integrated information theory - uzh
TRANSCRIPT
What does IIT say?
Consciousness is a special type of ‘information’
Information can be calculated
But, what is ‘information’?
The birth of a theory
The classical approach:
Describing matter in enough detail will explain why it feels at all
The hard problem:
Abstractions only produce more abstractions
The birth of a theory
The classical approach:
Describe phenomena using the language of matter
The alternative?
Describe matter using the language of phenomena
The birth of a theory
Let’s talk about how consciousness feels …
and then think about what kind of physics would support it …
instead of the (classical) reverse approach
The birth of a theory
For a theory, we first need ‘axioms’
These are statements we hold to be true
If we make some useful axioms about consciousness,
…maybe we can derive everything else
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence - ‘‘I experience therefore I am’’.
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence
Composition - Left, right, red, blue, triangle, square,
and a red triangle on the left.
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence
Composition
Information - Each experience differs from other possible experiences. It is this because it is not everything else
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence
Composition
Information
Integration - An experience is more than the sum of its parts. A word is experienced as a whole, not as it’s parts.
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence
Composition
Information
Integration
Exclusion - Each experience excludes all others;
It has a spatiotemporal resolution.
The birth of IIT
What does Integrated-Information Theory have as its axioms?
Existence
Composition
Information
Integration
Exclusion
From truths to guesses
These axioms don’t offer anything for our mainstream concepts about matter and causality … yet
We need to derive ‘postulates’ from them to link them to a physicalist world-view
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence - Mechanisms exist. System = a set of mechanisms
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence
Composition - Mechanisms can be combined
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence
Composition
Information - A mechanism can contribute to consciousness ifit constrains past and future states of the system
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence
Composition
Information
Integration - A mechanism can contribute to consciousness ifit is irreducible to the info of its components
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence
Composition
Information
Integration
Exclusion - In a system, there is only one conscious set of mechanisms. Why? Analytical reasons.
The postulates of IIT
What must a physical system satisfy to generate experience?
Existence
Composition
Information
Integration
Exclusion
A short note on ontology
Okay, we’ve found a way to describe conscious physical systems using a language of phenomena
But, it has subtext: Consciousness is identical to physical properties
How is IIT used?
If I want to assess the consciousness of a system, what does IIT give me as an output?
How is IIT used?
If I want to assess the consciousness of a system, what does IIT give me as an output?
Answer: A graph, and a number.
Wait, what?
How is IIT used?
Example:
I take 3 logic gates from a computer and connect them together
Gates active for specific inputs:
AND = 1,1OR = 1,0 / 0,1 / 1,1
XOR = 1,0 / 0,1
How is IIT used?
IIT then breaks this toy system into its components
and stimulates them to define
the cause-effect structure of
the system as a whole
How is IIT used?
Cause-effect structure?
This is a statistical description of the system
It uses high-dimensional statistics to answer:
Given the current state, how certain can I be about the states which led to this moment, and the states which are going to happen?
How is IIT used?
What does this mean for the system’s conscious experience?
If the cause-effect structure of the system is more thanthe sum of its parts, and constrains itself
dynamically, it exists from its own intrinsic perspective
The cause-effect structure, according to IIT, is identical to the
experience being had by that system
How is IIT used?
The cause-effect structure, according to IIT, is identical to the
experience being had by that system
Which ‘qualia’ are present, and ‘how much’ do they exist?
How is IIT used?
But there’s more:
We can now look for the constellation of qualia which is maximally irreducible – a set of qualia which are the most real from the system’s own perspective.
How is IIT used?
But there’s more:
We can now look for the constellation of qualia which is maximally irreducible – a set of qualia which are the most real from the system’s own perspective.
The value of that maximum irreducibility is called Phi (Φ)
Φ tells us how conscious the system is
How do we calculate Φ?
In this lecture, we won’t
But here’s where you can try for yourself:
Demo: http://integratedinformationtheory.org/calculate.html
And I’ll put extra slides on the course website:
What do you think of IIT?
Does this theory sound useful to you?
Does it tackle the ‘hard problem’ of consciousness?
What does IIT predict?
IIT has some interesting consequences
- Consciousness is lost due to a loss of integrated-information
- Feedforward neural networks have no conscious experiences (e.g. computer vision, and the cerebellum)
- Some non-biological devices have small Φ values (e.g. photodiodes)
- Some systems potentially contain multiple conscious entities (e.g. split-brain patients)
Problems with IIT
IIT has many supporters, but also several critics
What is wanted is a new way to think about consciousness in research
IIT is agreed to be a great starting point …
but by no means a satisfying solution
Is IIT practically useful?
We cannot manipulate a whole brain to calculate Φ(neither physically nor computationally)
But we can use the theory to make ‘approximations’
Bedside diagnosis
What do you do when you have a patient in hospital who is unresponsive to stimuli, and yet may be conscious?
“Is it locked-in syndrome?”
“They took too much ketamine at a party?”
“Is the patient successfully anaesthetised?”
Bedside diagnosis
The Perturbational Complexity Index (PCI)
Inspired by the ‘Information’ and ‘Integration’ postulates, a measurement procedure was designed
Time (milliseconds)
Time (milliseconds)
Time (milliseconds)
Sarasso et al., 2014
Bedside diagnosis
The Perturbational Complexity Index (PCI)
A combination of Transcranial Magnetic Stimulation (TMS), Electroencephalography (EEG), and classical Information Theory
What does this all mean?
Integrated-Information Theory is currently the most clearly defined theory of consciousness
It is currently called IIT 3.0 because it is an evolving concept
It is creating fresh debates, novel ideas, meaningful experiments, and extra lectures to teach.
Final thoughts: Is describing the same as understanding?When Louis Armstrong was asked
“What is Jazz?”
He replied
“Man, if you gotta ask, you'll never know.“
We could say the same about attempts to understand consciousness
Further reading
∞ Oizumi, Albantakis, Tononi, 2014 - From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0
∞ Tononi et al., 2016 - Integrated information theory: from consciousness to its physical substrate
∞ Integrated-Information Calculator - integratedinformationtheory.org/calculate.html
∞ Internet Encyclopedia of Philosophy on IIT - iep.utm.edu/int-info/
∞ Herzog et al., 2019 - The unfolding argument: Why IIT and other causal structure theories cannot explain consciousness
∞ Mindt, 2016 - The Problem with the ‘Information’ in Integrated Information Theory
∞ Casali et al., 2013 - A Theoretically Based Index of Consciousness Independent of Sensory Processing and Behavior
∞ Sarasso et al., 2014 - Quantifying Cortical EEG Responses to TMS in (Un)consciousness