Dear Hector,

Personally I agree that algorithmic information theory and the related concepts 
of randomness and Bennett’s logical depth are the best way to go. I have used 
them in many of my own works. When I met Chaitin a few years back we talked 
mostly about how unrewarding and controversial our work on information theory 
has been. When I did an article on information for the Stanford Encyclopaedia 
of Philosophy it was rejected in part becausewe of fierce divisions between 
supporters of Chaitin and supporters of Kolmogorov!  The stuff I put in on 
Spencer Brown was criticized because “he was some sort of Buddhist, wasn’t he?” 
It sounds like you have run into similar problems.

That is why I suggested a realignment of what this group should be aiming for. 
I think the end result would justify our thinking, and your work certainly 
furthers it. But it does need to be worked out. Personally, I don’t have the 
patience for it.

John Collier
Emeritus Professor and Senior Research Associate
Philosophy, University of KwaZulu-Natal

From: Hector Zenil []
Sent: Thursday, 30 March 2017 10:48 AM
To: John Collier <>; fis <>
Subject: Re: [Fis] Causation is transfer of information

Dear John et al. Some comments below:
On Thu, Mar 30, 2017 at 9:47 AM, John Collier 
<<>> wrote:
I think we should try to categorize and relate information concepts rather than 
trying to decide which is the “right one”. I have tried to do this by looking 
at various uses of information in science, and argue that the main uses show 
progressive containment: Kinds of Information in Scientific 
Use<>. 2011. 
cognition, communication, co-operation. Vol 9, No 

There are various mathematical formulations of information as well, and I think 
the same strategy is required here. Sometimes they are equivalent, sometimes 
close to equivalent, and sometimes quite different in form and motivation. Work 
on the foundations of information science needs to make these relations clear. 
A few years back (more than a decade) a mathematician on a list (newsgroup) 
argued that there were dozens of different mathematical definitions of 
information. I thought this was a bit excessive, and argued with him about 
convergences, but he was right that they were mathematically different. We need 
to look at information theory structures and their models to see where they are 
equivalent and where (and if) they overlap. Different mathematical forms can 
have models in common, sometimes all of them.

The agreement among professional mathematicians is that the correct definition 
of randomness as opposed to information is the Martin Loef definition for the 
infinite asymptotic case, and Kolmogorov-Chaitin for the finite case. 
Algorithmic probability (Solomonoff, Levin) is the theory of optimal induction 
and thus provides a formal universal meaning to the value of information. Then 
the general agreement is also that Bennett's logical depth separates the 
concept of randomness from information structure. No much controversy in in 
there on the nature of classical information as algorithmic information. Notice 
that 'algorithmic information' is not just one more definiton of information, 
IS the definition of mathematical information (again, by way of defining 
algorithmic randomness). So adding 'algorithmic' to information is not to talk 
about a special case that can then be ignored by philosophy of information.

All the above builds on (and well beyond) Shannon Entropy, which is not even 
very properly discussed in philosophy of information beyond its most basic 
definition (we rarely, if ever, see discussions around mutual information, 
conditional information, Judea Pearl's interventionist approach and 
counterfactuals, etc), let alone anything of the more advanced areas mentioned 
above, or a discussion on the now well established area of quantum information 
that is also comletely ignored.

This is like trying to do philosophy of cosmology discussing Gamow and Hubble 
but ignoring relativity, or trying to do philosophy of language today 
discussing Locke and Hume but not Chomsky, or doing philosophy of mind 
discussing the findings of Ramon y Cajal and claiming that his theories are not 
enough to explain the brain. It is some sort of strawman fallacy contructing an 
opponent living in the 40s to claim in 2017 that it fails at explaining 
everything about information. Shannon Entropy is a counting-symbol function, 
with interesting applications, Shannon himself knew it. It makes no sense to 
expect a counting-symbol function to tell anything interesting about 
information after 60 years. I refer again to my Entropy deceiving paper:
I do not blame philosophers on this one, phycisists seem to assign Shannon 
Entropy some mystical power, this is why I wrote a paper proving how it cannot 
be used in graph complexity as some phycists have recently suggested (e.g. 
Bianconi via Barabasi). But this is the kind of discussion that we should have 
having, telling phycisists not to go back to the 40s when it comes to 
characterizing new objects. If Shannon Entropy fails at characterizing 
sequences it will not work for other objects (graphs!).

I think the field of philosophy of information cannot get serious until serious 
discussion on topics above starts to take place. Right now the field is small 
and carried out by a few mathematicians and phycisists. Philosophers are left 
behind because they are choosing to ignore all the theory developed in the last 
50 to 60 years. I hope this is taken constructively. I think we philosophers 
need to step up, if we are not be leading the discussion at least we should not 
be 50 or 60 years behind. I have tried to to close that gap but usually I also 
get convenently ignored =)

I have argued that information originates in symmetry breaking (making a 
difference, if you like, but I see it as a dynamic process rather than merely 
as a representation) Information Originates in Symmetry 
Breaking<> (Symmetry 1996).

Very nice paper. I agree on symmetry breaking, I have similar ideas:
(published in the journal of Natural Computing)
On how symmetric rules can produce assymetric information.


Hector Zenil

I adopt what I call dynamical realism, that anything that is real is either 
dynamical or interpretable in dynamical terms. Not everyone will agree.

John Collier
Emeritus Professor and Senior Research Associate
Philosophy, University of KwaZulu-Natal

From: Guy A Hoelzer [<>]
Sent: Wednesday, 29 March 2017 1:44 AM
To: Sungchul Ji <<>>; 
Terry Deacon <<>>; John Collier 
<<>>; Foundations of Information 
Science Information Science <<>>

Subject: Re: [Fis] Causation is transfer of information

Greetings all,

It seems that the indigestion from competing definitions of ‘information’ is 
hard to resolve, and I agree with Terry and others that a broad definition is 
preferable.  I also think it is not a problem to allow multiple definitions 
that can be operationally adopted in appropriate contexts.  In some respects, 
apparently competing definitions are actually reinforcing.  For example, I 
prefer to use ‘information’ to describe any difference (a distinction or 
contrast), and it is also true that a subset of all differences are ones that 
‘make a difference’ to an observer.  When we restrict ‘information’ to 
differences that make a difference it becomes inherently subjective.  That is 
certainly not a problem if you are interested in subjectivity, but it would 
eliminate the rationality of studying objective ‘information’, which I think 
holds great promise for understanding dynamical systems.  I don’t see any 
conflict between ‘information’ as negentropy and ‘information’ as a basis for 
decision making.  On the other hand, semantics and semiotics involve the 
attachment of meaning to information, which strikes me as a separate and 
complementary idea.  Therefore, I think it is important to sustain this 
distinction explicitly in what we write.  Maybe there is a context in which 
‘information’ and ‘meaning’ are so intertwined that they cannot be isolated, 
but I can’t think of one.  I’m sure there are plenty of contexts in which the 
important thing is ‘meaning’, and where the (more general, IMHO) term 
‘information’ is used instead.  I think it is fair to say that you can have 
information without meaning, but you can’t have meaning without information.  
Can anybody think of a way in which it might be misleading if this distinction 
was generally accepted?



On Mar 28, 2017, at 3:26 PM, Sungchul Ji 
<<>> wrote:

Hi Fisers,

I agree with Terry that "information" has three irreducible aspects --- amount, 
meaning, and value.  These somehow may be related to another triadic relation 
called the ITR as depicted below, although I don't know the exact rule of 
mapping between the two triads.  Perhaps, 'amount' = f, 'meaning' = g, and 
'value' = h ? .

                                      f                               g
               Object --------------->  Sign -------------->  Interpretant

Figure 1.  The Irreducible Triadic Relation (ITR) of seimosis (also called sign 
process or communication) first clearly articulated by Peirce to the best of my 
knowledge. Warning: Peirce often replaces Sign with Representamen and 
represents the whole triad, i.e., Figure 1 itself (although he did not use such 
a figure in his writings) as the Sign. Not distinguishing between these two 
very different uses of the same word "Sign" can lead to semiotic confusions.   
The three processes are defined as follows: f = sign production, g = sign 
interpretation, h = information flow (other ways of labeling the arrows are not 
excluded).   Each process or arrow reads "determines", "leads", "is presupposed 
by", etc., and the three arrows constitute a commutative triangle of category 
theory, i.e., f x g = h, meaning f followed by g ledes to the same result as h.

I started using  the so-called  ITR template, Figure 1,  about 5 years ago, and 
the main reason I am bringing it up here is to ask your critical opinion on my 
suggestion published in 2012 (Molecular Theory of the Living  Cell: Concepts, 
Molecular Mechanisms, and Biomedical Applications, Springer New York, p ~100 ?) 
that there are two kinds of causality -- (i) the energy-dependent causality 
(identified with Processes f and g in Figure 1) and (ii) the information (and 
hence code)-dependent causality (identified with Process h).  For convenience, 
I coined the term 'codality' to refer to the latter to contrast it with the 
traditional term causality.

I wonder if we can  view John's idea of the relation between 'information' and 
'cause' as being  an alternative way of expressing the same ideas as the 
"energy-dependent causality" or the "codality" defined in Figure 1.

All the best.


From: Fis <<>> 
on behalf of Terrence W. DEACON 
Sent: Tuesday, March 28, 2017 4:23:14 PM
To: John Collier
Cc: fis
Subject: Re: [Fis] Causation is transfer of information

Corrected typos (in case the intrinsic redundancy didn't compensate for these 
minor corruptions of the text):

 information-beqaring medium =  information-bearing medium

appliction = application

 conceptiont =  conception

On Tue, Mar 28, 2017 at 10:14 PM, Terrence W. DEACON 
<<>> wrote:
Dear FIS colleagues,

I agree with John Collier that we should not assume to restrict the concept of 
information to only one subset of its potential applications. But to work with 
this breadth of usage we need to recognize that 'information' can refer to 
intrinsic statistical properties of a physical medium, extrinsic referential 
properties of that medium (i.e. content), and the significance or use value of 
that content, depending on the context.  A problem arises when we demand that 
only one of these uses should be given legitimacy. As I have repeatedly 
suggested on this listserve, it will be a source of constant useless argument 
to make the assertion that someone is wrong in their understanding of 
information if they use it in one of these non-formal ways. But to fail to mark 
which conception of information is being considered, or worse, to use equivocal 
conceptions of the term in the same argument, will ultimately undermine our 
efforts to understand one another and develop a complete general theory of 

This nominalization of 'inform' has been in use for hundreds of years in legal 
and literary contexts, in all of these variant forms. But there has been a 
slowly increasing tendency to use it to refer to the information-beqaring 
medium itself, in substantial terms. This reached its greatest extreme with the 
restricted technical usage formalized by Claude Shannon. Remember, however, 
that this was only introduced a little over a half century ago. When one of his 
mentors (Hartley) initially introduced a logarithmic measure of signal capacity 
he called it 'intelligence' — as in the gathering of intelligence by a spy 
organization. So had Shannon chose to stay with that usage the confusions could 
have been worse (think about how confusing it would have been to talk about the 
entropy of intelligence). Even so, Shannon himself was to later caution against 
assuming that his use of the term 'information' applied beyond its technical 

So despite the precision and breadth of appliction that was achieved by setting 
aside the extrinsic relational features that characterize the more colloquial 
uses of the term, this does not mean that these other uses are in some sense 
non-scientific. And I am not alone in the belief that these non-intrinsic 
properties can also (eventually) be strictly formalized and thereby contribute 
insights to such technical fields as molecular biology and cognitive 

As a result I think that it is legitimate to argue that information (in the 
referential sense) is only in use among living forms, that an alert signal sent 
by the computer in an automobile engine is information (in both senses, 
depending on whether we include a human interpreter in the loop), or that 
information (in the intrinsic sense of a medium property) is lost within a 
black hole or that it can be used  to provide a more precise conceptiont of 
physical cause (as in Collier's sense). These different uses aren't unrelated 
to each other. They are just asymmetrically dependent on one another, such that 
medium-intrinsic properties can be investigated without considering referential 
properties, but not vice versa.

It's time we move beyond terminological chauvenism so that we can further our 
dialogue about the entire domain in which the concept of information is 
important. To succeed at this, we only need to be clear about which conception 
of information we are using in any given context.

— Terry

On Tue, Mar 28, 2017 at 8:32 PM, John Collier 
<<>> wrote:
I wrote a paper some time ago arguing that causal processes are the transfer of 
information. Therefore I think that physical processes can and do convey 
information. Cause can be dispensed with.

  *   There is a copy at Causation is the Transfer of 
 In Howard Sankey (ed) Causation, Natural Laws and Explanation (Dordrecht: 
Kluwer, 1999)

Information is a very powerful concept. It is a shame to restrict oneself to 
only a part of its possible applications.

John Collier
Emeritus Professor and Senior Research Associate
Philosophy, University of KwaZulu-Natal<>

Fis mailing list<><>

Professor Terrence W. Deacon
University of California, Berkeley

Professor Terrence W. Deacon
University of California, Berkeley
Fis mailing list<>

Fis mailing list<>

Fis mailing list

Reply via email to