Let's Read: some mind-bending papers from Anil Seth

Anil Seth studies the neuroscience of consciousness. Here's a presentation where he talked about it for 50 minutes. 

Basic points from the talk:

• Chalmers (1995) proposed two problems: Easy: explain of how a human brain works/functions/does its job. Hard: why consciousness exists/why do we feel.
• Anil Seth says: the hard problem is too philosophical! The real problem is what physical systems are conscious and what are not, aka giving a Physical Correlate of Consciousness.
• Seth talked about three aspects of consciousness: level, content, self.
• Level: deep sleep < drowsy < fully awake.
• Wakefulness is not consciousness. Vegetatives are awake but unconscious.
• 
• Cerebellum is not conscious despite having lots of neurons, and having no cerebellum [complete primary cerebellar agenesis] is actually fine for consciousness.
• One possible way to measure consciousness is by transcranial magnetic stimulation (TMS). If you TMS a conscious brain at a point, the excitation spreads in a complex way. Unconscious brain, simple way. "Simple" and "complex" were measured by information complexity, approximated by Lempel-Ziv compression algorithm (make a brain recording, then turn it into .gif with maximal compression, and see how big the result is!)
• Content:
• Bayesian Brain Theory: The brain uses Bayes probability to build a model of what's going on outside, updated by the sensory inputs.
• There's a feedback loop between the brain's expectation of what it should see and the sensory data on what is actually seen. If the brain's expectation cannot be conveyed (for example, by TMS interference), conscious vision is lost.
• The alpha wave in the visual cortex seems to regulate this loop. During the peaks, the expectation has most impact, valleys, the sensory inputs. See (Sherman 2016).
• Self: body, perspective, volition, narrative, social. Anil Seth works mainly on the body self.
• The "body self" is the brain's best (Bayesian) explanation for the proprioceptive and interoceptive senses. Proprioceptive: where the limb and muscles are, what they are posing. Interoceptive: the concentration of chemicals in the blood, heart rate, how full the guts are, gut feelings.
• Emotions depend on external and interoceptive inputs. Seeing a bear is external input. Feeling heart rate go up and muscles tensing is interoceptive. Taking in these inputs, the brain figures out: "Oooh, I am feeling fear!" and fear enters the conscious mind. Bear does not cause fear. Bear causes unconscious reflexes that generate the bodily responses. The brain's interpretation causes fear.
• In short: emotion = brain's best guess of the cause of interoceptive inputs.
• This can explain why people tend to experience emotion as happening in a particular place in the body.
• Descartes said that humans are (beast machine + soul). But really, the soul is generated by the beast machine trying to do Bayes prediction on itself and its environment. I predict, therefore I am.

The drug-trip without drugs is published in Nature: A Deep-Dream Virtual Reality Platform for Studying Altered Perceptual Phenomenology (2017). A nice summary is here. Basically, using Deep Dream to change what you see in VR is already enough to make you have similar feelings as when on a shroom trip. This shows that the main effect of shroom on consciousness is through visual hallucination, rather than the other physiological effects like heart rate.

Bonus paper:

Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms (2018).

Superrecognizers are people who are (naturally?) great at recognizing faces. This research pitted them together in facial recognition tests. 

forensic examiners = superrecognizers

(human + algorithm) > (two humans), (a human), (algorithm)

It shows that humans and the algorithm are seeing something that the others don't, and that's how their combined efforts are better.

The next step should be to find out what the humans are seeing that the algorithm don't, and program the algorithm to see them. The end result should be

(human + algorithm) = (algorithm)

And to hell with human + algorithm cooperation! Algorithms rule!