(soon)
Macrocognition bead thread -> .
4/17/2004 7:46 PM
Leo Obrst
Mitre
There is a position that abduction and induction can be simulated by computational processes acting on data structures if by simulation one means something specific, reserving and recognizing the difference between something simulated and something natural and physical. Robert Rosen and Peter Kugler are two of the many scholars who have addressed the essence of this difference.
Fully understanding the difference would require a perfect understanding of nature, which may never be achieved; as well as a perfect understanding of computer science – which may in fact soon be achieved.
Note: this hypothesis of an “end to computer science” is subject to a
narrow definition of computer science, not a stratified computer science that
admits to a multiplicity of foundational logics, as suggested by Peter Krieg.
Our position about abduction and induction follows the communication to the BCNGroup made by Peter Krieg – specifically in regards to his three systems:
·
natural physical (including biological) systems,
·
engineered physical systems and
· theoretical systems
A few precise points can be made, and discussed, about Peter’s viewpoint. Physical and biological systems are treated together, which is exactly how I treat them. The material-physical (non-living) system is stratified in its nature and exists in time. This is not true for abstractions. Living systems can exhibit intention and even perception and cognition, so one might think that living systems and non-living systems are quite different. But various scholars, Robert Shaw in particular (personal communication), have worked on a theory that an elementary particle approaching the event horizon of a black hole and the event horizon MUST mutually anticipate the future. In other words, material, non-living physical reality is anticipatory at root. Understanding living systems deeply cannot occur without really understanding the essence of physical reality. Shaw’s theory is not so precise to make this theory real in a practical sense, but it seemed to be a good start.
Beginning in a 1999 discussion between myself and Robert and myself and Nan Gelhard, a series of discussions lead us to invent the term “Anticipatory Web”. This notion was developed in the context of e-commerce system design that “anticipates” the client’s needs and arranges an electronic storefront. One of the needs was for a very small engine and data management system that could be constructed while and individual was visiting a web site, and then release abstract (not personal) structural information about the social discourse in real time to an aggregating agency.
In our prototypes, the arrangement of electronic store fronts are being based on “computational anticipatory responses” made using purely computational processes interacting with social reality (web harvest of social discourse) and with the individual (profile generation based on observation of individual response behaviors.)
Anticipation is essential to my notion of an Anticipatory Web of Information. How anticipation is grounded in the so called tri-level architecture for computational intelligence can be suggested. Note this tri-level architecture for computational intelligence is quite different from the tri-level architecture proposed by David Marr and by Z. W. Pylyshyn.
As I will illustrate, the difference between my work and almost all other work is an ontological commitment to organizational stratification. In my work the three levels are physical organizational strata where time differentiates the levels in two ways:
·
By
time scale, faster events are separated from slower events by an empirically
observed gap where no events having an intermediate duration can be observed
(this is somewhat like the Plank constant in the definition of electron shells
– where the existence of an electron is confined to shells whose distance from
the atom is an integer multiple of Plank’s constant.)
·
By
past and future being separated by the present.
The point is that the greater difference exists between natural systems and theoretical systems (i.e. formalism such as mathematics) than between non-living systems and living systems. We can set this discussion aside as a given, I would hope.
The key to my present communication is Krieg’s engineered physical systems.
The reason why, is that I wish to suggest a specific relationship between the following four categories
( memory, awareness, anticipation )
( induction, deduction, abduction )
( past, present, future )
(fast time scale, “our scale of observation”, slow ecological time scale)
Stratification allows something to be said, and something to be done that is useful and perhaps might be clear to many scholars, regarding the replacement of XML and relational (Codd-type) databases with a hash table encoding of ordered triples of the form
< a, r, b >
where a and b are “locations” and r is a relational variable. Following the applied Russian semiotics community’s usage, we call these ordered triples “syntagmatic units”. Aggregations of syntagmatic units are represented as simple graph constructions, such as the Orb constructions.
An ongoing process can encode “all” of the results of measurement into the abstract form of the syntagmatic unit. These units are then encoded with a hash table, or the Orb “key-less” hash table, in order to build a “memory” of the past. This memory, like human physical memory stores – the substrate form which memory is re-membered – has not global organization. The organization is instantiated in real time in response to affordance structure.
The process of induction is then an assembly of elements from memory AND a stepping into the present with this assembly. The stepping into the present produces a specific logic from the ground up. In biological system this logic will be minimalized in complexity so that there is an efficacy in how it is obtained and how effective it is when measured by evolutionary pressures.
In Russian quasi-axiomatic theory (QAT), the intent (of V. Finn and D. Pospelov and others) was that the facts and the inference rules are all made up, or inducted, into a single simplest control system. But in natural intelligence this induction does not occur by itself. It is meet with abduction from the environment. QAT could not address this (Peter Kugler, personal communication).
The abduction from the environment identifies what J. J. Gibson called an ecological affordance.
(Side note: There are some historical notes that
need to be made about Gibson’s precise notion of affordance. Karl Pribram suggests that Gibson’s
behavioral upbringing lead him to incorrectly insist that affordance was
something perceived, or environmental, rather than coming from the inside out. (Personal communication.) Ecological scholars like Michael Turvey
would have more to say on this than I.)
In any case, classical “deduction” is then a specific type of real instantiation of a mixing of anticipatory mechanisms, mostly organized by the environment, and memory of the structural invariances that have been encoded into a memory construction. In the human memory the construction is not fully understood even today. In the tri-level architecture for computational intelligence the induction is an aggregation of substructural elements under the global constraints of the abduction that is occurring due to a set of rules governing coherence and completeness.
The result is a single logic, or perhaps a set of logics
with some polylogical language. These
polylogical systems then can control engineered physical systems with deductive and
therefore algorithmic obtainable computations.
The mutual induction between environmental affordance and the emergence
of aggregated substructural elements (invariants) from memory produced a
deductive and thus rational system focused on the pragmatic of the moment.
I am not suggesting that I have this relationship between the four
categories correct. But it is
interesting to me that things seem to line up so well. Comments are welcome.