Science and Faith

[tongodeon]

There are a few related comments that I keep coming across in various forms:

"Science requires faith (or is a kind or faith) just like religion. Everybody has faith in something."
"Science can't replace religion because it can't answer questions as well as religion."
"It's not a question of whether we have faith, it's what we have faith in ...

Comments

[tongodeon]

If you raise a plant from seed in a situation where dirt is up and gravity is down, it won't be able to infer that gravity and dirt are opposing and will grow wonkily.

The principle of uniformity says that general principles are generally true, not universally true. Dirt is generally, but not universally 'below'. Plants (and everything else) sometimes make incorrect inferences in unusual situations based on overly broad assumptions of uniformity. A dog might sit, expect a treat, and get scolded in the unusual situation where the person is just some neighbor kid who wants to mess with the dog. A person might incorrectly infer that a skeleton they've never seen before must be an animal that they've never seen before. Inference isn't an absolutely reliable method of acquiring perfectly reliable knowledge, it's just a principle of reward-driven adaptive systems.

I'm not sure what you mean by "collective genetics", but I wouldn't say that genetics "accept empiricism". It's the other way around. You have to accept empiricism to accept that genes exist or that adaptation happens. Genetic codes might create an organism with sensory organs, and those sensory organs might inform that organism of changes in its environment, and the organism might 'decide' to change its behavior because it 'believes' the external stimulus represents a reality about the world that should respond to, but the genetic code itself is just information.

[flwyd]

I was taking issue with the claim that plants are capable of inference. Inference is a thought process, and plants are incapable of thought (AFAIK).

[tongodeon]

Inference is a thought process

Inference is a way to weight hypotheses with success heuristics using logic, which often but not always involves "a thought process". It's possible to write inference engines that discover knowledge without conscious thought.

Every organism uses either direct inference or quasi-inferential processes to whatever extent they are capable. Dogs perform inference without rigrous formal logic, but they do this trick as well as they're capable of doing, and enough that most people can call it inference and give them a pass. Plants are even less capable of performing rigorous formal logic than dogs, but they still perform inference as much as they are capable of.

I'd actually argue that the whole process of genetic mutation and natural selection is a kind of biochemical inference engine. New "hypotheses" are generated via mutation, the success or failure of those genes are weighted by a survival heuristic, and the system 'intuits' strategies that will be successful for the following generation. Like computational inference engines the system arrives at 'knowledge' of encoded inheritance without any conscious thought.

[flwyd]

It's possible to write inference engines that discover knowledge without conscious thought.

Agreed. I think we disagree about what "knows" the things discovered by the mechanical process. Does the machine know it, or is it the agent that interprets the machine's results?

Every organism uses either direct inference or quasi-inferential processes to whatever extent they are capable.

"Uses" or "takes part in?" I'd argue that grass has a capability of 0 on this front.

I'd actually argue that the whole process of genetic mutation and natural selection is a kind of biochemical inference engine.

That's what I meant when I was talking about collective genetics. I think it may be possible to helpfully describe grass qua species as using a process of inference while individual grass organisms are not themselves inferring. It's something of an ultimate/proximate distinction. I also don't think you've fully made the case for this view, but it's not a simple argument.

If you are, in fact, suggesting that individual blades of grass are capable of inference, what are some things that are incapable of inference?

[tongodeon]

Does the machine know it, or is it the agent that interprets the machine's results?
"Uses" or "takes part in?" I'd argue that grass has a capability of 0 on this front.

It's frustrating that I've written "to the extent that they are capable" twice so far, and you seem to think that this a point on which we disagree. Of course grass, computers, and the process of evolution aren't capable of understanding the knowledge that they discover or the formal process of that discovery. They engage in quasi-inductive, knowledge-yielding behaviors even in the absence of comprehension. Dogs might (or might not) understand that "good behavior" earns them "treats" without realizing that this constitutes a concept called "knowledge". People might understand that a shortcut to the grocery store constitutes knowledge, without understanding that the process they employed to find it is called "inference". My point is that inferential processes are broadly applicable and generally useful, not that everyone maintains a formal and self-aware understanding of that process while they're doing it.

I think your distinction about "uses" or "takes part in" is both a false choice and not particularly relevant. Were Dorian ironworkers "using" or "taking part in" empirical observation to determine optimal forging processes before Aristotle noted that physical experience impresses memories upon the tabula rasa of the mind? Were British generals "using" or "taking part in" inductive reasoning before Thomas Bayes formalized the rule that even if all foreign spies are French people this doesn't mean all French people are were foreign spies?

Basically, the philosophical underpinnings of empiricism, uniformity, and induction are post-hoc explanations for a system that served us well long before people knew whether it was superior to the alternative, let alone why it worked. Knowing how it works can be both helpful and satisfying, but induction also works just fine without consciousness, comprehension, or self-awareness.

[flwyd]

It's frustrating that I've written "to the extent that they are capable" twice so far, and you seem to think that this a point on which we disagree.

Grass also plays the tuba to the extent that it's capable, but saying that is kind of weird. So I'll agree that it does it to the extent to which it's capable, I just think it's not capable.

I'm not arguing that you need a theory of mind to use induction. I'm arguing that inductive reasoning is still reasoning and you need some form of concepts (or at least symbols) in order to use it. And I'm not convinced that plants are capable of concepts, but I think dogs are.

[tongodeon]

Grass also plays the tuba to the extent that it's capable, but saying that is kind of weird. So I'll agree that it does it to the extent to which it's capable, I just think it's not capable.

This is a useful metaphor, because it lets me know you don't understand what I'm trying to explain.

Grass can certainly grow in the presence of a tuba, but its actions are completely tuba-agnostic. It is not drawn to, repulsed from, or otherwise influenced by any particular part of the tuba. Its roots might grow *around* a tuba, purely because the tuba is impermeable to roots, not due to the instrument's tuba-ness. Grass would be capable of identifying the mouthpiece and valves of a tuba and growing particularly well in those areas, but it doesn't even play the tuba that well. It does not receive any tuba specific stimulus, nor does it provide any tuba specific response. Therefore cannot be said to be even slightly "playing", or doing anything else, with tubas specifically.

On the other hand grass is extremely gravity-aware. By virtue of its internal chemistry, grass is able to determine the direction of gravity and send its roots in that direction, just as its leaves are drawn away from gravity. The roots of grass grown in a centrifuge run perpendicular to the acceleration of the centrifuge. It receives gravity specific stimulus and provides very specific response. Gravity is a demonstrably familiar and important 'concept' to grass in a way that tubas are not.

I'm arguing that inductive reasoning is still reasoning and you need some form of concepts (or at least symbols) in order to use it.

As you've previously told me, I think you need to do a lot more background work to support the assertion. It's true that most of the induction that you and I come into contact with is based on intelligence, concepts, and symbols, but I'm not so sure that it's a necessity. For example I think it would be possible to train a variety of otherwise non-intelligent animals to run unfamiliar mazes quickly once they infer a code like "red means go left, blue means go right".

[tongodeon]

If you are, in fact, suggesting that individual blades of grass are capable of inference, what are some things that are incapable of inference?

Blades of grass are capable of inferring the direction the sun and earth, the season, the presence of nearby grasses, etc. Knowledge within its area of competence. Blades of grass aren't capable of inferring knowledge outside its area of competence, like what's wrong with my DSL connection. I can't (on my own) infer the direction of a bank robber but a bloodhound can, because that's its area of competence not mine.

Depending on how you define "knowledge", and how "quasi" you want to get with your definition of quasi-inferential processes, almost anything can end up in a situation where past outcomes feed back into future optimizing modifications. Water can reveal the "knowledge" of a smooth path of decreasing elevation from source to sea, or the crystalline cleavage planes of rocks as it freezes and expands, but obviously the water isn't conscious of its situation or the knowledge these processes reveal.

[flwyd]

I think I'm significantly more to the less-quasi spectrum than you are. I wouldn't describe grass as inferring where the sun is so much as growing toward the light. In a climate-controlled room with constant artificial lights, grass will grow toward the light no matter what evidence is available to it that it's not the sun.

I don't think it's helpful to say that water infers the easiest path downhill, (you could just as easily say that the Earth infers the potential energy of the water or something). However, it's entirely appropriate to say "I inferred that downhill is that way because that's the direction water is flowing." We have a concept of up and down, but water doesn't and neither does grass.

[tongodeon]

In a climate-controlled room with constant artificial lights, grass will grow toward the light no matter what evidence is available to it that it's not the sun.

Sure, but so what? A plant in a climate-controlled room can also be fooled into inferring that it's in a different season and flowering or going to seed. A dog can be fooled into inferring that I've thrown a stick when I'm hiding it behind my back. A child can be fooled into inferring that I've taken a quarter from his ear. Shrewd investors can be fooled into inferring that Bernie Madoff offers a sound investing product. Inferences are not guaranteed reliable.

I don't think it's helpful to say that water infers the easiest path downhill, (you could just as easily say that the Earth infers the potential energy of the water or something).

No you can't. An inference isn't just anything that happens, it's an end state reached by the consequences of a series of applied heuristics which includes an intuited next step in the pattern. The continuous or intermittent flow of water over an irregular surface alters the areas of the surface whose contours attract the water, shaping the earth in a feedback process that creates a more efficient path than existed previously, a mechanical analog to the process of inference. It's obviously extremely tenuous to call this "induction", but this is exactly the sort of far-reaching bordeline case that you asked me to illustrate my concept with. I might as well ask you what earliest and most primitive invention might be called a "motorcar", and then take you to task because it bears little resemblance to modern automobiles. This is the most remote and tenuous quasi-inferential 'behavior' that I can imagine, so of course it's going to be missing all but the most essential of hallmarks.

[flwyd]

I think the most succinct statement of my position is "Mere stimulus response is insufficient for induction." I'm not sure exactly where the dividing line is, and there's probably some fun fuzziness around it. I think you're saying that evolved stimulus response counts as induction. It's a semantic disagreement.

I wasn't trying to say that ability to be fooled rules out induction; in fact, the ability to be wrong about the result is what distinguishes induction from deduction. Rather, I was trying to note that given a light source, there's no evidence that can lead the grass to "believe" that light is not the sun. A dog, after searching for a while, can modify its belief based on the lack of stick evidence. And if you fake it out often enough, it will stop inferring that arm motion implies there's a stick to chase.

[tongodeon]

I think the most succinct statement of my position is "Mere stimulus response is insufficient for induction."

I agree. When a doctor performs a patellar reflex test, your knee jerk response to the doctor's stimulus is definitely not induction. Induction is obviously more than that.

I think the most succinct statement of my position is that there is a broad spectrum of behaviors within the general category of "inference", from professional logicians making deliberate, formal inductive proofs to laypeople performing informal induction to very simple quasi-inductive animal responses to even simpler bacterial and plant responses to purely deterministic mechanical processes. The borderline cases are called "borderline" because they're almost completely dissimilar from the stuff at the center. You're "refuting" my position by pointing out that the extreme borderline are missing nearly all the familiar hallmarks of induction. That's not a refutation, that's my point. You're also ignoring the passing similarities, which are also my point.

You might as well look at a gradient from black to white and argue that the lightest shade of gray has barely any black in it at all. Of course it doesn't - that's why it's an extreme example of a very light shade of gray.

I was trying to note that given a light source, there's no evidence that can lead the grass to "believe" that light is not the sun. A dog, after searching for a while, can modify its belief based on the lack of stick evidence.

Your analogy almost has it right, but I need to tweak it a little.

A dog doesn't care whether the thing you're throwing is a natural stick or an artificial frisbee or piece of PVC pipe any more than a plant doesn't care whether the light is actually the sun or not. A dog will happily continue to chase after an artificial stick, and a plant will happily continue to grow toward an artificial light. If you just move your arm around without throwing the stick the dog will eventually realize that the stick is gone and modify its behavior by not looking for the stick anymore. If you just put a lightbulb near the plant without turning it on the plant will eventually "realize" that the light is gone and modify its behavior by turning pale, not putting its dwindling energy reserves into making useless chlorophyll.

I picked gravitropism because it illustrated the simplest of quasi-inductive plant response, but plants make more complicated "decisions" than that. For example plants take input from their environment and modify their "belief" about the season or available rainfall or existence of nearby plants or existence of nearby predators or symbiotic pollinators. Using these behaviors, shrewd gardeners can "train" some plants to grow in certain ways, just as animal trainers can train dogs to behave in certain ways. But obviously there are very many things that a plant will never be able to "understand" or do, just as there are a few things that a dog will never understand or be able to do, because dogs are a lighter shade of grey and plants are an extremely light shade of grey on the broad spectrum of inductive ability.

[tongodeon]

I was taking issue with the claim that plants are capable of inference. Inference is a thought process, and plants are incapable of thought (AFAIK).

Granted I wasn't aware of this research at the time I wrote this post, but it's related and interesting nonetheless.

http://www.bbc.co.uk/news/10598926