ASA9

Another Antarctic Ice Shelf Diasappears Ahead of Schedule

2009-04-04T19:57:00.004-04:00

Tragically, polar melting is outpacing most climate models.
http://www.reuters.com/article/environmentNews/idUSTRE5332BU20090404

Which models you ask? Those ones that predict global desertification in 90 years. So granted that these climate models might be wrong and they have large error bars. But it seems to me that we are seeing evidence that warming is occurring at the UPPER end of those error bars. So maybe Earth will be a desert in 70 or 50 years instead of the 90 predicted.

The Great Filter Discovered

2009-03-24T15:58:00.006-04:00

We have never discovered extraterrestrial life. Why not? Something must be stopping life from getting sufficiently advanced. This something, mysterious as it is, has been called The Great Filter.

What is The Great Filter? It could be as banal a idea that intelligent life like ourselves is just a very rare thing. So many conditions have to be just so, that in fact it arises only once per trillion stars say. But more ominously, The Great Filter could be the inescapable tendency for intelligent civilizations to wipe themselves out.

As for the first possibility: We are finding a lot of planets orbiting stars already. Clearly there are many, many planets out there. Organic chemicals are quite common, we know from comets. Are planets like ours really that scarce? Admittedly they have to be watery, far from big clusters of stars and associated frequent gamma ray bursts, and in a system with some outer gas giants to shield them from collisions. But I doubt such conditions are all that rare.

So we're left with that sticky little second possibility: that civilizations invariably wipe themselves out before getting to the technological state to spread through the galaxy.

Sounds nutty that we can observe no little green men and conclude therefore that we are likely doomed to be wiped out soon. Yet there are 100 billion stars in our galaxy, and they have had billions of years of time to colonize the galaxy (or to build space probes called "Von Neuman Probes"). It would only take about 20 million years to reach every star in the galaxy, assuming probes could reach 1 percent of the speed of light. So little green colonists or Von Neuman Probes should have been here for billions of years. Yet they are not. Why not?

Well I think that recent events have provided us with a likely answer. The Great Filter is... climate change. Here is my reasoning.

Over the millions of years required to develop intelligent life, planets tend to accumulate carbon. This carbon gets sequestered in various places around the planet, such as gaseous carbon in cold places (methane trapped in permafrost, methane stores in the ocean floor) as well as liquid and solid forms, stored in warmer, more accessable places (oil fields, coal deposits). Such carbon is too easy a fuel source, and an intelligent civilization will invariably learn to burn the warmer sources of carbon, achieving rapid growth in population.

Every planet has a limit at which it will be overwhelmed with the greenhouse effect achieved from burning carbon (i.e. from CO2), to the point that a warming trend eventually releases the trapped gaseous carbon, sealing the planet's fate. It so happens that planets can hit the released carbon tipping point much sooner than they can learn to build intragalactic rockets or Von Neumann Probes. They hit the tipping point before they can develop technologies to reverse it (carbon sequestration). Even before they have time to convert to carbon neutral energy sources (wind, nuclear).

It turns out that our gaseous carbon tipping point will likely occur very soon, like within a decade, according to climate scientists who build predictive models. Our planet will likely be a very hostile place, mostly desert by the end of this century. One climatologist predicts Earth's population to be just 1 billion by 2099.

It's fashionable at this point in the discussion to put a positive spin on things and say we can solve our problems. I see no reason to do this. Positive spins just promote complacency and inaction. There is no reason for positive spin here. We can pretty much be certain that 6.76 billion Earthlings are not going to come together and stop burning fuels. We are not even going to come together and slow the rate at which we increase burning fuels any time soon. Carbon sequestration technologies are not going to happen on a large enough scale.

Our planet, like the 10 to the x ones before it, will wipe itself out due to climate change. And that will probably occur over just the next couple centuries. Even today's child is likely to experience major hardship or early demise because of this problem. Sorry folks.

That said, I do not believe in fatalism. I intend to do all I can to delay this. The more generations we can eek through our post-industrial Fools Paradise the better.

Q: How many vicious cycles does it take to kill a planet?

2009-03-06T16:18:00.004-05:00

Answer: 1

But my occasional read of the popular science magazines finds 5.

(1) Warming planet melts polar ice caps, which reduces polar reflectivity, which increases absorption of sunlight, which warms planet, ...

(2) CO2 acidifies the oceans to the point that calcium carbonate dissolves, killing the organisms that have been sequestering carbon, which increases CO2, ...

(3) Melting permafrost releases methane, which has ~23X the global warming potential of CO2, which causes warming, which melts permafrost, ...

(4) Warming oceans release methane trapped near the bottom, which warms oceans, ...

(5) Human technology (a) invariably through history increases human consumption and (b) increases human technology, which ...

See this alarming article. It says that many climatologists' models predict that the world will be mostly uninhabitable just 90 years from now.
http://www.newscientist.com/article/mg20126971.700-how-to-survive-the-coming-century.html

We humans need to wake up and get our heads around the very real likelihood that we will destroy our world in but a century.

What differentiates INQLE?

2009-02-13T21:55:00.004-05:00

People often ask me: Dave, what's so special about this project you've been working on?? Well people seldom do. But you, dear Hypothetical Reader, will get a response nonetheless.

Here are INQLE's innovations, for the record.
(1) I have not seen a wizard-driven semantic data importer like INQLE's. Since ~all data is captured nonsemantically, a good data importer is critical functionality. Specifically, INQLE's data importer (a) makes the process of capturing semantic data and URIs somewhat intuitive, (b) introduces a coherent semantic framework for organizing (measured) data from unchanging attributes, which are more identifiers, and (c) it captures both explicit data (data in the table) and implicit data (data that describes the table)

(2) INQLE introduces sampling (i.e. gathering learnable data), by issuing SPARQL queries against internal data stores and/or remote SPARQL endpoints. Our first sampling algorithm, Simple Subject Sparql Sampler, demonstrates a method whereby semantic arcs are discovered, and used to identify learnable data withing an RDF graph. Semantic graphs (Jena Models) are assembled by issuing SPARQL CONSTRUCT query to capture these arcs. Multiple RDF graphs/models can be combined easily, to form an aggregate learnable data set. So local and remote data can be combined andmined.

(3) INQLE introduces automated machine learning on semantic data. I have not seen that elsewhere.

(4) INQLE exposes the pieces of the learning cycle, such as data harvesting algorithms ("samplers"), as plugin-friendly extension types (in an OSGi container). So it expects to have new sampling & learning automated algorithms plugged into it.

(5) Whereas there has been some work toward applying data mining to the semantic web, I have not seen the reverse. INQLE applies semantic technology to solve problems with data mining. Most notably, data mining typically works with nonstandard formats of data, with nonstandard naming conventions for the methodologies/learning algorithms/statistical measures. Data mining offers no standard format for publishing experimental results. Data mining offers no means for combining 2 data sets into one. These are examples of problems with data mining, for which semantic technology offers solutions.

That about does it. I am sure I forgot something.

INQLE 0.1.4

2008-10-01T00:47:00.004-04:00

Finally uploading the next release. Skipped 0.1.3 to emphasize the grandeur of 0.1.4. And doesn't it have an air of glory about it? "0.1.4".

Bugs notwithstanding, this version has quite an advanced spreadsheet importer. I have not seen a RDF importer wizard of this sophistication. It can do all of the following things:

It captures as RDF not only explicit data (i.e., that which sits in the spreadsheet) but also implicit data (i.e., data about the spreadsheet as a whole, like "this spreadsheet is a bunch of data collected in Wilmington, Delaware by so-and-so organization").
It does not require users to enter in RDF URIs, as I considered this to be a major usability impediment.
It has a lookup capability for finding subject classes.
It automatically retrieves known properties for the subject in question and provides form elements for entering literal values of thos properties or for specifying which column in the spreadsheet contains the values in question.
It permits users to add their own subjects and/or properties.
It registers new subjects and properties in the Central INQLE Server, and makes these discoverable by other INQLE users running the importer wizard.

The other big step forward off course is release of the Central INQLE Server (CIS), for looking up RDF subjects and properties. The CIS will play a larger role in the future, in coordinating the cooperation between INQLE servers.

Quite a nifty bit o work if I do say so myself! And this is my excuse as to why it was almost 4 months in coming. Hopefully all this reengineering will pay off and make the process of importing data as RDF as easy and comprehensive as possible. My reasoning is that if you can import most any structure of CSV data, then you make it a whole lot easier for folks to use INQLE.

Future work: hook the thing up to the full UMBEL subject database, so we can look up subjects and properties from a richer source. Plus all the billions of other things in the INQLE roadmap.
http://code.google.com/p/inqle/wiki/INQLE_Roadmap

Releasing INQLE version 0.1.2

2008-07-07T22:57:00.003-04:00

This version of INQLE fixes a few minor bugs. It adds relatively little new end-user functionality. However, we have added a new plugin type: org.inqle.datasets. This plugin type permits plugins to specify a new internal dataset for the purposes of containing data associated with them. This facilitates quicker lookups. This was also a prerequisite for us to be able to store incoming info (in the Central INQLE server). This is an important feature which will come in the next release in a few weeks. More later.
Davo Donohue

INQLE version 0.1 is born!

2008-06-22T22:25:00.003-04:00

My open source project INQLE (Intelligent Network of Querying and Learning Engines) has reached the ripe old age of 0.1! I had intended this version to be very bare bones, such that it would barely work. But I found that a few features were needed to make the bloody thing usable. Most notably I added 3 big features:

a wizard for loading spreadsheet data as RDF. This is a pretty powerful feature.
a setup wizard which runs upon starting INQLE for the first time
an embedded database. This dramatically simplifies the process of installing INQLE. To my delight, I discovered that Jena has recently begin supporting the fantastic H2 database, which performs very well as an embedded database (in fact it outperforms non-embedded databases significantly). I find INQLE runs much faster using an embedded H2 database than an external PostgreSQL database.

Next up: probably will add security, probably using OpenID and Google accounts.

INQLE Scores 8.5 out of 10 on killer app scale

2008-06-18T22:53:00.005-04:00

I stumbled across this post on scoring a semantic web application on a 10 point killer scale. [OK, well actually I wrote it.]
So let me score my project INQLE on this scale.

At this writing, IQNLE is very "early doors" (version 0.0.9). Currently, INQLE scores 6 on the 10 point scale. However, our vision/roadmap puts INQLE on a path to score about 8.5:

Immediate Value to User
+1: The tool adds immediate value to the human user. INQLE permits automated machine learning experiments. Users must merely load data and they can then immediately start running experiments.
+1: We are aware of no product that does this.
+1: The tool is free.

Generation of Semantic (RDF) Data
+0.5: INQLE allows users to generate data in spreadsheets (as they are want to do). Users must then use the INQLE interface to import that data.
+1: The new semantic data that INQLE generates are assertions about the correlations that exist between different things in the universe.
+1: Those things which INQLE correlates are real world objects. Some of INQLE's future sampling algorithms will combine local data with remote, pre-existing RDF entities.

Consumption of Semantic (RDF) Data
+1: In future versions of INQLE, users will be able to annotate how valid or trivial or novel or spurious a correlation is.
+0: Such human annotation will require use of INQLE's interface.
+1: Future INQLE algorithms will be able to discover the results of past experiments.
+1: INQLE can then use the power of linked data and semantic reasoning, to perform repeated or related experiments. INQLE servers can therefore accrue an expanding body of knowledge.

So 8.5 out of 10 is pretty decent. But we have to remember that we (and by "we", we mean "I") wrote the damn thing, thru our own myopic specs.

So is INQLE the killer app for the semantic web? Um if your standard for a killer app is Google then probably not. But if you could live with lower expectations and if INQLE could really deliver on its ambition to effect true artificial intelligence and/or revolutionize the way research & discovery is done, then it could deliver some degree of killerness.

Semantic Web Killer App Scale

2008-06-18T22:35:00.004-04:00

Many smart people have asked this question:
"What is the killer app for the semantic web?". Well I do not have the answer to that question. But I can tell you some of the attributes that characterize a killer semantic web application.

I came up with a scoring system you can use for evaluating semantic web technologies. The maximum score is 10.

Immediate Value to User
1 point: The tool adds immediate value to the human user.
1 point: That immediate value to the user is novel functionality that is not available for free elsewhere.
1 point: The tool is free.

Generation of Semantic (RDF) Data
1 point: Use existing human workflows to generate new semantic data.
1 point: Automated computer process generates new semantic data, without direct human involvement.
1 point: Generated semantic data links extensively to pre-existing semantic data, hosted remotely.

Consumption of Semantic (RDF) Data
1 point: Humans may annotate the semantic data through a simple procedure, increasing the value thereof.
1 point: Such human annotation occurs automatically, using existing human workflows.
1 point: An automated computer process can consume the generated semantic data in some useful way. That is, humans are not the sole consumers of the generated semantic data.
1 point: Such automated processing increases the value of the body of semantic data, thereby facilitating cumulative accrual of value by the computer.

Not sure how accurate the above model is for capturing the key features of a semantic web application. For example, maybe it puts too much emphasis on machine processing of data. But that's what the semantic web is all about, right? Most agree that it's not just another paradigm for presentation.

So assuming that above scoring system is good enough, let's try to answer: "What is the killer app for the semantic web?"
Well it will be a tool for generating semantic data, of immediate value, using simple, human + automated methods. Such semantic data is processable by automated agents, in such a way that its value grows with time.

What's the best creative medium?

2008-04-10T20:49:00.004-04:00

Great question, Dave. Well let's break it down. What attributes might distinguish the creative media?

Most features of creative media don't seem to differentiate one versus another. For example most creative media give you a significant buzz (that is, they are fun). True, some are more fun than others. E.g. stamp collecting I would wager is not lighting up much of the elation centers of the brain. Learning Klingon maybe down on the list as well. Just above learning Klingon I would rank my first profession: a lab researcher in a molecular biology lab. This was creative work albiet slow. It seemed that one made a creative decision about once per year. But for the most part, they are a wash here.

All creative endeavors have some form of environmental toll as well. I suppose ones that generate a lot of byproduct would lose in this attribute. Like nickel smelting-4-fun.

What about the value of the end product to society? Now here is a point of differentiation. Clearly some creative exploits are not worth a hill of beans save for the transient PET pattern they create in the beholder for about 5 seconds. I would put most modern art in this category. What about good art? Or musical performance? Well the benefits are subtle, and I would argue small.

What about its potential to make you money? This is a good point of differentiation because it indicates how much the person is completely wasting his time. Measured by this standard, most creative exercizes are a fool's errand. But some media outperform others clearly. I suspect that here again, art fares poorly relative to more technological-related exploits.

Where the hell are you going with this Dave? I hear you ask. Well my thesis is that the finest creative medium ever is [envelope please]. Computer programming!

Eh? A hush fell over the Readership like a choking cloud of chlorine gas. Good thing nobody would ever read this. Well here is why software development is such a fabulous exploit for the few who are lucky/squashy enough to do it.

High buzz per unit time invested. Imagine scientific advancement sped up by a 1000 fold. That's what computer programming feels like to me. It is pretty easy to get yourself humming on a project in which you are tinkering with code and running a new experiment every half minute or so. No pesky gels or radioactive phosphorus or carcasses either.

High reward to society [I think]. This is hard to figure. Well what is the internet worth? Now throw in the value of non-internet devices. Costly. And all this wealth was created in just the last few decades. I bet well over half of the value of the internet was programmed in just the last few years. Yes I hear your point that the actual content on the internet gets some credit. If this post is at all representative, then I think it's clear that content is overrated.

High potential for financial reward. Yes the days of everybody-who-knows-html-gets-rich are over. But you stack the median joe programmer against the median seth actor and i think in the former case he has a ranch house and a kid and 2 cars versus the latter is sleeping on his friend's couch, still chasing a forlorn dream.

So in sum, I retract all above statements.
Dave

Evidence-Based Everything

2007-10-19T22:27:00.000-04:00

Subtitle: Everything is about prediction is about data mining is about everything.

Here's what I mean:

1. "Everything is about prediction"
All information (books, articles, lectures, even conversations) is basically intended to provide you with a model from which to make future predictions. In some cases, the predictive models are explicitly spelled out, as in "the moral of the story". Other predictive models are more subtle. When your coworker says she thinks your employer is short-sighted, she is predicting that in the near future, that employer will do short-sighted things. When a self-help book tells you that people ate half the number of candies when the candy jar was moved 6 feet away from their desk, that book is providing you with a predictive model that says that mindless snacking goes down as snack food is moved away.

2. "prediction is about data mining"
All predictions are measurable using data mining. What is data mining? It is basically the scientific method, with particular focus on finding statistical correlations from tables of data. Data mining and associated scientific methods can be used to make just about any prediction. The best route to developing predictions about the real world is to use scientific methods, armed with -- you guessed it -- data mining.

3. "data mining is about everything."
Whether the subject matter is interpersonal relationships, dietary habits, or social hierarchies among crack dealers, it can be measured with data. And it should be! Because hiding with data everywhere are shocking findings that often argue against the conventional wisdom. Thus, data mining can dispel our misconceptions and enable us to better predict and manage our future, in all walks of life.

My broader point is that we are very subjective creatures, with an amazing capacity to see the world through tinted glasses. In my field of internal medicine, we have only in the last few decades appreciated the critical need to practice evidence-based medicine. This basically means that we rely on scientific proof that an intervention is justified, and that in the absence of that proof we should proceed with caution.

Before the evidence-based medicine movement, doctors relied exclusively on intuition and basic research into the pathophysiology (the nuts and bolts) behind diseases. The trouble with this approach is that intuition and even the science behind studying pathophysiology of diseases both can and do lie, frequently. Medicine has done many famous about-faces when it finally got around to studying whether an intervention is helpful. Examples include medicine's historic practices such as useless low-protein diets for kidney health, recommending toxic vitamin E to prevent cancer, widespread use of toxic anti-arrhythmic drugs and estrogen replacement therapy.

Bottom line: everything, not just medicine, would benefit from increased use of scientific data mining, to provide us with new and improved predictive models about... everything.

The Global Environment: Let Me Go On Record

2007-10-09T22:47:00.000-04:00

Some things are important enough to restate the obvious. I write this for the benefit of some distant future reader combing thru ancient posts, trying tounderstand why we ruined the planet.

For the record, I understand and firmly believe the following environmental facts:

We are increasing CO2 concentrations to levels never before seen in at least 400,000 years.
This plus other human factors are contributing to major environmental damage in the form of global warming and other forms of global climate change.
Sea levels will almost certainly rise to their maximum levels within the next century or 2 (100 feet above current level) with catastrophic consequences (imagine all of Florida, trash and all, being washed out to sea).
Through climate change and through habitat destruction, we are visiting upon the earth one of the greatest extinctions ever. The pace of this mass extinction surely matches those from prior meteor impacts on a geologic scale. We will soon live in a planet without cheetas, pandas, large primates, many species of whale, most smaller primates, etc., etc.
There is a vast overpopulation of humans contributing to this. I recognize it is fashionable to say that world populations will level off around 11 billion +/- several billion. However: (a) there is no proof of this, (b) even if populations flatten, the environmental impact per person continues to rise, (c) 40,000 children starve to death every day, (d) habitat destruction and other consequences of human activity have wiped out most wild habitats, with the remainder to be consumed within the next few decades.

So obviously I am only touching on a fraction of the damages. As for policy, and the United States & Bush adminstration's role in exacerbating most of these problems, the NRDC's "Bush Record" speaks volumes.

As for solutions, let me go on record as endorsing massive diversion of resources to protect environments, incentivising rainforest countries to preserve what remains, massive effort to invent new energy technologies, etc. For my small part I contribute the the World Wildlife Fund and others.

link on vitamin d

2007-10-06T22:18:00.001-04:00

http://ods.od.nih.gov/factsheets/vitamind.asp

great stuff from the NIH on vitamin D.

How much vitamin D can children safely take? Accorfing to this
document and the NIH, the answer is a whopping 1,000 I.U. for infants
0-12 months, and 2,000 I.U. for all people over 12 months!

I expect we will be seeing recommendations from AAP that kids should
get more vitamin D, to prevent later Multiple Sclerosis plus cancer.

Vitamin D Totally Rocks

2007-09-26T22:48:00.001-04:00

"Totally rocks" is not your every day medical jargon. But the research behind vitamin D is getting to be highly cool. Here's the deal. Not very much of what we do in medicine has hard-hitting research behind it. And where research does exist, it seldom demonstrates a clear benefit to survival or longevity.

Well Vitamin D is different! Check out this meta-analysis.
http://archinte.ama-assn.org/cgi/content/short/167/16/1730

The authors looked at every study ever done in which patients took either vitamin D or placebo, and summed up all that data. Bottom line: vitamin D increases longevity.

2007 is really shaping up to be the year of vitamin D. Earlier this year, 2 meta-analyses came out demonstrating that vitamin D at a daily dose of 1000-2000 I.U. daily appears to decrease the risk of breast cancer by 50% and colon cancer by 2/3.
http://www.physorg.com/news89984353.html
And now add to this a 3rd meta-analysis showing lower death rate.

I don't believe that ever before in human history have we had a pill that we could give you and confidently tell you: "Here, take this. it will prevent cancer and make you live longer."

I know that medicine tends to reverse itself on frequent occasions. However, these are 3 large studies, summing the results of many other studies. This provides considerable assurance that we are on solid ground taking vitamin D. I oficially signed on to vitamin D earlier this year, and I have been giving the stuff out like candy. Now that we can add this meta-analysis showing improved survival, I'm even more sold!

Healthcare Matrix

2007-09-26T20:09:00.000-04:00

On September 26, 2007, I delivered a talk on the US health care system entitled "Healthcare Matrix: Inferior Quality for Twice the Price". The audience is the Arden Georgist Gild in Delaware.

(Georgists are those who promulgate the economic philosophy of Henry George, a 19th century American economist. More on that lovable bunch another time perhaps).

While I find activism about as appealing as door-to-door sales, the thing about the predominantly corporate US Health Care System is that it is so outrageously expensive, wasteful, inequitable, and ineffective that the story tells itself.

To me the biggest surprise of my research into this topic was the Veterans Health Administration (VHA). Roll back the clock to 1995. The VA was the laughingstock of our health care system. But around 1995, the VA embarked on a major restructuring, which included release of its vaunted Vista electronic health record, and aggressive measurement of performance. The results speak for themselves: every single health outcome measure jumped dramatically from 1995 to 1997, and continued to improve thereafter. The same study stacks the VA's performance up against Medicare's from 1997 to 2001. The surprise to me was that not only did the VA outperform Medicare in all measures but 1 (annual eye exam for diabetics in year 2000), but that it bested Medicare by such wide margins. Parenthetically: Medicare outperforms HMOs (specifically, fee for service Medicare outperforms HMO-mediated "Medicare Advantage") and Medicare is much more cost efficient than HMOs.

So if your keeping score,

VA beats Medicare
Medicare beats HMOs

How could the socialistic health system of the Veterans Administration outperform the free-market driven remainder of our health care system? Many Americans are hard wired to think that free markets are always more cost effective than government programs (even in cases like this one, where the profit incentive runs counter to the mandate to deliver health to society). Rather than try to argue against peoples' ideology to explain why the socialistic VA's is supreme over the capitalistic HMOs and private doctors of Medicare, I will say this: It just is.