leaving soon, I would start from the second one. But like so far, am I doing a good job? Am I? Am I moving too fast?
So here you see it's for, I think it's for software development, but it applies the same. So we lost you exactly. So here we could have, like, some sequence, and then at some time we could have test with multiple prototypes, and then we can have a sequence again. So in concept development or software development or car development, you can iterate on one thing you do, and you do it better and you improve. Or you can test like AB testing like two products or ABCD, and you put all of them on market, and you see the one that is competing. So it gives you some kind of evolutionary approach where the one that fits the most most of the market might win, compared to making one bet on one thing, and if the bet is not a good one, you lose everything. So it's again, how do you allocate resources in a finite environment? And I did my PhD about evolutionary psychology and taking concept from that, and with Angie. Now we are trying to use some of these concepts and operationalize them in terms of the mathematics, the optimization. So we can apply this model to entrepreneur related to their resource operations that are limited, choice of market, choice of VC, which VC should I choose? And so on and so on. Because engineer entrepreneurs, they have to make choices all the time, and they search for their next future strategic move. This is the search in parallel search or sequential search. What are the parameters for entrepreneur to choose their next step, being this or that, and this is what we try to model together. I hope it's it's understandable. We are still making it, but it's basically we are comparing two very simple things, either you iterate on something or you do many things at once, which is most of the environment of entrepreneur Angie,
yeah, I mind flipped your time. So like you may ask, how it relates with what we were discussing and the second one was parallel, but I put in the third one when we had the software and hardware, or software or hardware that has two different direction and a speed, then you can imagine how the allocation would change if the environment changes, and how the agent can adapt to the new environment. So I put like for the format for all the three questions I wanted to like chat with you today, I put question and my answer to it, so I would be really happy to continue, either through mail or another Zoom meeting, to continue discussing this
no, you're, then you jump way up to some kind of vector thing, which isn't what predators and preys are doing in their life, in their action, they're, they're competing with each other. You're, you're you're marketing or operations are not competing with each other. So it's a different situation.
that was in our last chat we you asked me to come up with three prediction questions. So I had the EV manufacturing, and another column was the ad tech. So that somehow gave me the idea, yeah, oh, when I was making that, the physical industry and digital industry had some interesting differences. And like, I tried to summarize them in this table again. So yeah, so I thought perhaps these are the three prediction questions that we discussed last time about, like, how to do the expected cost reduction. And, yeah. So long story short, while I was making this, I realized that a lot of questions raised in the physical part is a little tilted toward the cost saving, because it's kind of however good market they have. If you don't produce the EV, then you don't you cannot capture the value, whereas software, it's a little straightforward to come up with a product like the developer can just spend one night, not not always. That's straightforward, but comparing to, compared to EV, where you need to actually factory labor and all the machines, it's here. People's attention is more focused on the market side and in the hardware industry, people's attention is more focused on the cost reduce saving side. Does that somehow?
no, no. What I was explaining was with the same model I can describe how agent in like so this was this graph where agent was the same, prior started imagine a twin. Twin was the same. Belief system was dropped into one as a EV and one as a ad tech, and as they do the decisions and interpret the signal, they learn their prior is evolves as they interact with the environment. So later, a person who dropped, who was dropped into the EV now learned that, oh, in this environment, it's more beneficial for me to allocate like seven to three to the operations and marketing, whereas in the ad tech industry, this agent that started from the same prior now learned though, like three to seven might be the optimal ratio. So this is, like, what I call as An evolution or adaptation. Oh, I 
you hope to have a model that here I am, here I am an entrepreneur, and you hope to to and I've got a product, and it's been selling, and and I'm, I've got a new investor, the investor is going to put X million into the company, and possibly, but has asked, you know, what are you know, what are you going to spend this on? Right? Yes. Are you going to spend it on making the product better, or are you just going to keep doing the same, pretty much the same product? Or are you going to spend it on set? You know, it's really sending whatever marketing and sales involved, yeah, so you already have, I would, you know, as the investor, I would be asking, well, how you know, how big a a potential sale is there? You know? How many you know over the next five years? How big, how big a market do you think you have, and what is the competition? And and so forth. But that doesn't seem to be in your any in your model.
like just simple, simple three steps. This is an agent. All inside agents had, and there is an environment module and Agent start with some belief of the this PC is the rate of return, and sigma is the processing speed related variable, meaning, if you if the speed is very fast, sigma is very low, and if the speed the feedback is very like clock speed of the industry, like digital, is much faster, so they would have lower sigma, whereas the hardware would have much higher sigma, because there is a lag. And so that's where the PC from, the belief of rate of return in cost side and the revenue side affects the predictive predicted cost and predicted revenue and your next action to invest, whether in marketing or the operations, is computed in a art max of additional profit, which is a function of minus cost plus revenue. And at each time point you can invest only one of them. So this is just how each step agent updates its belief and makes their decision of investment. So that's that's as simple as that. 
Well, you know, I can see how, I mean, I can't see how 30 years would be of any use at all. I can, I can see how trying to to think about the investment decisions between production and and marketing are important. Are important. Who would be important? And they would depend on beliefs that I, the, you know I, or the the management team that that I'm a part of, purport assign any capital that we have to it, and we would, you know, depend, depending on the nature of the product and the nature of the market, which are really, you know, they're, they're not even if you even if you say the software market or the automobile market, are not just all The same. So depending on a whole lot of of detail, we would be making decisions about, you know, what to what to spend the capital that we have on for the next quarter or the next year, depending on where our decision process is yes and if And and then you think with your system that you either you can, or I, as the entrepreneur, can assign a some kind of a value,
a prior value, to what the proportion of spend, you know, we're going to have to spend it on, you know, given that we already have the product, we're going to The proportion is going to be zero, to, to, to production, and 100% to marketing, or some other proportion, right? And I'm going to have, and that's going to depend on my, on my belief system, which you're going to, which has some sort of, not, you know, some sort of get, you know, my some sort of belief, and I go in as an entrepreneur, and I believe that with this product in this market, my
perhaps one more thing and it can explain is, I didn't touch a lot on this part, but hardware people tend to have longer time. Step is what it can also Explain as well.
than the people who don't keep track of this data and where it can be most helpful this thing is when you need to pivot from hardware industry to software industry, for instance, and knowing that the two industry has different underlying like ground truth of the signal. If you know this, then you can somehow try to rejuvenate your prior, like the prior that you build on so far. You just try to abandon this prior and then start a new learning, a new whereas, if you don't know that, you would just keep on kind of having a very sticky learning based on the environment that is not relevant to where you're currently at,
is a great comment. Thanks for that. Yeah, I will prepare next time about how changing the environment can be modeled as a hyper priors. So, 



was a project for the European Union. By by the way, I think all of this is on on my website, under the poll, about there this and about four or five other downloads, PDFs, so you get the whole, the whole thing. But here, here's what we here's what the modeling process was. We described three different alternative scenarios for the European Union which does not have very much energy or material supplies. It's a very small The European Union is a very small area, and it just doesn't they import practically everything, including a lot of food so and they wanted to figure out they paid this a whole team of us. There were seven or eight or 10 of us on the team, including some very good simulation modelers. And we created three different scenarios. Here they are, global cooperation, Europe goes ahead, and civil society leads again. This is, you know, I'm sorry, it's I should have sent this ahead if it was relevant. And all of those model ideas were put into a stimulation model, a very big one, and the simulations were then run again, also against the reference scenario, a sort of keep
doing what you're doing now, basically, and then the output for the model was this, I'll give you just a chance of I 


well, first I'll show you what one scenario looks like. Here's um, Europe goes ahead. That's, that's what the scenario looks like. It has to do with resource governance, power mobility, buildings, industry, land and global developments. And the time frame is over 10, the 2010s the 2020s the 2030s and 40s, and then the outcome, the numerical outcome of the modeling process, against what the targets were for the project. So project. Now there are two other models, like two other scenarios like this. Oh, I should probably show you what the vision for the whole project is. And just a 





We started with the goal and, and, and then worked on three different sets of assumptions, three different sets of assumptions, for Europe goes ahead, civil society leads and so forth. So here, here's the here's the overall project. Start at the top. Europe wants to improve its quality of life. That's why they're paying us, right, all right. And you know, in all of these areas, this is nothing other than we'd like to have we like to have something better in our homes. We like to have better solidarity between countries, blah, blah, blah, then, however, in the real world, let me I'm zooming in a little bit here so you can read it. I in the real world, we're actually dangerously exceeding planetary boundaries and approaching a whole bunch of others. Well, they took that seriously. So what does Europe do when you're facing that well, and what do you have? What are you doing when there's all of this resource depletion and global independence? Well, this requires enabling the economy to grow while total resource use decreases. That's the problem for Europe and for the rest of the world also, by the way, but, but we were just working for Europe, and so Europe has to have new targets and those that was what was on the far right hand side of those scenarios. The change these were these resource efficiency targets. Were the goals. They're the goals. 
Well, we had to think through, you know, they said, you know, the project was, what do we do? So then our team had to think through, you know, what, what we had to do all the way, you know, the way through. And I think some, maybe something like this could be useful to you. Then, yeah, then we created those scenarios. And now I'm gonna really overload you, but I will send this.
here, these are the numerical outcomes of the simulation. And these are, again, interesting. 

let's see this over. Wish we had bigger screens, bigger bigger screens so you can see that again, you've got resource governance, power, mobility, buildings, industry, and then the outcomes that were found from the simulation for for the business as usual, one and the reference one, and then for global cooperation, the same thing, the same targets, Europe goes ahead and civil society, those are the main scenarios.