18 ms·

Along similar lines, nobody "really knows" how Google search works. Sure, you have some experts who can talk about how parts of the algorithm work, you have the

by qqtt 5y ago

Along similar lines, nobody "really knows" how Google search works. Sure, you have some experts who can talk about how parts of the algorithm work, you have the underlying bones of page rank, you even have SEO experts who can get you reasonably close to the top search results. But is there a single individual on earth who knows all the ins and outs of all the minutiae of every single aspect of Google search - from search result algorithm to ingestion to crawling to artificial intelligence? Not a chance. You don't even have to get as complicated as Google search - any reasonably complex software system with hundreds of engineers working on it will be beyond the comprehension of any human being - fully understanding that is (which also entails knowing exactly all the edge cases, all the bugs, all the unexpected results possible, etc.) How does the economy work? Nobody knows - but that doesn't mean nothing intelligent can be said about it, or that the Fed can't make reasonable decisions on imperfect information - just like any other incredibly complex system that humans struggle to understand.

jawarner 5y ago
I don’t know. Google search is an engineered system, but the economy is an emergent system.
_xnmw 5y ago
This is a really bad analogy. We can understand Google Search far, far better than the economy. Google itself is just one small part of the economy. We can test hypotheses, run concrete tests in seconds, and we have SEO experts which can reasonably reliably get you to the top in a few months. There is no equivalent for the economy. We can't run macro-scale real-world tests. Hypotheses are basically unverifiable.
omegalulw 5y ago
Software (and economy) get big and complex due to emergent properties that arise from base rules. This is why, for most large projects, no one _needs_ to know the entirety and all the edge case to make modifications and improvements, you focus on base rules and build off from there. For software that would be individual components and whatever invariants you have set up. And then you you build off of these to reason at higher levels. Attempting to know the whole thing at once "and all the edge cases" is quite wasteful.
arcticbull 5y ago
Right, or the way that gravity works. We know that it works and that's good enough for the day to day. Even good enough to plan.
phkahler 5y ago
That's missing the point. Economics is treated more like thermodynamics. In thermo you have lumped parameter models (parameters like temperature and enthalpy) that are actually correct even though they dont take tiny details into account. IMHO the problem with economics is that their simple models are not over a closed system. A supply and demand curve is fine for a specific product, but it fails if you try to use it at the macro level where other factors come into play. I'd love to work on better models for the fed, with the data and experts they have.
- tibbetts 5y ago
  Also people working on thermodynamics get to test their models a lot, and it’s noticed when their models fail. I’m not even sure supply demand curve is good for a specific product, having spent some time around consumer packaged goods analytics. Consumer behavior around frozen pizza alone is massively complicated, never mind cleaning supplies, candy, beer, or cosmetics.
  - scott00 5y ago
    I'd love to hear more about this, sounds fascinating.
    - spaetzleesser 5y ago
      Thermodynamics is always the same. People’s tastes change all the time constantly. The same person may do something different tomorrow under the same circumstances as of today.
    - kristopolous 5y ago
      For instance, lowering the price of something doesn't necessarily increase sales - it can decrease sales if say the consumer believes it's cheaper because it's an inferior product. So sometimes increasing the price is the way to go to increase sales. So the price operates in weird ways; what about other things. If the product is placed next to more expensive items it could do a number of things. For instance, it could be a decoy product and increase the sales of the adjacent item. Or perhaps the adjacent item could be a decoy product. And what about numbers? Pricing things ending in 9s perform better. Selling things at $39 tend to do better then lower numbers like $34 or $37 because of the 9. And then there's the packaging and placement itself. Tall or wide packaging of the same volume emote different signals and respond differently to prices - look at the great variety of how bottled water is packaged next time you're at the store. Then there's the quantity on a shelf at any given time. Sometimes intentionally overstocking the shelves will increase sales as consumers will take more than they need, sometimes understocking will creating anxiety that the product won't be available when needed. Depends on the product, the buyer, the conditions of the market, lots of things. Then there's the brand itself. Depending on the market you're in, a name brand can either help or hurt you. In some markets, the name brand is associated with premium pricing and poor value so it doesn't carry your product; in other markets it's just the opposite. And then there's discounts and sales. In some stores you want to never have a sale, ever. In other stores you want to barrage your shoppers with a complicated array of perpetually churning sales. And what about sales, there's an artificial price that gets advertised before the actual one; nobody is paying the higher price, so it can more or less be chosen arbitrarily. You can mark your bubblegum as 99.99% off from being $10,000, but then the illusion is broken. Instead, what sales do is they give the perception of a higher price signal through the presentation of the sale without actually charging the price. That's why when you go into clothing stores they'll say the product was $95 or whatever and what luck, through many discounts it's now only $30. Well they all sold at about $30, we know that, but the product decoy pricing itself tries to capture both worlds. This is just honestly just a tiny bit of how demand actually works. It's a vast study. The point is that people aren't mechanical rational automotons without any strategy or agency who respond to prices through unconscious reflexes. Nor do they have perfect information or maximize their utility value through arbitrarily complex deductions and always choose the most optimum outcome. Things are totally messy and the classical models are about as accurate as saying "winter brings cold weather and summer brings hot". Sure, in general. But on any specific day it's non predictive. A $10,000 pack of gum will likely perform poorly. But if you hyped the heck out of it and made it really unique, you might even sell out your inventory of that alarmingly quickly because humans are weird like that. Proof? Let me introduce you to Palessi https://m.youtube.com/watch?v=-2gaNq-rOkI https://m.youtube.com/watch?v=-2gaNq-rOkI
MuffinFlavored 5y ago
> But is there a single individual on earth who knows all the ins and outs of all the minutiae of every single aspect of Google search Would you go as far as to wager that the top 15 engineers/architects on the project (over the years) know about 85% of the ins and outs of the system? Generally speaking at least, edge cases excluded.
- indigochill 5y ago
  I work on a video game that's approaching 20 years old. I don't think in our case that this distribution would hold, because for instance if you took our top engine programmer, he probably wouldn't know off-hand how to work with our texture pipeline. Nor our web store. Nor our customer support tools. He'd probably be capable of figuring any of them out, but in terms of "knowledge at rest", I doubt anyone knows even a third of everything there is to know, even if we constrain ourselves only to knowledge of things within the engineering discipline.
  - sprafa 5y ago
    World of Warcraft ?
  - JCharante 5y ago
    Eve Online? If so the knowledge at rest is probably much less than 30%
- jeremysalwen 5y ago
  As a SWE working on Google Search, I can tell you this is not the case for Google Search (Nobody understands anywhere close to 85% of it). The amount of different moving pieces is enormous, and most parts are constantly being changed. The best you can do is to have an understanding of which system is responsible for what, and delegating to someone who is an expert in that system when you need to dig into what it is doing.
  - TrumpRapedWomen 5y ago
    Who thought it was a good idea to have Google just randomly drop words I searched for so I must put double quotes around words so that Google actually searched for them? This is an incomprehensibly stupid thing for a search engine to do.
albertgoeswoof 5y ago
No one even knows how to make a pencil: https://en.m.wikipedia.org/wiki/I,_Pencil https://en.m.wikipedia.org/wiki/I,_Pencil
anonu 5y ago
Are you saying the core Google search engineers don't know how the product they are building actually works?
- Dma54rhs 5y ago
  Why would it be surprising when the product is complex snd huge?
  - anonu 5y ago
    I don't believe that a successful product operates in some magical realm where the operators just don't know how it works. I'm sure it's complex but it's probably very well laid out and managed complexity. An opaque system cannot be changed or improved upon. But yet, we see that Google Search changes every now and then with different methods of indexing and new features that bring relevant content to the top. Its evolving, which means that there are people who understand the system enough to effect change.
    - denton-scratch 5y ago
      I have never worked in a place where anyone knew everything about how the product worked. Google Search has been getting worse rapidly. I get more and more irrelevant ads crowding out the organic results, and more and more results that simply ignore my search terms. Nowadays I only use Google Search if I haven't found what I'm looking for by result-page-3 on DDG; and that usually doesn't help. I think the people at Goo that want search to be "the bestest" possible, are now in the passenger seat; the driver is obviously advertising. Wouldn't it be nice if "different methods of indexing and new features" really brought relevant content to the top? Instead, they seem to make relevant content disappear completely. Google could easily have the best search engine in the world, again, if they cared enough. But I guess money is more important; revenue from pure search: $0. Revenue from advertising: I dunno, $1billion PA?
- spaetzleesser 5y ago
  Is there anybody who knows all parts of Windows 10? I doubt it.
joelbluminator 5y ago
The difference is your average pro can anticipate what Google search will look like tomorrow and make pretty accurate predictions. With the economy the average pro can barely even explain stuff after they happened, let alone anticpiate what's coming next month. Also it feels like the whole set of axioms thought of in the 19-20th centuries (stuff like interest rate / the desired amount of debt) is being questioned. It may be that the whole foundation of economics is shaky.

by qqtt 5y ago

jawarner 5y ago

I don’t know. Google search is an engineered system, but the economy is an emergent system.

_xnmw 5y ago

This is a really bad analogy. We can understand Google Search far, far better than the economy. Google itself is just one small part of the economy. We can test hypotheses, run concrete tests in seconds, and we have SEO experts which can reasonably reliably get you to the top in a few months. There is no equivalent for the economy. We can't run macro-scale real-world tests. Hypotheses are basically unverifiable.

omegalulw 5y ago

Software (and economy) get big and complex due to emergent properties that arise from base rules. This is why, for most large projects, no one _needs_ to know the entirety and all the edge case to make modifications and improvements, you focus on base rules and build off from there. For software that would be individual components and whatever invariants you have set up. And then you you build off of these to reason at higher levels. Attempting to know the whole thing at once "and all the edge cases" is quite wasteful.

arcticbull 5y ago

Right, or the way that gravity works. We know that it works and that's good enough for the day to day. Even good enough to plan.

phkahler 5y ago

That's missing the point. Economics is treated more like thermodynamics. In thermo you have lumped parameter models (parameters like temperature and enthalpy) that are actually correct even though they dont take tiny details into account. IMHO the problem with economics is that their simple models are not over a closed system. A supply and demand curve is fine for a specific product, but it fails if you try to use it at the macro level where other factors come into play. I'd love to work on better models for the fed, with the data and experts they have.

tibbetts 5y ago
Also people working on thermodynamics get to test their models a lot, and it’s noticed when their models fail. I’m not even sure supply demand curve is good for a specific product, having spent some time around consumer packaged goods analytics. Consumer behavior around frozen pizza alone is massively complicated, never mind cleaning supplies, candy, beer, or cosmetics.
- scott00 5y ago
  I'd love to hear more about this, sounds fascinating.
  - spaetzleesser 5y ago
    Thermodynamics is always the same. People’s tastes change all the time constantly. The same person may do something different tomorrow under the same circumstances as of today.
  - kristopolous 5y ago
    For instance, lowering the price of something doesn't necessarily increase sales - it can decrease sales if say the consumer believes it's cheaper because it's an inferior product. So sometimes increasing the price is the way to go to increase sales. So the price operates in weird ways; what about other things. If the product is placed next to more expensive items it could do a number of things. For instance, it could be a decoy product and increase the sales of the adjacent item. Or perhaps the adjacent item could be a decoy product. And what about numbers? Pricing things ending in 9s perform better. Selling things at $39 tend to do better then lower numbers like $34 or $37 because of the 9. And then there's the packaging and placement itself. Tall or wide packaging of the same volume emote different signals and respond differently to prices - look at the great variety of how bottled water is packaged next time you're at the store. Then there's the quantity on a shelf at any given time. Sometimes intentionally overstocking the shelves will increase sales as consumers will take more than they need, sometimes understocking will creating anxiety that the product won't be available when needed. Depends on the product, the buyer, the conditions of the market, lots of things. Then there's the brand itself. Depending on the market you're in, a name brand can either help or hurt you. In some markets, the name brand is associated with premium pricing and poor value so it doesn't carry your product; in other markets it's just the opposite. And then there's discounts and sales. In some stores you want to never have a sale, ever. In other stores you want to barrage your shoppers with a complicated array of perpetually churning sales. And what about sales, there's an artificial price that gets advertised before the actual one; nobody is paying the higher price, so it can more or less be chosen arbitrarily. You can mark your bubblegum as 99.99% off from being $10,000, but then the illusion is broken. Instead, what sales do is they give the perception of a higher price signal through the presentation of the sale without actually charging the price. That's why when you go into clothing stores they'll say the product was $95 or whatever and what luck, through many discounts it's now only $30. Well they all sold at about $30, we know that, but the product decoy pricing itself tries to capture both worlds. This is just honestly just a tiny bit of how demand actually works. It's a vast study. The point is that people aren't mechanical rational automotons without any strategy or agency who respond to prices through unconscious reflexes. Nor do they have perfect information or maximize their utility value through arbitrarily complex deductions and always choose the most optimum outcome. Things are totally messy and the classical models are about as accurate as saying "winter brings cold weather and summer brings hot". Sure, in general. But on any specific day it's non predictive. A $10,000 pack of gum will likely perform poorly. But if you hyped the heck out of it and made it really unique, you might even sell out your inventory of that alarmingly quickly because humans are weird like that. Proof? Let me introduce you to Palessi https://m.youtube.com/watch?v=-2gaNq-rOkI https://m.youtube.com/watch?v=-2gaNq-rOkI

MuffinFlavored 5y ago

> But is there a single individual on earth who knows all the ins and outs of all the minutiae of every single aspect of Google search Would you go as far as to wager that the top 15 engineers/architects on the project (over the years) know about 85% of the ins and outs of the system? Generally speaking at least, edge cases excluded.

indigochill 5y ago
I work on a video game that's approaching 20 years old. I don't think in our case that this distribution would hold, because for instance if you took our top engine programmer, he probably wouldn't know off-hand how to work with our texture pipeline. Nor our web store. Nor our customer support tools. He'd probably be capable of figuring any of them out, but in terms of "knowledge at rest", I doubt anyone knows even a third of everything there is to know, even if we constrain ourselves only to knowledge of things within the engineering discipline.
- sprafa 5y ago
  World of Warcraft ?
- JCharante 5y ago
  Eve Online? If so the knowledge at rest is probably much less than 30%
jeremysalwen 5y ago
As a SWE working on Google Search, I can tell you this is not the case for Google Search (Nobody understands anywhere close to 85% of it). The amount of different moving pieces is enormous, and most parts are constantly being changed. The best you can do is to have an understanding of which system is responsible for what, and delegating to someone who is an expert in that system when you need to dig into what it is doing.
- TrumpRapedWomen 5y ago
  Who thought it was a good idea to have Google just randomly drop words I searched for so I must put double quotes around words so that Google actually searched for them? This is an incomprehensibly stupid thing for a search engine to do.

albertgoeswoof 5y ago

No one even knows how to make a pencil: https://en.m.wikipedia.org/wiki/I,_Pencil https://en.m.wikipedia.org/wiki/I,_Pencil

anonu 5y ago

Are you saying the core Google search engineers don't know how the product they are building actually works?

Dma54rhs 5y ago
Why would it be surprising when the product is complex snd huge?
- anonu 5y ago
  I don't believe that a successful product operates in some magical realm where the operators just don't know how it works. I'm sure it's complex but it's probably very well laid out and managed complexity. An opaque system cannot be changed or improved upon. But yet, we see that Google Search changes every now and then with different methods of indexing and new features that bring relevant content to the top. Its evolving, which means that there are people who understand the system enough to effect change.
  - denton-scratch 5y ago
    I have never worked in a place where anyone knew everything about how the product worked. Google Search has been getting worse rapidly. I get more and more irrelevant ads crowding out the organic results, and more and more results that simply ignore my search terms. Nowadays I only use Google Search if I haven't found what I'm looking for by result-page-3 on DDG; and that usually doesn't help. I think the people at Goo that want search to be "the bestest" possible, are now in the passenger seat; the driver is obviously advertising. Wouldn't it be nice if "different methods of indexing and new features" really brought relevant content to the top? Instead, they seem to make relevant content disappear completely. Google could easily have the best search engine in the world, again, if they cared enough. But I guess money is more important; revenue from pure search: $0. Revenue from advertising: I dunno, $1billion PA?
spaetzleesser 5y ago
Is there anybody who knows all parts of Windows 10? I doubt it.

joelbluminator 5y ago

The difference is your average pro can anticipate what Google search will look like tomorrow and make pretty accurate predictions. With the economy the average pro can barely even explain stuff after they happened, let alone anticpiate what's coming next month. Also it feels like the whole set of axioms thought of in the 19-20th centuries (stuff like interest rate / the desired amount of debt) is being questioned. It may be that the whole foundation of economics is shaky.