Following my previous blog post on Understanding Ourselves with Virtual Machine Concepts, I exchanged e-mails with Aaron Sloman on virtual machines (VM’s) and mathematics. Google Books had served me part of a chapter he wrote on the Poplog VM, published in Artificial Intelligence & Software Engineering (a book edited by Derek Partridge.) I asked for a PDF, which Aaron couldn’t find. But he kindly shared some more other papers on VM’s.
I thought I would share the information for anyone who was interested in my previous post. Below, you’ll also find references to two papers on five senses of mechanism, and some thoughts about choosing knowledge-building projects.
Aaron suggested the said file might overlap with the following.
http://www.cs.bham.ac.uk/research/projects/cogaff/81-95.html#65
Filename: smith-gibson-sloman-1992.pdf (10MB OCR-PDF)
Title: POPLOG's Two-level Virtual Machine Support for Interactive Languages
Authors: Robert Smith, Aaron Sloman and John Gibson
Date published: 1992
Where published:
In Research Directions in Cognitive Science Volume 5: _Artificial Intelligence_, Eds. D. Sleeman and N. Bernsen, Lawrence Erlbaum Associates, pp. 203--231, 1992,
Abstract:
Poplog is a portable interactive AI development environment available on a range of operating systems and machines. It includes incremental compilers for Common Lisp, Pop-ll, Prolog and Standard ML, along with tools for adding new incremental compilers. All the languages share a common development environment and data structures can be shared between programs written in the different languages. The power and portability of Poplog depend on its two virtual machines, a high level virtual machine (PVM — tne Poplog Virtual Machine) serving as a target for compilers for interactive languages, and a low level virtual machine (PIM — the Poplog Implementation Machine) as a base for translation to machine code. A machine-independent and language-independent code generator translates from the PVM to the PIM, enormously simplifying both the task of producing a new compiler and porting to new machines.
This does in fact have a very helpful section on the VM: Section 2, pages 205-219. Poplog’s two-level virtual machine is a good example of how a system can have stacked virtual machines. Poplog also provides different routes for compiling to a virtual machine. While human brains do things different, they most certainly also will provide different ways of compiling from one virtual machine to another.
Aaron Sloman also pointed me to the following tutorial, which he referred to as imperfect but a good starting point: Web page: Virtual Machine Functionalism (VMF)
(The only form of functionalism worth taking seriously
in Philosophy of Mind and theories of Consciousness), and in PDF.
Causation
On a related note, I attended a very interesting Defining Cognitive Science talk at SFU Tuesday (March 6) on “Causal Metaphysics and Modeling”, by Prof. Holly Andersen. In 2014, she also published “A Field Guide to Mechanisms” in two parts. Part 2 is here.
As you can see from the second abstract, Prof. Andersen distinguished five senses of mechanism that have no overarching meaning. On skimming the papers (which is all I have done so far), a question in my mind was, are different senses of mechanisms required for characterizing VM’s? My sense was yes. However, articulating the intuition would be require more time. I’m interested in these questions (as you can gather from the next section), but I am not a philosopher.
Notes on intellectual compatibility: causal reasoning, CogAff, evolution, …
Here’s a little tangent on how one’s intellectual interests can lead one to meet people with whom one is intellectually compatible, and from whom one can develop. I intend to write about this more abstractly later, as it is a major component of meta-effectiveness. For now, I will just discuss some personal examples related to the topic of this post and CogZest.
The thesis problem I proposed (in 1989) for my D.Phil scholarships at Sussex University was causal reasoning. Aaron Sloman was the supervisor I requested. As an undergraduate, I had done a self-directed reading course contrasting and assessing Kantian and Humean (covariation) models of causal reasoning in humans. I concluded that our causal intuitions are not merely based on covariation. While Aaron accepted my research proposal, and is an expert on Kantian understanding of understanding, he also gave me the opportunity to work on his Cognition & Affect project. As an undergraduate, I had worked for two years as an RA on motivation, in a neuroscience lab —Studying motivational priming effects in rats self-stimulating their lateral hypothalamus; I had become disenchanted with the lack of theoretical rigour in neuroscience. I don’t think I had previously encountered information-processing approaches to motivation. After much reflection, I realized the Cognition and Affect was a unique opportunity to understand the kinds of “hot” phenomena that got me interested in psychology in the first place. So I accepted Aaron’s invitation and have never regretted it.
In 2008, Aaron wrapped up the CogAff project and suggested I resume, where I had left off soon after my Ph.D., on CogAff work. I’ve been chipping away at this (part-time, on and off) ever since. Aaron himself returned to his own initial interests: understanding how brains understand.
In the email exchange mentioned above, he wrote:
I’ve recently started trying to specify a non-discrete alternative to digital computer (as Super-Turing multi-membrane machine) that could support ancient mathematical reasoning, and reasoning by human toddlers, crows, squrrels, elephants etc.
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/super-turing-geom.html
That’s a tantalizing part of his very interesting meta-morphogenesis project. The meta-morphogenesis project is concerned with the evolution of cognition. In particular: how could evolution had produced brains that understand and know that they understand?
Coincidentally, early in my academic path I was concerned with the evolution of cognition. My undergraduate thesis paper was A computational investigation of the evolution of vision. I haven’t returned to that project. However, meta-effectiveness (which is the main problem addressed by Cognitive Productivity) can be read as a subset of meta-morphogenesis, concerned with deliberate ontogenesis, rather than evolution.
I mention Aaron Sloman a lot on this blog, because he’s deeply influenced my thinking. But as the foregoing notes illustrate, it’s not by coincidence that I wound up working with Aaron. I had scoured the Commonwealth for a prof to do my Ph.D. thesis with, and heeded the suggestion of Claude Lamontagne. I consider Aaron’s work to be some of the most important foundational work in cognitive science/AI.
If you’re curious, check out the Wikipedia article on him. Or better, read this collection of papers in honour of him: From Animals to Robots and Back: Reflections on Hard Problems in the Study of Cognition. Be sure to read Maggy Boden’s, “Aaron Sloman: A Bright Tile in AI’s Mosaic.” If you google that with “PDF” you will find the entire PDF on Springer’s website.
I intend to feature some of Aaron’s work in a future book, Great Contemporaries in Cognitive Science and Related Endeavours.
What research problems should one choose?
This raises a couple of related questions:
- What determines the research problems one chooses?
- How should one choose research problems?
Those are big questions that would call for long answers. I plan to offer some thoughts on them in a later blog post. For the time being, I will just recommend the latest Brain Science podcast interview of Seth Grant by Ginger Campbell:
He provides some answers to the second question based on his own experience. It’s worth listening to Seth. If I were to pick a scientist who, I predict, will wind up with a Nobel prize one day, it would be Seth Grant.
Now, I need to get back to writing a paper on somnolent information-processing theory, which is an offspring of the CogAff project.
Revisions
2018-03-13. Added some text about Poplog’s two-level VM’s, which was the initial rationale for this and the previous blog post. Reorganized the text in the “notes on intellectual compatibility” to make the connection more explicit between Aaron’s meta-morphogenesis project and my earlier work: both have to do with evolution of cognition. Thanks to Carol Woodworth for noting the link wasn’t clear. Added section “What Research Problems to Choose?”
