Sunday, March 9, 2014

Evidence and the Reasoning Subconscious

I began to suspect that nobody, even me, respects evidence. In particular, not Less Wrong nor reactionaries. On reflection, I'm equivocating on 'evidence.' (Or: Aumann's theorem will never obtain.)

First, the sample bias. I can only consider those who write about their beliefs. Perhaps many are convinced by evidence, but would never blog about it, indicating a strong negative correlation.

Second, my counterfactual: there should be heretic neoreactionaries. Someone who buys say HBD but also AGW, because, coincidentally, they've seen a particular subset of evidence for AGW. Similarly, it should be possible to change a NRx's mind, and back again, by using subsets of evidence. We're always using subsets of the evidence, the question is whether it is a representative subset, so it should be possible to see-saw someone's mind by defrauding them with non-representative subsets. This fraud should sometimes happen by ignorance or luck.

Or: too much tribal correlation. Humans make mistakes, and they should occur in a random direction.

On Less Wrong the evidence is less clear, which brings me to the point about evidence.

Third: my equivocation.
There's scientific / natural philosophical evidence. There's also evidence that is difficult to describe, explain, or pass along.

The latter kind dominates subconscious processing.

The subconscious functions much like a machine language layer, which the consciousness communicates to in a very high-level language. It is labour intensive to consciously think. (But, e.g. necessary when learning math. Passing add(2, 5) to the uneducated subconscious returns [?].) It's easy to pass think(apples, red) to the subconscious and get something back...but necessarily, the consciousness does not automatically understand the implementation details of think() and therefore almost can't debug it, as that literally means calling think(think(), bugs). If think() has an issue, you won't get the right answer.

(Note that to learn add() means calling think() in the right ways.)

The subconscious commands orders of magnitude more neurons than the voluntary consciousness, and it has access to the unprocessed data to use that power on. (See transcranial magnetism induced savantism.) Even after somehow gathering the implementation details of think(), it lacks a formal notation. Gathering the information means calling think() on itself. If you invent a formal notation (e.g, think()) a full account of the implementation details will presumably be orders of magnitude larger than a couple graphs and charts. Subsequently, it has to be manually checked by redundantly gathering the information it calls for (like adding by consciously counting) in case there's a typo. There will be feedback - the subconscious is watching the implementation get written and will change as a result.

At this point, I've re-derived the usual conclusion - humans can only communicate if they already agree on 99% of what they're trying to say.  When think() implementations happen to match. In principle we could limit ourselves to describing the differences, but...

My think(Less Wrong, subconscious evidence dominance) returns true. Similarly, think(Less Wrong, ~see saw) and think(Less Wrong, ~tribal dissent). If your think() returns differently, I have no way of finding out why not.

Think() contradicts the notion that complex structures must be common to all individuals in a sexually reproducing species.

My prediction is that your philosophy or political leanings are almost entirely determined by which version of think() you happen to have. Once exposed to Dark Enlightenment thought, whether you buy it is determined by whether your subconscious buys it, which is determined by whether the DE article's evidence profile matches that used by your subconscious.

Charts and statistics are almost irrelevant to this process. Few think() implementations even use them, and of those that do, there's byte asymmetry. Say a conscious visual representation of a chart is worth a thousand times the bytes of its description; then, the subconscious gathered a million times as many bytes on the way to the chart.

I'm interested in knowing if it is possible to summarize what the subconscious generally uses as evidence. (How does one distinguish one article's tone from another?) If the feedback isn't fortuitously arranged, it may make it truly impossible. At least, checking a putative description is easy, as it will predict when and how minds get changed. Or not, as the case may be.

I would predict that most of what's called persuasive, isn't, but I think it is a retrodiction at this point. Most implementations of think(persuasive, argument) are bugged, which is why persuasion in general targets the wrong thing.

For now I decline to address the fact that beliefs are usually decorative, and think() is called by aliefs only.

Like a good passivist, I will hereby refrain from policy recommendations.


Ironically, the subconscious can return things even if not called, but does so in an even higher-level language than think() calls: emotions.

Friday, March 7, 2014

ARFAQ Spot Check: Suicide

Nick Land provides the 1928 suicide rate among kids, almost by accident.

A quick google gets me 2006.

Unfortunately the numbers aren't directly comparable. But, some quick estimates from the available data.

10-14: < 0.5 to ~1.5 per 100k
15-19: ~4 to 7 per 100k.

Slightly more dudes than chicks in this cohort used to kill themselves. Now, it's twice as often in 10-14 and four times as often in 15-19, matching the 1928 adult ratio.

This is not 'fallen by 20%'

So, err, even worse than I thought. Once again, the graph is cut off at a very convenient time. Or perhaps magic. The stats I stumble upon first happen to support my hypothesis - Google/Nick Land can psychically know my thoughts, as well as Alexander's - and therefore the stats Alexander stumbles upon first support his hypothesis too. Very nice of magic, don't you agree? Neighbourly. I thought I would have to skip checking suicide, and then the stats fall right into my lap.

This is even discounting the fact we don't want suicides, we want suicides-that-would-have-successfully-completed-against-1928-trauma-surgeons-and-ambulances, 2006. This is even discounting the fact that we know governments fudge the hell out of their numbers.

I also noticed a double bind. Either murder rates should be going up and are being kept down by trauma advances, OR murder rates should be going down, but stagnation in trauma medicine is holding them steady. Either way, the progressive has to play god of the gaps...assuming they can even think in straight lines long enough to notice there's a problem.

I found the opening sentences of the 1928 report a huge relief. It would appear it was written by an adult. With, like, perspective and stuff. I still wouldn't call the rate large, it's just that wealth should make it go down, not up - the trend is worrying.

Well, would be worrying, were the national rate any of my business. I'm only in this for the epistemology.