Natural Morphological Computation as Foundation of Learning to Learn in Humans, Other Living Organisms, and Intelligent Machines

Round 1

Reviewer 1 Report

- The paper deals with the relevance of the info-computational approach to cognitive sciences, developed by the Author in other publications quoted among the  paper references, for a renewed epistemology of the Philosophy of Nature, emphasizing that it can provide an essential contribution to the better understanding of the learning process, both in artificial communication systems - AI systems before all - and in natural communication systems like animals and humans. As to the humans, it can solve the issue about the previous process of "learning to learn", essential for connecting the pre-symbolic (pre-conscious) with the symbolic (conscious) information processing in humans. In this sense the paper gives a relevant contribution to cognitive sciences and then to a naturalistic approach to the epistemology of the Philosophy of Nature. - The originality of the approach is therefore granted, because nobody before used the info-computational approach to solve the "learning to learn" issue in cognitive neuroscience. - The paper appears well-written and easy to read. It requires only a minor revision in formatting the bibliography because of the mistake in aligning the second reference. - Also the conclusions are well-written and convincing about the main thesis of the paper. The only remark is that are probably too synthetic in summarizing the steps of the argumentation. Therefore, if the Author wants to profit by the minor revision I suggested for solving the formatting problem, she could also refine and enlarge the conclusions paragraph, in order to make the paper more understandable. Generally indeed, people read before the abstract, the introduction, and the conclusion of a paper for deciding to read the rest of it. Therefore, the more care we give to these parts of a paper, the more understandable and profitable we make it, for its better diffusion.

Please control the formatting of ref. 3 lines 315-16

Author Response

I would like to thank the Reviewer for very constructive and helpful comments.

I have corrected errors in the references, improved the abstract and have re-written the conclusions - as suggested.

 

Gordana D-C

Reviewer 2 Report

This paper claims to argue that "natural morphological computation" is the basis for learning to learn in natural and artificial systems.  But the computational resources required for learning to learn are never actually discussed.  The paper points to various parts of the current literature on biological computation, but doesn't pull it together around the question of learning.  Indeed generalization is left as an open question (ln. 266).  To be acceptable, the paper needs to establish the claim made in the title as something more than trivially true.  It needs to say how morphology provides a basis for learning to learn.  This is often discussed under the rubric of "evolvability"; see e.g. Watson, RA & Szathmáry, E 2016, “Can evolution learn?” Trends in Ecology and Evolution, vol. 31, 147-157.

Some additional specific comments:

ln. 73: Explain briefly how an agent represents "its own behalf". See e.g. Froese, T.; Ziemke, T. Enactive artificial intelligence: Investigating the systemic organization of life and mind. Artificial Intelligence 2009, 173, 466–500.

ln. 86-87: This is a stretch. Why would a limbic system be a necessary condition for an attractive/aversive distinction? Even bacteria appear to experience the latter. See e.g. Lyon, P. The cognitive cell: bacterial behavior reconsidered. Frontiers in Microbiology 2015, 6, 264.

ln. 97-106: Bengio's position is basically that of Descartes and GOFAI, which should be noted. There are many who would take consciousness, in the form of awareness of changing stimuli, down to bacteria. How exactly is the word "consciousness" being used here? What abilities are being assumed as "parts of the package"?

ln. 116: Here and elsewhere, please list references in numerical order.

ln. 119: What about gene regulation?

ln. 134: What does "explicit" mean here? Metacognitive?

ln. 139-140: "too small" for what? Why "too short"? Bacteria are effectively immortal.

ln. 146: even in the case of macromolecules. See e.g. Fields, C.; Levin, M. Multiscale memory and bioelectric error correction in the cytoplasm-cytoskeleton-membrane system. WIRES Systems Biology and Medicine 2018, 10, e1410.

ln. 201-202: Many would argue that this is too broad. See e.g. Horsman et al. When does a physical system compute? Proc. Royal Society A 2014, 470, 20140182.

ln. 224: Biologists tend to use H. Simon's older term "satisficing" and recognize that all natural information processing is "merely" satisficing. See https://en.wikipedia.org/wiki/Satisficing

ln. 252-260: This needs to be up around ln. 73.

ln. 263: [64] is hardly an adequate reference for the physics of information movement. Cite some physicists. See e.g. Mermin, arxiv:1809.01639 or Muller, arxiv:1712.01826v5

ln. 266: I thought this paper was going to be *about* generalization and causality. Does the framework presented here have nothing more to say about these issues?

ln. 302: What about developmental robotics?

There are some incomplete references, e.g. 6, 10, 11, 43, 44, 48 (I may have missed some).

This paper seems incomplete.  I expect this author has something of significance to say on this topic, so I am recommending a revision to provide an opportunity to say it.

 

Author Response

I would first like to thank the reviewer for very careful reading and numerous valuable suggestions of references and improvements.

My detailed answers can be found in the file attached.

 

Gordana D-C

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The paper is much improved.  Some minor errors need correcting:

ln. 64-66: "It is not the development into the unknown, as some of it on the System 2 side, has earlier been proposed by GOFAI, and it is addressed in the new developments in cognitive science and neuroscience." I don't understand this sentence. What is it saying?

Punctuation needs to be checked for errors throughout, e.g. ln. 82 "computation,. [17–19]."

References [3] and [115] appear to be duplicates.

References [7] and [36] appear to be duplicates.

Remove caps in ref [84].