Physics - The Second Law of Thermodynamics
A week ago I stretched my rudimentary knowledge of high school biology to come up with an analogy that, for me, shed some light on web services. Today, I am going to try an analogy from physics which I know even less about.
Most industrial processes produce byproducts. More and more, we see the output of these processes (even the waste) as a potential input for another process. An example of this is co-generation – generating energy from the waste heat of existing industrial processes. Like co-generation in an industrial process, many web services are driven by the data that is created as a byproduct of a user’s interaction with another service.
When you use a search engine to find mortgage deals, you reveal your intention as a byproduct of your search. That intention data becomes the input to a different system – the search engines advertising system. The intention data is a byproduct of the search, not the purpose of the search; it is data exhaust. When Amazon recommends a book, they are using the exhaust from your prior purchases and those of people who have purchased similar books to show you books you may like. People who tag links in Del.icio.us do it so they can find those sites again. As a byproduct, Del.icio.us has created a curated list of interesting things on the web that has become a powerful discovery tool for lots of people who have never tagged anything.
The analogy, however, seems to break down when you compare the potential efficiency of “co-generation” in physics and in web services. In physics, the potential of a co-generation system is limited by the Second Law of Thermodynamics which says (I think) that the available energy in a closed system can never increase. This means that no matter how efficient your co-generation system, you always lose the use of some of the energy in the system. Otherwise we would have perpetual motion machines. It appears, however, that the exhaust of a web service can be more useful than the service itself. Webmasters linked to other sites to make their sites more useful to their users. That took a certain amount of energy and provided a certain utility to their users. They did not intend to power a search engine. But then Google captured this exhaust and used it to organize the web.
So the extra point question is… does the apparent increasing utility of data violate the Second Law of Thermodynamics? Is data different, or are we fooled by the inherent efficiency of data storage, transmission, and manipulation so that we don’t even notice the incremental energy lost? Perhaps I have not accounted for all of the energy in the system (i.e. the users interaction with the data adds energy). This is more than an academic question. If data is somehow different, even if only the efficiency of its reuse is greater, then it will impact the way web services companies package, deliver and get compensated for the use of their data assets.
January 25, 2006 01:46 PM, By Brad Burnham
Tags: data physics thermodynamics
Comments (9)
It occurs to me whilst reading your post is that the question of whether the exhaust is more useful than the service itself is an interesting one.
I think the physics analogy is helpful here. Just as energy can never be created, it is always transferred, so is true with value. Useful exhaust is only the by-product of a separate, more useful service.
Amazon's recommendation feature is useful but isn't useful enough that people will buy books simply in order to create the exhaust to power the recommendation system - that really would be perpetual motion.
Like energy, value is a conserved quantity. Any company must be very careful to ensure that the value it extracts is less than the value it adds. If the balance becomes the wrong way around, the company goes bust. Exhaust may sometimes be where the money is (the advertising industry) but it is only ever the by-product of value-creation elsewhere (the publishing industry).
Bad companies see money coming from exhaust and try to generate exhaust. Good companies create value and then figure out how to make money from what follows.
Posted by Peter Nixey , January 26, 2006 09:31 AM
Brad - as someone who majored in Physics in college, I can assure you that you are not alone in your struggle to wrap your brain around the 2nd Law of Thermodynamics. Even practicing physicists have a tough time explaining exactly what the 2nd law means (as opposed to simply showing you the formula).
That being said, like reality, your analogy suffers under one of the most inconvient facts about real-world systems - they are never closed. So, while information, value, heat, intelligence and stupidty may be approximately conserved, the lack of a fully bounded system guarantees that there will always be some flow in and out of the system.
I would also like to comment on something Peter said regarding value being conserved. Perhaps I am lacking the proper context for this comment, but I do not believe this to be true. Metcalfe's Law and, if you believe it, Reed's Law are direct consequences of this not being true. Bascially, these laws state that the value of a network increases as you add more people. I have postulated (although not proven) a corollary for data which is that the value of a data set increases with the diversity of the data. Time and again we have seen where a small amount of input leads to disproportionately larger output (one could claim that Google is a perfect example of this).
Posted by Joe Wilson , January 26, 2006 09:57 AM
Very interesting point Joe. You didn't say it but I presume what you were meaning to emphasise was that Metcalfe's laws state that the power of the network rises according to the *square* of the number of connections.
If the value of the network rises proportionally to the *square* of the connections then we are getting more value from each subsequent connection than we had from our original. Simply rising proportionally to the number of connections does not disprove the theory that value is a conserved quality since each person represents value and so adds to the total value of the system.
So Metcalf does indeed present a problem but only *if* you assume that the full value of each person in the network is being utilised. If it's not then the addition of each new person adds value in itself but also adds value by unlocking some of the latent value in the original members.
If I have a mobile phone and my mum has a mobile phone then I use the potential of it nowhere near as much as if all my friends have one too. With only one other person in my network I was using only a fraction of the potential of the phone. I can add new people to my phone book though and only tapping into the unused value of my phone – I’m not generating new value.
So, for a small network, we can still stay within the bounds of value-conservation. What happens when it gets bigger though? If Metcalfe's law is correct and value really does rise proportionally to the square of the number of members in the network then we are soon getting into very high-value territory.
Metcalfe’s law assumes that every person makes use of every other person in the network but of course that’s not true. I don’t phone every person with a mobile phone, I only phone those I know.
What actually makes more sense is that the value of a network rises proportionally not to N^2 but to N*log(N) (http://en.wikipedia.org/wiki/Metcalfe's_law). When this is the case I get a very rapid increase in the value of the network as new people/members join it but ultimately the value is proportional to the number of people.
If value rises proportionally to the number of people then value is still being conserved - it's not coming out of nowhere, it's coming out of new people and better use of the existing system. It also means after a certain point, increasing the size of the network is of no real benefit to me as an individual since the value per-person is almost constant.
This should come as no surprise. When I first got a mobile phone it was very useful. It became a disproportionately more useful as my friends got mobile phones. Even though it was only after that happened that most of the UK and the states also got mobiles, the value of my phone to me did not rise by Metcalfe's predicted 400M-fold. In fact it stayed pretty much the same, which if Odlyzko and Tilly are correct, is exactly what we'd expect.
So I stand by my original statement which is that value is conserved. Adding more members to a network may make it more efficient but it can't spontaneously generate value past that contained within the people or nodes in it.
(personally I don't think that N*log(N) is the correct proportionality but my point remains the same)
Posted by Peter Nixey , January 26, 2006 11:11 AM
Brad, I'm not into physics, but I think I understand the thrust of your thinking.
When the exhaust becomes sufficiently interesting, people start to game the input, at which point the value of the exhaust diminishes. Look what happens when too many del.icio.us users start tagging based on what they think others will understand, rather than what makes sense to them. The value of their tags as a reflection of their own interests, drops.
Or when people game Digg. Or game Google to manipulate results ranking.
Another point that seems related (though I'm not sure how) is that search engines are butting up against some kind of self-limiting mechanism. The growth in indexed content is expanding faster than is the growth in search precision/recall. Thus, ironically, we're likely to keep getting even more junk with future search queries. (That's why Google's Base and similar metadata-collecting initiatives are so critical.)
Posted by Terry Steichen , January 26, 2006 02:54 PM
Interesting analogy and a fun read.
Information is fundamentally different from physical materials. If you have an apple and share it with me, it's a zero sum gain - we both have a fraction of an apple that adds (at best) to one whole apple. If you have knowledge/information and share it with me, the effect is additive (and sometimes multiplicative) the sum of our knowledge is at least twice what you originally had.
This has bearing to some of the paradoxes that emerge from set theory. If one were to break down the universe into 'sets', you might have the 'set of all chairs' which includes all the chairs in the universe, and the 'set of all apples' which includes all the apples in the universe, and so on. The question is, does the 'set of all sets' include itself? If it does, it becomes an infinitely large set. (I might be partially munging this paradox)
How can this be?
It's because unlike the 'chairs' and 'apples' in the first two sets, which are physical entities, the 'sets' contained in the third set I cited are logical entities - they exist as information only. Information doesn't have to obey the laws of physics because it isn't physical - it exists only in our minds. Information can represent real things (ie. the set of all chairs refers to all the chairs that 'physically' exist) but that doesn't mean that the information itself is real - it's just a logical construct.
Posted by Yali Friedman , January 28, 2006 12:13 AM
Without having read every comment completely (my system is about to crash and I'd like to get this thought out quickly) the analogy breaks down for me because what you are labeling as exhaust that can be used as energy is actually another byproduct that can successfully be captured and put to good use. Intentional data is incredibly rich and interesting although most companies don't have good access or the ability to analyse it.
Forget physics and think about beer making. Fermentation basically takes yeast and sugar and creates alcohol and C02. If you ferment beer in the bottle that C02 is captured and retained in the beer. In the modern manufacture of beer, and with most modern drinkers preferring not to have that yeast sediment at the bottom of the bottle, beer is fermented in huge quantities and filtered. But the C02 is still captured, stored, and injected into the beer before bottling. Intentional data is more like that C02. If captured and used correctly it can really add value.
Posted by Ed Hodder , January 30, 2006 09:35 AM
ok, hopefully the computer is a little more stable now.
So, in the earlier days of beer making the fermentation would take place in a cask or bottle and the CO2 would get thrown off by the yeast, build up pressure and get infused into the beer.
Now, think about the small local retailer--say a wine shop. When a customer walks in the door and heads to the cabernet section, the merchant has a pretty clear idea about the customer's intentions. That intentional information was infused into the process of brick and morter retailing and of course could be used from one visit to the next.
Unless you take measures to capture and use the intentional data that is available in the web world then it is lost. Miller could let all that C02 escape into the air but then they would either have to put out flat beer or buy the C02 from someone else.
Intentional data is the bubbles in your beer or the sparkle in your champagne. It makes online marketing come alive and is a natural by-product of the online interactive experience.
Posted by Ed Hodder , January 30, 2006 10:02 AM
Some friends told me about this site, and now I'm glad they told me about it. when stake increase TV percieve: http://www.boxofficemojo.com/movies?id=poohsheffalumpmovie.htm universal grass becomes small gnome in final, full table becomes tremendous cosmos in final
Posted by gnome will plane unconditionally , May 11, 2006 09:43 AM
Brad,
I think that you've drawn an interesting analogy but I suspect that your labelling of which elements represent exhaust and which represent energy is perhaps awry.
Both Google and Del.icio.us are data enrichment engines. They suck in low-quality data with a low signal to noise ratio and use algorithms to reduce the volume of noise and increase the strength of the signal.
However, the front page of del.icio.us and the results page of Google are not the exhaust of this process any more than petrol is the exhaust of crude oil refinement.
To continue the analogy, let's think of the web as crude oil. There's good stuff in there but if you put average an average slug of in your brain it would probably seize up. The top results on Google and the most popular links on del.icio.us are the refined high octane fuel that is what our very refined minds power off.
In a pure physics sense, the exhaust of Google and Del.icio.us is the heat that their computers and brains generate. In a data sense though the exhaust is the millions of webpages and links that don't make it onto the front page of Google and the urls that are only ever tagged by one person in del.icio.us.
I like the idea of applying the second law to information but I think that both these companies are safe and sound within the physics compound. They start with a large amount of data, decrease the volume of it, increase the quality and use energy to do so.
Lots of low quality data in
= small amount of high quality data out (first 2 pages of Google)
+ large amount of low quality data (pages 3 thro 'ooooooogle')
+ heat
The final result is more enriched and has a better signal to noise ratio but for each 'good' result there will be a million more bad ones. I suspect that both intelligence and stupidity are conserved properties :)
Posted by Peter Nixey , January 26, 2006 07:25 AM