Does Human Knowledge Double Every 5 Years?

Global Warming and Hypothesis 3

If you've read the beginning days of this weblog (which today celebrates two weeks of actual existence), you might recall that I am trying to test three hypotheses. The third is that we are disproving (or falsifying) incorrect assumptions, hypotheses and opinions at a rate that effectively doubles human knowledge (knowing something is not true is really, really important). This is just as important in global warming as with anything else.

Critics of global warming have for years pointed to the fact that satellite observations, taken from a single dataset, showed slight cooling in the tropical troposphere since 1979. Since this conflicted with what climate models predict, this was a serious problem for those attempting to warn the public about global warming.

So they went and got more data, which is what you're supposed to do. This 2005 news release from the Lawrence Livermore Lab (close to where I grew up), shows that further research found better data that agrees with modern climate models.

As there is still considerable controversy about global warming, there's a whole lot more falsifying and disproving that needs to be done, probably on both sides of the fence. If you think about it, the  beneficiaries of this disproof of the previous data are the critics of global warming as well as those who strongly believe, as critics now actually have to examine their assumptions, look for more or better data, or find explanations for what this research showed.

And Still More About Human Knowledge and Global Warming

If you go to climateprediction.net you can volunteer your computer's spare time to help run climate models. I found the site thanks to an article at NewScientist.com. The project, started in 2002, appears to have run the Hadley model 172,229 times (at the time I checked), apparently with minor changes in assumptions to check on the different outputs that result. Check here for the aggregated results of the first 27,000 runs. (I think--I didn't spend a whole lot of time on the site.)

This is exactly what I would expect to find if human knowledge is advancing rapidly and incorporating multi-disciplinary tools and knowledge to forge ahead. Hope I find a lot more like this.

Yet More on Human Knowledge and Global Warming

Okay, stick with me here. The IPCC summary for policy makers (PDF) of their draft report on Global Warming was issued in February--the draft will be released in May. It used data from over a dozen climate models in preparing their reports. The 2001 report relied heavily on a model developed by the UK's very own Hadley Centre. The policy summary notes the following problems with the climate models they used:

• systematic biases in simulation of the Southern Ocean, which is important for the transfer of heat between the atmosphere and the oceans
• on-going problems in simulating the El Nino - Southern Oscillation (ENSO) cycle, which is a major factor in the Earth's climates
• poor simulation of precipitation events: "in general, models tend to produce too many days of weak precipitation (<10 mm per day) and too little precipitation overall  in intense events (>10 mm per day)"
• substantial uncertainty in the simulation of feedback from sea-ice, which are coupled with polar cloud formation and transpor of heat through the polar oceans

Given the potential gravity of worst-case scenarios of the effects of global warming, you cannot wait for perfect data before attempting to make policy decisions. But these are not trivial issues.

This should be a hotbed of scientific activity, and we should see lots of research being done on the issues above. In particular, we should see attempts to harness Moore's Law, which baldly states that computer power has doubled every 18 months to 2 years in the recent past to make climate models more detailed, easier to test and quicker to run. So what do we find?

Here are some positive indications:

• (I thought this was much more exciting until I looked at the publication date of 2002) (From Science Daily) "Atmospheric scientists from Lawrence Livermore National Laboratory have performed the first global climate simulations with spatial resolutions of roughly 50 km (30 miles). This capability will be used to assess climate change and its societal impacts. Typical global climate simulations use spatial resolutions of about 300 kilometers (186 miles), which limits their ability to simulate climate and climate change on a regional scale "
• This (also from Science Daily, but from 2007) is the type of article I was hoping to find: Open Source Software Toolkit Plays Key Role In New Climate Simulations  Science Daily — The Model Coupling Toolkit created by the U.S. Department of Energy's Argonne National Laboratory played a key role in the climate simulations used in preparing the new U.N. report "Climate Change 2007: The Physical Science Basis." The Model Coupling Toolkit (MCT) is an open source software library for constructing parallel coupled models from individual parallel models. MCT is designed for high performance and portability. All of the simulations of the Community Climate System Model used the Model Coupling Toolkit." This is good--although the skeptic in me wonders if it means that any errors in the Model Coupling Toolkit are therefore replicated in all models that use it...
• The University of Arizona researcher Joellen Russell, working with a group of researchers from National Oceanographic and Atmospheric Administration/Geophysical Fluid Dynamics Laboratory, Princeton, New Jersey, published (in December 2006) a paper called "The Southern Hemisphere Westerlies in a Warming World: Propping open the Door to the Deep Ocean,” which may specifically address some of the IPCC study's self-identified areas of weakness. Although I'm just interested in the fact that the research took place within the past 5 years, the results (that the Southern oceans can absorb more carbon dioxide and additional heat, perhaps reducing the rate of global warming), are not in strict accordance with what the IPCC is publishing.

I'll do a sequel to this post very soon. I don't like blog posts that go on forever, and besides, breakfast is waiting.

Authorgeddon and Book Publishing

Well, I guess we need to hope that the Singularity happens before 2052. That's the year when Authorgeddon is predicted to happen, according to Wikipedia. In case you're wondering, Authorgeddon is the moment in time when the number of books published in the U.S. is expected to surpass the number of people in the U.S. who are willing to admit having read a book in the previous year.

"A study by print on demand company Lulu.com pinpoints 2052 as the year when Authorgeddon will arrive. The study arose from statistics published by R.R. Bowker, the company that controls ISBN distribution in the U.S., which reported that the number of books published in America in 2004 hit a record 195,000 -- a 14% increase over the previous high of nearly 175,000 recorded the year earlier. Bowker reports that the average annual rise in published titles over the last three years has been 14.6%."

That 14% increase looks pretty interesting--if it held up it would mean that the number of books published in the U.S. would double every 5 years. So, how does that stack up to real publishing statistics?

From Bookwire (owned by RR Bowker):

1993: 104,124

1995:  113,589

1999:  119,357

2000:  122,108

2001:  141,703

2002:  147,120

2003:  171,061

2004:  190,078

Sadly, in 2007 the number of books published in the U.S. plummeted to 172,000 new titles (I'll bet it's that damn Internet thingie). The overall CAGR is 3.65%, which would double every 19.33 years. But even looking at statistics to 2004 only (2005 might have been a blip, after all) it gives us CAGR of 5.14%. So even by cherry-picking the time frame to include the rosiest picture of growth, we get about 1/3rd of the growth we need. The concept of human knowledge doubling every five years did not originate with US book publishing statistics, I'm afraid.

Back to Work

I'll be writing a lot more about the Singularity later. Let's go back to the basic building blocks that will educate my later writing.

We return to the numbers of people in higher education. If this number is growing exponentially, there might be reason to think human knowledge will too, in future. Specifically, in India, we find "India's provision for higher education rose from a meagre 200,000 at the time of Independence to an astronomical 5,000,000 college and university places by 1998." Those words come from Gajaraj Dhanarajan, President and CEO, The Commonwealth of Learning, Vancouver, Canada. He was speaking in August of 2000 at the 11th General Conference--Universities as the Gateway to the Future.

Let's see--India achieved independence in 1947 (right? Might be off a year--Nope. Google says August 15, 1947). If Indian university students had doubled in number every five years since then, starting from a base of 200,000, they should now have 516 million university students. So, if they have started growing exponentially, it is recently. (How many university students should India have? Using figures from Dhanarajan's speech, to have the same rate of university enrollment as Canada {5%}, for example, India should have had 50 million university students in 2000. Room for improvement here).

This is actually fairly important. The Compound Annual Growth Rate (CAGR) for the Indian university population is 5.51%. The CAGR needed to double in five years is 14.87%. A CAGR of 5.51% means the population doubles every 13 years, instead of five. We won't get where we need to be at this rate.

If you've been reading from the bottom up, you'll be able to join in the refrain--"Great growth. Not doubling."

One... Singular Dilemma

Singularitarians say that computing will advance to the stage where artificial intelligence outstrips human capabilities and then will happily begin designing even better artificial intelligence. I hope they're right.

But hand on heart, I honestly believe that if we had the magic box (or network of boxes) that could compute as quickly and efficiently as it will in 2035 (one of the dates proposed for arriving at the Singularity), we couldn't make it work. We couldn't program it to do what we expect AI to do at the time of the Singularity.

Human knowledge will have to increase in order to make the box work. I suspect (and am searching for evidence) that it will have to increase exponentially. (The alternative is finding a genius, which could happen at any time, but we can't count on it. That genius may exist, but have died yesterday in Darfur or Baghdad.)

If human knowledge is not increasing exponentially (or if we don't find and train the right genius), I don't think the Singularity can happen. Ever.

Time to Get Serious

If you type the search string "human knowledge doubles every five years" into Google today, it will return about 1,580,000 results.

The first result is from a Bellagio newsletter reporting on a conference, and it uses the sentence as if it were a given. The second is from President Clinton's 1998 speech to the AAS, where he asserts it as a given. The first post in this weblog quotes from it.

The third result is why I started this weblog. Here is part of the opening from The Low Beyond, by Eliezer Yudkowsky.

1: The End of History

It began three and a half billion years ago in a pool of muck, when a molecule made a copy of itself and so became the ultimate ancestor of all earthly life.

It began four million years ago, when brain volumes began climbing rapidly in the hominid line.

Fifty thousand years ago with the rise of Homo sapiens sapiens.
Ten thousand years ago with the invention of civilization.
Five hundred years ago with the invention of the printing press.
Fifty years ago with the invention of the computer.

In less than thirty years, it will end.

At some point in the near future, someone will come up with a method of increasing the maximum intelligence on the planet - either coding a true Artificial Intelligence or enhancing human intelligence.  An enhanced human would be better at thinking up ways of enhancing humans; would have an "increased capacity for invention".  What would this increased ability be directed at?  Creating the next generation of enhanced humans.

And what would those doubly enhanced minds do?  Research methods on triply enhanced humans, or build AI minds operating at computer speeds.  And an AI would be able to reprogram itself, directly, to run faster - or smarter.  And then our crystal ball explodes, "life as we know it" is over, and everything we know goes out the window. "

Yudkowsky, like Ray Kurzweil and other advocates of what is known as the Singularity, extrapolates progress in computing sciences from past performance increases in computer efficiency, networking and falling prices for computation and memory. At some point in our lifetimes, computing will be so fast, so efficient, so networked and have access to so much memory, that we will be able to create artificial intelligence (AI). We will then be able to assign this AI the task of creating an even better computing machine, as well as assisting human intelligence in solving real world problems. Because of the increased capabilities of AI (which include indefatiguability and no need for coffe breaks as well as very high intelligence and concentration), progress will be so rapid that life as we know it will change dramatically and in a very short timeframe.

I really want this to be true. But I am afraid that a) we are exaggerating our progress on many fronts and b) even if we can build this better box, we will need to improve our knowledge exponentially before the box is built so we know what to put into it. And I see no evidence as yet that we'll be able to fill the box.

We'll be talking a lot about this--and I hope to involve some of the protagonists in the discussion. But if this isn't your cup of tea, you can adjust your visits to this site accordingly.

The Other Half Of This Experiment

Of course, the other half of this experiment is to see if I can do something relevant and important with free tools available to all. If I can double human knowledge regarding the doubling of human knowledge, all for free and for fun, I think that might be important in and of itself.

Happy Coincidence

A chart I put up on the weblog cited Derwent's World Patents Index as a source. It is found at Thomson.com, and it has information regarding 31.66 million patent documents. (Okay, let's future proof this--in 2012, will it have information on 63 million patent documents?)

They charge to get in and look around, so I'm hovering on the periphery, seeing if I can find more meta-data about the database. For example, an article titled 'Open access to the medical literature: How much content is available in published journals?' by Marie McVeigh and James Pringle, published in 2005, indicates that about 20% of medical literature published in 174 journals since 1992 is 'open access', that is, freely available online. They estimate that a bit less than 60% of medical literature in that time period can be accessed electronically. We also find that Thomson's ISI Web of Knowledge has 20 million users from 81 countries accessing a database of 22,000 journals, 23 million patents and much more. All good supporting data for my second hypotheses, about the increasing availability of human knowledge. But I would love to see the growth rate for new data coming into that database, and I can't find it.

However, I did find this--an article called 'Eastward ho! --The geographic drift of global R&D' written in March 2007 by Bob Sternbridge. It touches on a number of topics I've written about below, and has some interesting numbers:

• The number of computing patents filed in China has risen from 2,000 to 20,000 in five years! Clear doubling in the right time frame. (Pity the increase was not replicated in Japan or Korea, which were also included in the study)
• The number of communications patents filed in China has also risen in the same time frame, from 8,000 to 30,000--again a doubling! Sadly, again not replicated in Japan or Korea.
• More importantly, the total number of Chinese patent applications is growing by 25% per annum. But the article then says that Japan and US patent filings have been flat or declining in recent years. And a lot of the patent filings are by multinational corporations
• R&D spending in China has risen from 89.6 billion yuan in 2000 to 245 billion yuan between 2000 and 2005
• The number of researchers in China grew 77% over the ten years from 1995 to 2004

I think it will take me a couple of hours to digest this and incorporate it into previous posts. More later.

An Anecdote Regarding Hypothesis 2

The second hypothesis I am investigating with this weblog is that the availability of existing human knowledge is expanding exponentially. If true, I think it would reduce the burden of expectations on those who think new human knowledge is doubling every five years, if that turns out not to be the case. If a brilliant Chinese programmer can see the appropriate article online today, where he couldn't in the past, it might guide his efforts more effectively--pretty common-sensical, I think.

So, this article from the Dalhousie News is interesting. In it, we see that the Nova Scotian university now has access to 15,000 academic journals--up from 4,000 five years ago. Essentially, these common-sense Canadians formed a buyer's club and got good deals on academic journals online. Good for them.

I'm starting to like Hypothesis Number Two. Every time I search for evidence regarding the increased availability of human knowledge, I feel like I'm swimming with the river, easily finding data that seems reliable and pertinent. Every time I look for evidence about the primary hypothesis (that actual human knowledge is doubling every five years), I really feel like I'm swimming against the current.