Monday, May 01, 2006
The Stuff of Science
One of my fantasy project ideas - one that could keep a small army of graduate students busy - and potentially be a tool for the human race - is what I've been calling CLEW - the causal logic evidence web. This would be a internet-based sort of scientific hypothesis wikipedia thing. A networked hypothesis space that links to the evidence behind each hypothesis. I've thought about this a fair amount. For a while, I thought I had figured out how to do this. But, then, I began to wonder if science and hypotheses were really so interchangeable. This is when I began to think I needed to do more groundwork.
Here's the question: What is the stuff of science? In 5th grade, they taught you all about hypotheses, evidence, double-blind experiments and such. If you are a scientist, you might have listened. If not, you probably were annoyed and daydreamed.
I think hypotheses are an artifact of the human brain - and not the real stuff of science. Humans don't think in "systems" - they think more simply. Science is really a model of the system of the world. If we make the mistake of limiting science to techniques that are easy for humans to think about, then we will really limit our understanding and exploration.
Most science is posed in one of two forms: mathematical or cause/effect relationships. Both are pretty compact and easily communicated to someone who understands math or the vocabulary of their field. But, systems require more than math or cause/effect to describe them effectively. For example, a car is a system of sorts. You can describe many things about a car with math and cause/effect relationships. Acceleration is math. Gas mileage is math. "Turning the key starts the engine" is cause/effect. You will know a lot about a car from these descriptions. But, it's not enough. Physics, astronomy, and chemistry lean heavily on math. But biology is big on cause/effect. My CLEW idea was born out of medicine - so, I was thinking cause/effect.
Now, rumbling around in the back of my mind is all my experience as a computer scientist - especially a computer language specialist. I invented languages in my early career. It was lots of fun and I understand a lot about computer languages at a theoretical level. There are very sparse ways to describe computational capability, like Turing's. But, in practice people program with some more descriptive and powerful tools. If I want to specify a system - like the system of the world - I can do this the most effectively with a slightly larger toolbox. Functions are there - the math. Cause/effect forms are there - conditional execution like "if.. then .. else". But there is also sequence and iteration. There are also objects, attributes, and inheritance. There are variables and constants - the "x" of algebra and the numbers that the "x" can really be. The "mother" of "mother knows best" and your actual mother. I've programmed in languages that are mostly math (APL) and languages that are mostly cause/effect (CLIPS, Prolog). These languages are pretty limited at describing things. All commonly used languages have all the features. You need them to effectively simulate the real world. And you must understand that almost all programming has an element of simulation or description. Even a banking system is simulating the physical banking process. You have a logical account - that is ultimately little magnetized spots on a hard drive platter somewhere - and can withdraw abstracted money from it and deposit abstracted money to it. These operations are really simulations that mimic the processes of putting money is a box and taking it out - the action that would have happened 300 years ago.
So, I am thinking about whether science is better described with a richer language - one that has all the features of a good programming language. A good test of this would be the Piagetian concepts that are described in the book I am reading (Political Reasoning and Cognition: A Piagetian View). There is sequence there - the four levels. There is iteration over each conceptual focus as it is transitioned to the new level. There are "objects" - the concepts. Could you describe Piaget's concepts with a "science language"?
Interestingly, I suspect a major objection to this approach is its complexity. This seems funny to me to even think about. If I could describe Piaget's ideas succinctly in 3 or 4 pages of formal language - this is far more succinct that hundreds of pages of text - the form it is in at this time. If a system is complex, then it cannot be described with a few equations. Simplicity would be misleading - and ultimately not very useful at predicting behavior. No point in pretending that the simple view is adequate. So, if science is going to scale up, and address complex systems, like living things, maybe we need better tools.
One of my fantasy project ideas - one that could keep a small army of graduate students busy - and potentially be a tool for the human race - is what I've been calling CLEW - the causal logic evidence web. This would be a internet-based sort of scientific hypothesis wikipedia thing. A networked hypothesis space that links to the evidence behind each hypothesis. I've thought about this a fair amount. For a while, I thought I had figured out how to do this. But, then, I began to wonder if science and hypotheses were really so interchangeable. This is when I began to think I needed to do more groundwork.
Here's the question: What is the stuff of science? In 5th grade, they taught you all about hypotheses, evidence, double-blind experiments and such. If you are a scientist, you might have listened. If not, you probably were annoyed and daydreamed.
I think hypotheses are an artifact of the human brain - and not the real stuff of science. Humans don't think in "systems" - they think more simply. Science is really a model of the system of the world. If we make the mistake of limiting science to techniques that are easy for humans to think about, then we will really limit our understanding and exploration.
Most science is posed in one of two forms: mathematical or cause/effect relationships. Both are pretty compact and easily communicated to someone who understands math or the vocabulary of their field. But, systems require more than math or cause/effect to describe them effectively. For example, a car is a system of sorts. You can describe many things about a car with math and cause/effect relationships. Acceleration is math. Gas mileage is math. "Turning the key starts the engine" is cause/effect. You will know a lot about a car from these descriptions. But, it's not enough. Physics, astronomy, and chemistry lean heavily on math. But biology is big on cause/effect. My CLEW idea was born out of medicine - so, I was thinking cause/effect.
Now, rumbling around in the back of my mind is all my experience as a computer scientist - especially a computer language specialist. I invented languages in my early career. It was lots of fun and I understand a lot about computer languages at a theoretical level. There are very sparse ways to describe computational capability, like Turing's. But, in practice people program with some more descriptive and powerful tools. If I want to specify a system - like the system of the world - I can do this the most effectively with a slightly larger toolbox. Functions are there - the math. Cause/effect forms are there - conditional execution like "if.. then .. else". But there is also sequence and iteration. There are also objects, attributes, and inheritance. There are variables and constants - the "x" of algebra and the numbers that the "x" can really be. The "mother" of "mother knows best" and your actual mother. I've programmed in languages that are mostly math (APL) and languages that are mostly cause/effect (CLIPS, Prolog). These languages are pretty limited at describing things. All commonly used languages have all the features. You need them to effectively simulate the real world. And you must understand that almost all programming has an element of simulation or description. Even a banking system is simulating the physical banking process. You have a logical account - that is ultimately little magnetized spots on a hard drive platter somewhere - and can withdraw abstracted money from it and deposit abstracted money to it. These operations are really simulations that mimic the processes of putting money is a box and taking it out - the action that would have happened 300 years ago.
So, I am thinking about whether science is better described with a richer language - one that has all the features of a good programming language. A good test of this would be the Piagetian concepts that are described in the book I am reading (Political Reasoning and Cognition: A Piagetian View). There is sequence there - the four levels. There is iteration over each conceptual focus as it is transitioned to the new level. There are "objects" - the concepts. Could you describe Piaget's concepts with a "science language"?
Interestingly, I suspect a major objection to this approach is its complexity. This seems funny to me to even think about. If I could describe Piaget's ideas succinctly in 3 or 4 pages of formal language - this is far more succinct that hundreds of pages of text - the form it is in at this time. If a system is complex, then it cannot be described with a few equations. Simplicity would be misleading - and ultimately not very useful at predicting behavior. No point in pretending that the simple view is adequate. So, if science is going to scale up, and address complex systems, like living things, maybe we need better tools.
Subscribe to Posts [Atom]