Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

Giulio Tononi's "Integrated Information Theory" of Consciousness

2 Post author: DataPacRat 07 October 2012 02:59AM

My daily browsing has come across an idea I haven't seen before, though it has been mentioned on LW occasionally. Created by Giulio Tononi, the basic premise seems to be that consciousness can be quantified, by measuring how much information is contained within the overall patterns of a system in excess of the information contained within its subsystems - that the whole is greater than the sum of its parts.

Some further browsing has let me find a somewhat uninformative Wikipedia article, a printed book using a fictional narrative, a scientific paper, and a 'Provisional Manifesto'.

The general idea doesn't seem to fall prey to most of the more obvious flaws that theories about consciousness tend to end up suffering. But I'm far from an expert in the field, or potentially related fields such as considering Φ's possible use as a basis for developing AGI/FAI. So: what does the wisdom of the crowd of LW have to say about this concept?

Comments (6)

Comment author: Manfred 07 October 2012 04:44:20AM 3 points [-]

Instant pattern-match: Dreams of AI Design

But this is probably doing it a little too little justice, and also perhaps underestimating the effects of hyping. Maybe it's just a plan for measuring the amount of information stored specifically in artificial neural networks that got rebranded into measuring "consciousness".

Comment author: Oscar_Cunningham 07 October 2012 07:38:33AM 6 points [-]

It doesn't make consciousness any less confusing i.e. in the same way that the sequences make free will less confusing.

Comment author: torekp 08 October 2012 12:50:46AM 2 points [-]

The Provisional Manifesto was intriguing. I view it as part scientific hypothesis (roughly "integrated information will correspond to the states we are independently inclined to call conscious, and predict some relationships among phenomenal features") and part proposal ("so let's use consciousness to refer to integrated information").

Comment author: BlazeOrangeDeer 08 October 2012 07:13:11PM 1 point [-]

At first glance it seems his definition of consciousness works better as a definition of some subset of intelligence, and not the common concept of consciousness. Usually "consciousness" implies self-awareness more than straight information processing.

Comment author: Angela 16 January 2014 11:20:18PM *  0 points [-]

There is a paper on both IIT and causal density here:

Comment author: capybaralet 27 January 2015 06:32:31AM *  0 points [-]

One potential issue with the theory in practice is that it requires calculating the integration (or whatever it is called) for EVERY sub-system and super-system of a given system in order to determine if it is conscious.