neededalj (neededalj) wrote,

Tilting At Windmills: Why Ogi Ogas and Sai Gaddam Are Bad Scientists (Neuroscience edition, round 2)

[addendum added just before posting].....fuck, ya'll, what can I say. I was an inspired insomniac. I hope this makes sense.

In reply to the response my first essay received, I have decided to attempt a further deconstruction of the Fail Neuroscience of Drs. Ogas and Gaddam. In a different approach from my first post, I am going to take this one from the ground up and start with the basic building blocks of neuroscience. I'm trying to write for an educated layperson; if I'm not clear about something, please ask. I am also limited at this time from being outside the university system (I would dearly love to do a proper *academic* criticism with citations and everything, but that's out of reach right now).

Ok. The Brain.

The brain is composed of networks of neurons. Neurons are single cells in the nervous system that 'talk' to each other, in various ways. One way is through electrical impulses, and another is with chemical signals called neurotransmitters that travel between the cells. A change in one neuron will send a series of messages to the neurons that border it, and then they communicate to the neurons bordering *them* and so on and so forth. An active brain is one where the neurons are continuously changing in response to each other. If they aren't, you're dead. When doctors say that someone is 'brain dead' it means that the neurons have died and stopped communicating to each other.

There are BILLIONS of neurons in the human brain. Billions. They are arranged into differentiated areas and structures. Some of these are very clearly physically defined, like the hippocampus. Others blur together such that you can tell that the areas of different, but it's hard to draw sharp line between them.

Now it's time to take a step back and have a short interlude about Science. The holy grail in scientific research is a theory with predictive ability. That's what scientists are doing when they experiment, theorize, and experiment again. Anyone can make a theory that can be retrofitted onto data you already know. The key to a good theory is to have it predict what happens in the future. Remember this: predictive ability. It's the goal of Ogi and Sai's research. It's what they think they have figured out how to do, and I'll return to this point later in the discussion about models.

Human's can't create a synthetic brain right now. It's far to complex. Your brain as you read this essay is doing something that the most advanced supercomputer can't. Your brain is better and faster than the world's greatest artificial machine at anything that requires a complex response. (Want to know why voice recognition systems are so bad? Scientists initially thought voice recognition would be a piece of cake because language was just a series of discreet inputs. WRONG. Turns out the brain is actually parsing a near continuous nondifferentiated stream of sound. Oops).

Some neuroscientists start by looking at behavior, stimuli, and inputs and attempt to sort things out through imaging and lesions studies (looking at when a brain is damaged to see what cognitive deficits appear). This is more the kind of stuff I studied.

What some neuroscientists and computer scientists do is model brain behavior. Neural models involve using computers to simulate neurons and programming those neurons using *very* complicated math to respond in a certain way, and then applying a stimulus and see if your model makes sense/is doing what it's supposed to/actually resembles the human brain/is properly predicting things (take your pick or all of the above).

Taking a very simple example from vision (vision modeling being the field Ogi Ogas held up as 'more complicated than' subcortical structures and sexual behavior).

Let's say I have neuron A, which is a retinal neuron (a neuron in the eye) and responds to a certain frequency of light. Neuron A is connected to Neurons B and C which are connected to each other, and which are connected each of them to Neurons D and E, and neuron E is connected to neuron F which is also connected to neuron B.

Now I give the system an input; because this is a visual system, our input is light. And neuron A is the only neuron that actually is exposed to the light. But neuron A now fires a signal, which triggers neurons B and C, and they trigger D and E...but remember that E is connected to F which is connected to B. So now F is *also* signaling B. Many responses later down the line the brain recognizes that neuron A fired and you see the light (literally). But it's not a simple chain; it's an overlapping, interwoven network with communication and response flowing 'up' *and* 'down'.

Neural systems are non-linear in structure and response. See how complicated my little example got with just 6 neurons and an input with simple physical properties (light)? The computing power required to model complicated neural networks is intense; take my little example and multiple it thousand thousand of times to get what our most sophisticated modeling systems do now, and multiply it a billion billion times to get what the brain actually does. So modelers write their models and ask 'does this system respond the way the brain does to this stimuli? or 'does this system 'remember' this stimuli'? etc.

Everything following from this point is speculation. Educated speculation, yes, but please do not quote me as saying what Ogi Ogas and Sai Gaddam were actually doing because I DO NOT KNOW for sure.

As far as I can figure out--and I have to caveat this because I literally had to look away from many of their explanations in pure disgust--Ogi Ogas and Sai Gaddam want to do this kind of modeling. Only instead of neurons, they want to use *structures*. And instead of simply physical stimuli, their inputs are...sexual? erotic? pornographic?

My BEST GUESS (and it is at best a guess, because it's so appallingly wrong-headed I can barely bring myself to type this) is that their 'survey' was designed to gather either a)inputs to the system or b)data to program the system; ie, women find THIS sexy which will cause the amygdala to respond'.

Models require data from the beginning. You look at a real-worl example, you say 'Neuron A does this when it is exposed to green light, but not red light or blue light!' and you program that in. If their survey gathering was to get programmable data for the neural network itself I...don't have much to say. They have no way of linking responses to *actual* neural responses because as many people have pointed out, they can't image us through their computers. So they would be firmly in the 'making shit up' category.

If they are using the survey responses as *inputs*, that's a little different. In that case they must be building their model off of a thousand sources which I believe I have seen them list--primate studies, former imaging studies, physiological responses such as blushing or pupil dilation to stimuli (in other people's experiments), and god knows what else. So basically they think they can take complex, complicated real-world stimuli such as an NC-17 slashfic, and say that it calls up a specific set of neural responses (responses that would theoretically include both electrical responses and chemical neurotransmitter responses). And they think they can program a model which predicts that.

They keep harping on about subcortical structures, but scientists barely understand subcortical structures. Here are some of the things subcortical structures have been implicated in: language, movement, learning, emotions, sleep, attention, and yes, arousal. Not by themselves, of course; the cortex is involved in all of this as well.

I don't really want to go into this further, because it goes more and more into speculation. But you probably don't need my help to see where 'deviancy' would enter into this situation. Deviant inputs should elicit different responses than 'normal' inputs. I assume this is where slashfic and 'shemale' pornography would enter into the picture. But I can only guess. I suspect that their model will assume universals as deeply laughable as their survey was a methodological travesty.

They're utterly wrong, of course. Their privilege racism and sexism and homophobia have been detailed at great lengths elsewhere. We don't know much about the brain, relatively speaking, but we do know this: all human beings have one. And human beings are wonderfully diverse, unpredictable creatures. Our cultures are different, our tastes are different, we interact differently in social situations. We have different morals and different traditions, different tastes and different emotions.

And our science is not yet at the point where this can be modeled in any of our artificial technological creations (ETA: there is a very good comment detailing a computer science perspective on neural nets here).

Ogi Ogas and Sai Gaddam's 'research' is a tragic act of biological reductionism.

  • Post a new comment


    default userpic

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.
← Ctrl ← Alt
Ctrl → Alt →
← Ctrl ← Alt
Ctrl → Alt →