Causal Connections and Working Memory
Kenneth
I'm going to put two bits of signal together and see what people make of it.
First: http://lesswrong.com/lw/khd/confound_it_correlation_is_usually_not_causation/
"Confound it! Correlation is (usually) not causation! But why not?" by Gwern. He takes this mantra for granted and explores it.
He concludes that "causal networks' interconnections drive up possible correlations faster than causal connections."
Second: This little tidbit from some site somewhere.
"""
The average human can store 5-9 objects in their working memory, pros
with several years of training can work with around 13. At any given
time, your ability to make connections between unconnected objects, say
pieces of information on the price of oil, the price of wheat, and
drought, you are limited to that number. These factors create
interdependent relationships which do not add or subtract, but multiply
in complexity. In other words, if your method of studying things
involves just reading a lot and seeing what comes out, you aren't going
to be able to make any serious connections. This means in any given
situation you are managing hundreds if not thousands of potentials,
whose data fits multiple mutually exclusive hypotheses. This means that
if you have one of the best working memories in the world and you can
hold 13 objects, you will only work out 1.5% or less of the total
problem in your head. Your goal is to sift through them effectively and
quickly. This is not humanly possible. Your problem needs to be solved
anyway. To do this you must augment your working memory and awareness.
This is what makes Morphology and Analysis of Competing Hypothesis
approaches effective, it is a method for managing working memory as much
as anything else.
"""
This is a fun thing to note. We might have a theoretical driver of why Working Memory, as measure in humans, is correlated with Fluid Intelligence. Furthermore, we possibly have *a unique, testable operational definition of intelligence*. Thirdly but not lastly, we would have given ourselves a whole new way of joining neuroscience to cognitive enhancement to analysis and problem solving in a mathematical manner - as we already do with OLAP cubes and other forms of dimension management.
In this way, we can draft external applications that can extend our problem solving ability in the same way that Google replaced our memories. We're getting new interfaces that can turn Roman Rooms from something inside your head, to something beaming down an Oculus Rift. (Starlight 3D in actual 3D! https://encrypted.google.com/search?q=starlight+analysis+3d&source=lnms&tbm=isch) Except right now we're dicking around with glorified checklists and killing fake people or something.
(In other news Gwern is using more colloquial language and exclamation points lately. Did he move to the Bay Area? Is that why he's so happy? IS THERE A CAUSE HERE OR IS IT JUST A CORRELATION?)
Wolf Tivy
I'll take my usual position of hard skepticism of general purpose software or practice-based intelligence enhancement techniques besides the hard ones (solve lots of hard problems, get stuff done, take responsibility, don't go off the rails philosophically). The analogy to lifting suggests it ought to be possible, but until someone writes Starting Intelligence and can convince a skeptic, I'll defy any given theory.
My model of the lifting/intelligence analogy is this:
The ability to modulate resources spent on strength depending on external conditions is actually a highly nontrivial affair. That doesn't just happen by accident. The mamalian muscular system has a 500 Ma heritage of constant refinement with relatively fixed requirements. Thus you would expect crazy tech like environment-responsive muscles.
Our intelligence, despite being based on 500 Myo basic hardware, is largely the product of a 1 Ma or even 10 ka intelligence arms race in which all stops were pulled and all corners were cut in pursuit of more general/social intelligence. Thus we would would not necessarily expect anything like stress-response hypertrophy to have been kept around, even if it had existed. All applicable knobs are already maxed out.
Still, we know that plasticity and learning exists, so if you practice X, you get better at X. I hypothesize that this has an opportunity cost in that in comes at the expense of Y. Unlike strength, which comes only at the expense of food. That said, Y might be entropy much of the time, which is why it's good to do something hard and useful rather than nothing. There is also natural overlap of knowledge between fields, of course, but what I'm saying is that I don't believe in general hardware hypertrophy *in humans*.
So why are some people smarter or more capable than others? This seems a good place to look for ways to increase cognitive performance.
Mutational load, is as usual the relevent concept to look at. Within a race, genotypes are basically a pure ideal prototype plus a bunch of harmful crud mutations peculiar to specific families. People vary in how much mutational load they have. People descended from nobility have lower mutational load, and thus higher phenotypical performance. (I can explain if necessary). The rest of us have more. For example I'm quite smart, but I have rather savage ADHD and some kind of possibly related social retardation and whatever else I've inherited from my disproportionately crazy family. These are my biggest limitations, and otherwise the cognitive machinery works about as well as you could reasonably expect.
Also, mutational load self-correllates because of assortive mating, so someone with high intelligence likely has high working memory, even if those things were conceptually and mechanically distinct.
So on one hand, you could identify and target your worst cognitive defects, but on the other hand, your phenotype is already relatively fixed.
So yeah, there's your daily dose of cognitive enhancement pessimism. Do hard stuff, etc. Work around your flaws. Cultivate good habits. Seek good environments. Don't look for silver bullets.
(The anti-software argument will have to wait until someone says something that prompts it)
--
---
You received this message because you are subscribed to the Google Groups "Vancouver Rationalists" group.
To unsubscribe from this group and stop receiving emails from it, send an email to vancouver-rationa...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Kenneth
I would challenge your skepticism of cognitive enhancement in general, on the basis that despite the expectation, it's only recently that we've been able to directly observe where intelligence differences may come from at the level of cortex and below; with manipulation ability lagging behind but otherwise increasing as well. Filtered through criteria like Nick Bostrom's Evolutionary Optimality Challenges, and we could find exploits that can at least normalize most humans towards that nobility archetype of yours while not costing much time or otherwise.
A concrete example of what I mean is methylation cycle deficiency treatment. The methylation cycle is a particularly well studied pathway that has a collection of genes encoding for the conversion rates of e.g. choline, TMG, vitamin B12, vitamin B6 and so on, into methyl groups. Methyl groups are everywhere, so naturally if you have a SNP that does suboptimal conversion, you'll be affected in some way. Dr. Amy Yasko studied the link between methyl group deficiency, the methylation cycle and autism or depression. Out of such research came Deplin, a form of folic acid that is describable to such cases.
To unsubscribe from this group and stop receiving emails from it, send an email to vancouver-rationalists+unsub...@googlegroups.com.
Kenneth
As for software and training based cognitive enhancement - I'll go into it later, but I think the reason why there's a lot of lag in software based cognitive enhancement is due to the opacity of cognition itself, and the relatively difficulty of description of cognitive strategies. In order to get better at math I believe that I needed to become more aware of my own cognition in order to understand what's expected of me when confronted with a mathematical task. Once you get rid of most hurdles to thinking, you're left with the strategy of how to think, which I believe should be considered a part of intelligence amplification.
In terms of formally studied psychobiological considerations, tDCS combined with training is showing promise greater than either tDCS alone or training alone.
Wolf Tivy
Note that biomedical state is included in "phenotype", and that there's no way beyond nanosurgery or maybe really intense corrective training to correct a miswiring.
Drug-borne interventions to correct short-timescale chemical imbalance is totally plausible. Hence the effectiveness of things like ritalin. Note that these are neither general or complete. Can correct some of the worst abuses, can't correct all. Some are cast in from day 1, like damage to the genes controlling visual memory, for example.
At the point where we have a mechanistic theory of intelligence and intervention capability to the point where we can fix and upgrade live brains, we have other problems. That's not either now or soon, though.
Why is tdcs likely to work? What's the mechanism? Or is it purely empirical? Do we have any guesses? It seems to me a bruteforce intervention like that, unless guided by understanding, is very unlikely to improve the function of a complex and highly tuned machine like a brain. You can't do that to a modern internal combustion engine, for example, and it's substantially less complex.
Again, I'm somewhat open to possibility here, but skeptical until math coaches are teaching out of an intelligence enhancement textbook.
To unsubscribe from this group and stop receiving emails from it, send an email to vancouver-rationa...@googlegroups.com.
Kenneth Bruskiewicz
Here we have options: There's mitochondria efficiency, power, and quantity. I know I get looks for wanting to fire my brain up with lasers, but like tDCS it's been a hidden gem of a treatment paradigm since the 1960s (see http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3288797/ for some synthesis on what we know). Also like tDCS, we don't necessary have a complete understanding of how it works despite seeing results. Yet it can do wonders from people with traumatic brain injury while not dealing with the messiness of chemical implements. It even upregulates the presence of certain brain transcription factors, which can screen off short-length SNPs even if you're genetically unlucky.
Biomedical research is diseased in a lot of ways, and I would say to a greater degree than other parts of academia as we've come to know it. It's much more driven by financial incentives than other fields, while also having a sizable share of complexity and risk. You have weird idiosyncrasies like tDCS being a dead simple, forty year old technology that suddenly gains popularity now, for whatever reason.
So if even I'm a bit of a crackpot, I don't think it's beyond the realm of plausibility that the above plan would work raise IQ by five or so points (I'm trying to be conservative). But I wouldn't except research to follow the paradigm in step just because the facts have been laying around for awhile; it still takes someone to put the pieces of the puzzle together.
> tDCS
A better metaphor for the brain might be a highly-optimized and purposeful ecology, as it's diverse and produces non-linear effects. Parts of the brain can interoperate with other parts while not being dependent on other parts. It's flexible to many changes in its operating parameters, even if it can catastrophically fail with others. Still painting broad strokes, but I think this is more correct.
> teaching out of an intelligence enhancement textbook