[THS] Brain delusions

[Peter Webster]

http://www.wired.com/medtech/drugs/magazine/16-04/ff_kurzweil_sb

Many computer scientists take it on faith that one day machines will
become conscious. Led by futurist Ray Kurzweil, proponents of the so-
called strong-AI school believe that a sufficient number of digitally
simulated neurons, running at a high enough speed, can awaken into
awareness. Once computing speed reaches 1016 operations per second
— roughly by 2020 — the trick will be simply to come up with an
algorithm for the mind. When we find it, machines will become self-
aware, with unpredictable consequences. This event is known as the
singularity.

These techno-utopians should pay closer attention to developments in
neuroscience. Sure, artificial intelligence techniques like neural networks
have led to better spam filters. But research suggests that the current
approach to AI won't result in a conscious machine on anything like
Kurzweil's timeline. The latest evidence shows that, when it comes to
consciousness, the brain simply doesn't work the way computer
scientists think it does. Almost nothing is known about how the brain
produces awareness, and current models of brain function don't accord
with the little that is known.

Singulatarians would respond by predicting that exponentially growing
scientific progress will fill the gap. This notion sweeps under the rug a
messy philosophical problem: An algorithm is only a set of instructions,
and even the most sophisticated machine executing the most elaborate
instructions is still an unconscious automaton. Philosophy aside, a
constellation of recent scientific findings indicates that no matter how
fast CPUs become in future decades, they'll be no more aware than a
toaster. Building a conscious machine will likely require paradigm shifts
in brain science — conceptual leaps that, by definition, won't come on a
schedule. Here, then, are five reasons why the singularity is not near.

The mind is synchronized, but no one knows how. New York University
neurologist E. Roy John has established that the hallmark of
consciousness is a regular electrical oscillation, or gamma wave, readily
detected by electrodes attached to the scalp. More recently, Wolf Singer
and his colleagues at the Max Planck Institute for Brain Research in
Frankfurt, Germany, confirmed that brain cells flicker in time with the
gamma wave. This flickering takes place among widely dispersed
neurons throughout the brain with no apparent spatial pattern. What
keeps these ever-shifting, widely distributed groups of cells in sync?
Neurochemical reactions take place too slowly to explain the
phenomenon. This mystery alone seems to demand a wholesale
rethinking of AI's underpinnings.

Current brain maps are of little use in explaining awareness. For more
than a century, the brain cell, or neuron, has been seen as a tiny
switching station with multiple signals coming in through many input
wires, known as dendrites, but only one signal going out through a
single output wire, or axon. AI is based on this circuitry model. When it
comes to consciousness, though, the model has its wires crossed.
Singer has discovered that gamma waves — the indicators of
consciousness — issue from the neuron's supposed inputs, not its
output. Confusing matters further, researchers, including Takaichi
Fukuda and Toshio Kosaka of Japan's Kyushu University, have revealed
that many inputs interconnect, forming an altogether different set of
networks. In other words, the vast strides made by neuroscientists in
their attempt to map the brain may reveal little about consciousness.

The brain is faster than singularity theorists think. AI assumes that the
neuron is analogous to a single computer bit. But it turns out that each
neuron is supported by a supercomputer's worth of additional circuitry.
MIT bioengineer Andreas Mershin and UCLA psychologist Nancy Woolf
have independently confirmed the importance of microtubules, the
scaffolding that undergirds each neuron, in animal memory and
learning. At the University of Alberta, physicist Jack Tuszynski has
developed computational models suggesting that these supposedly
dumb structures could be smarter than previously recognized. Stuart
Hameroff at the University of Arizona argues that trillions of
computations per second take place in the microtubules of each
neuron. If he's right, the brain's speed is 1028 operations per second —
a trillion times faster than is generally thought — which pushes the
vaunted singularity back by decades.

The on/off switch isn't where it's supposed to be. As it happens, doctors
have a handy way to flick the switch of consciousness: anesthesia.
When you're under, awareness is disabled, but everything else in the
brain operates normally. So how does anesthesia work? Hameroff has
come up with a simple model in which anesthetic drugs interact almost
exclusively with microtubules; the rest of the neuron plays only a
marginal role. This model is the closest anyone has come to a unified
theory of anesthesia — yet it flatly contradicts the notion that
consciousness arises from firing neurons.

Understanding consciousness may require new physics. In his 1989
book, The Emperor's New Mind, Oxford physicist Roger Penrose
proposed that the classical physics ruling neurobiology can't explain
consciousness. The mind, he declared, relies on the baffling mechanics
of quantum physics. Although his point remains controversial, evidence
in its favor is accumulating. Most recently, physicist Efstratios
Manousakis at Florida State University showed that certain confounding
quirks of visual perception are most easily explained by quantum
mechanics. If consciousness is indeed a quantum phenomenon, then AI
becomes an entirely new game. The singularity will have to wait for
engineers to catch up.