Alphabetical listing of contemporary AI theories by inventor's surname
A
B
Baars, B. Global Workspace Theory
C
Chomsky, N. Minimalist Program (MP). Chomsky's MP is ostensibly a theory of language. However, since Chomsky also espouses the idea of linguistically structured cognition, it is also a theory of mind. That is, MP's internal and external merge concepts describe mental as well as linguistic operations.
D
Dyer, M.C. GOLEM/TDE Theory. Research started in 2012 at Flinders University, South Austrtalia, as an Honours Science thesis. Mind-body problem is solved by semantic grounding, where semantics are compositional, combinational and set-theoretic. Semantic states describe ranges of perceptual measures (percept classes), in a not-dissimilar way to high-level (first order) data types in programming language theory. That is, program semantics are schematically akin to human level-3 cognolinguistic semantics. The key example is as follows- reading or understanding speech at cognolinguistic level 3 uses fundamentally the same type of semantic states and semantic state transitions as does subjective experience of 3D space at spatiotemporal level 2. In each case, global, holistic semantic composition is a combinational code built from semantic components. As with language, there is no direct specification of the detailed reality represented. Rather, semantic codes, be they abstract and symbolic (ie cognitive and/or linguistic) at level 3, or sensory and subjective (ie spatial trajectories/keyframe animations built from situation images [5]) at level 2, act as permissible solution delimiters, much as data types do in computer programming. The preferred mathematical analogue is that of combinations of inequalities which act as boundaries to the solution space, rather than simultaneous sets of equations which may or may not have an exact set of solution vectors
G
Graziano, M.S. Attention Schema Theory (AST). Graziano seems to take the approach that consciousness is magic, it is unfathomable, so we should instead ask our selves, "What is the adaptive advantage of this style of self-description, ie that I am conscious, and having a subjective experience?" Like Dennett, Graziano believes that, since he can't work it out, no one can!
M
Metzinger, T. (2015) M-Autonomy. J.Cons.Studies, 22, No. 11-12 pp270-302
In all his research, Tom Metzinger investigates the following- what are the necessary and sufficient conditions for the appearance of a phenomenal self? Self-Model Theory (SMT )(Metzinger 2003a, chapter 3) offers a set of ten such potential constraints, the two most important of which are functional integration with a system's "virtual window of presence" (i.e., its internal representation of time and an extended "Now") plus ongoing, dynamical integration into a single, overarching world model (i.e., a multimodal representation of the current situation as a whole).
The self is not, of course, an independent physical object, or even an independent virtual object (ie a fixed data structure). Rather, what does exist is the experience of being a self, as well as the diverse and constantly changing contents of self-consciousness. This is what philosophers mean when they talk about the ''phenomenal self'. GOLEM theory claims to know the solution to this problem - that the 'container' for conscious experience is the semantic state hierarchy (SSH) type of neural memory. Like a Klein bottle (a higher-order Moebius strip), this container can 'contain' itself, ie represent itself as a sub-representation, because it has rudimentary recursive properties. All recursive systems contain images or copies of themselves [5].
GOLEM theory divides cognitive computations into two kinds-
Metzinger's research also concurs with GOLEM theory on another point- that the standard form of conscious thought can be non-agentive, ie it sometimes lies below the level at which we conceive ourselves as individual persons, or 'selves'. Metzinger uses the example of 'daydreaming' to illustrate this possibility.
GOLEM theory claims that consciousness first evolved for animals, not humans. For the vast majority of animals, therefore, self-conceptualisation is computationally impossible or at least, unlikely. In GOLEM theory, the self is a level-3 construct, existing at the cognolinguistic level, while consciousness is a level-2 construct, existing at the spatiotemporal level. However, note that Metzinger includes three orders of embodiment/physical-semantic grounding in his discussions (first-order embodiment, second-order embodiment, and third-order embodiment).
Metzinger points out: some mental activities are not autonomously controllable, because one centrally important defining characteristic does not hold: they cannot be inhibited, suspended, or terminated. (Metzinger, 2015, p. 275). However, he identifies M-autonomy as another kind of mental activity in which the person is able to exert veto control over thoughts.
Metzinger's crude treatment of self-conceptualisation illustrates the degree to which GOLEM theory (GT) is more advanced. GT is based upon the VxC Cartesian product framework, ie it clearly distinguishes volition (a motor-side concept) from consciousness (a sensor-side concept). As well as distinguishing motor and sensor side concepts, it also identifies three levels of semantic abstraction and grounding (3SAG) - see figure Beta.1 below.
Metzinger fails to specify the relationship between M-autonomy (and the conditions under which it fails, the mental 'blink' for example) and PP theory. The main idea behind PP is that cognition consists in a hierarchically organized process of generating predictions about the states of the sensory periphery and minimizing the error of said predictions (Rao and Ballard, 1999). As Dyer (ie me) and Feldman have (exhaustively) pointed out, the real mechanism is the 'opposite' of PP, ie variable setpoints. That is, PP is a flawed (ie wrong) idea.
P
Pulvermueller proposes the existence of four semantic mechanisms. A direct quote follows-
These seem [3] to match the four abstraction levels of GOLEM's semantic hierarchy, as depicted in figure beta.1 below [1]. Very broadly speaking, Pulvermueller derives his tetramodal semantics from a bottom-up approach (ie detailed analysis of brain imaging data, with particular focus on the exceptions due to localised brain lesions), while GOLEM theory [2] derives a remarkably similar four-level model of integrated semantics from top-down considerations derived from a novel interpretation of Chomskyian (cognitive) linguistics.
GOLEM top-down theory (based on the Marr-Chomsky hierarchy) and Pulvermueller's 'bottom-up' concept of 'topotopic' [4] interconnectivity complement one another. This means that fact-by-fact comparisons are fair and reasonable, and not weakened by variable-context caveats.
T
Tononi Integrated information Theory (Tononi, 2008)
Figure Beta.1
1. Note that the level-for-level matching shown in figure beta.1 is an assumed mapping proposed by this author, Dyer, not Pulvermueller.
2. Dyer, M. (2012) TDE: A biologically plausible Turing Engine with a circular tape memory. Honours thesis, Flinders University of South Australia
3. A more thorough and detailed comparison between the theories is beyond the scope of this section.
4. A neologism used to define virtual or meta-topographies.
5. TDE is a recursive acronym meaning TDE Differential Engine, similar to GNU (GNU's Not Unix)