![Search icon](https://hackernoon.imgix.net/search-new.png?w=19&h=19)
2 Brief Overview of CtmR, a Robot with a CTM Brain
2.2 Conscious Attention in CtmR
2.3 Conscious Awareness and the Feeling of Consciousness in CtmR
2.4 CtmR as a Framework for Artificial General Intelligence (AGI)
3 Alignment of CtmR with Other Theories of Consciousness
4 Addressing Kevin Mitchell’s questions from the perspective of CtmR
7 Appendix
7.1 A Brief History of the Theoretical Computer Science Approach to Computation
7.2 The Probabilistic Competition for Conscious Attention and the Influence of Disposition on it
We have presented an overview of the CtmR model. Now we indicate how, at a high level, the model naturally aligns with and integrates key features of major theories of consciousness, further supporting also our view that CtmR provides a framework for building a conscious machine.
CtmR aligns broadly with the architectural and global broadcasting features of the global workspace theory of consciousness (Baars, Bernard J., 1997), and at a high level with the global neuronal workspace theory of consciousness of neuroscientists Stanislas Dehaene, Jean-Pierre Changeux (Dehaene & Changeux, 2005), (Dehaene S. , 2014), and others.[22]
However, CtmR differs from GW in significant ways. For example: CtmR has a formally defined natural competition for information to become globally broadcast; it constructs world models; and CtmR has no Central Executive, a feature, not a bug.
CtmR’s ability to construct and utilize models of CtmR’s worlds (inner and outer) and the key role they play in CtmR’s conscious awareness aligns closely with Michael Graziano’s attention schema theory of consciousness (Graziano, Guterstam, Bio, & Wilterson, 2020). AST proposes that the brain is an information processing machine that constructs a simplified model of attention, like it constructs a simplified model of the body, the Body Schema. According to AST, this Attention Schema provides a sufficiently adequate description of what it is attending to for the brain to conclude that it is “aware”.
Predictive processing asserts that the brain is constantly inferring, correcting and updating its predictions, generally based on motor outputs and sensory inputs. CtmR’s predictive dynamics (cycles of prediction, testing, feedback, and learning/ correcting), locally and globally, align with various incarnations of predictive processing (von Helmholtz, 1866; 1962), (Friston K. , 2010), (Cleeremans, 2014), (Clark A. , 2015), and (Hohwy & Seth, 2020) and others.[23]
CtmR’s ability to construct (and utilize) models of its worlds containing rich Brainish labeled sketches (leading to its feelings of consciousness) derives in part from a mix of its embodied, embedded, enactive, and extended mind. CtmR is embedded in its outer world and, through its embodied actuators, can enact in this world, thus influencing what it senses and experiences. CtmR’s “mind” is extended by information it gleans from resources in its outer world, and from its embedded (or linked) off-the-shelf processors.
This aligns with the “4E” view that consciousness, like cognition (Carney, 2020), also involves more than brain function (Rowlands, 2010).
Embodied: Phenomenal consciousness involves incorporating relations with the entity’s body parts and processes, (Damasio, 1994), (Edelman, 2006) and (Shanahan, 2005)[24].
Embedded, Enactive: Phenomenal consciousness involves the entity enacting and interreacting with its outer world, thus affecting its world and experiences (Maturana & Varela, 1972), in English (Maturana & Varela, 1980), (Varela, Thompson, & Rosch, 1991), (Thompson, 2007), and (Clark A. , 2008).
Extended: And consciousness is enhanced by the entity having access to rich external resources (such as libraries, Google, ChatGPT. Mathematica,...) (Clark & Chalmers, 1998).
IIT, the theory of consciousness developed by Giulio Tononi (Tononi, 2004), and supported by Koch (Tononi & Koch, 2015), proposes a measure of consciousness called Phi, that in essence measures the amount of feedback and interconnectedness in a system. CtmR’s extensive feedback (its predictive dynamics globally and locally) and its interconnectedness via global broadcasts contributes to its high Phi.
CtmR aligns with aspects of evolutionary theories of consciousness. Oryan Zacks and Eva Jablonka provide evidence for the evolutionary development of a modified global neuronal workspace in vertebrates (Zacks & Jablonka, 2023) reinforcing our suggestion that an AI with a global workspace architecture could possess access consciousness.
In Sentience, Nicholas Humphrey presents an evolutionary argument for the development of phenomenal consciousness in warm-blooded animals (Humphrey, 2023). In “The Road Taken” (Chapter 12 of Sentience), Humphrey spins a “developing narrative”, starting with “a primitive amoeba-like animal floating in the ancient seas. Stuff happens. …” The resulting story provides a roadmap on how an entity might create world models and, sense of self. Indeed, this roadmap closely parallels the way CtmR’s world models evolve, and how CtmR develops its sense of self and phenomenological conscious awareness.[25]
In The Hidden Spring, Marc Solms makes the case that the source of consciousness is the arousal processes in the upper brain stem (Solms M. , 2021). More generally, Solms cites the Extended Reticulothalamic Activating System (ERTAS) as the generator of feelings and affects, enabling consciousness. “Affective qualia” is the result of homeostasis. “[D]eviation away from a homeostatic settling point (increasing uncertainty) is felt as unpleasure, and returning toward it (decreasing uncertainty) is felt as pleasure” (Solms M. , 2019). Homeostasis arises by a system resisting entropy, i.e., minimizing free energy (Solms & Friston, 2018). This is enabled by a Markov blanket (containing the system’s input sensors and output actuators) that insulates the internal system from its outer world. The “system must incorporate a model of the world, which then becomes the basis upon which it acts[26]” (Solms M. , 2019).
At a high level, CtmR aligns with ERTAS + FEP:
Although GW models generally consider processors as performing cortical functions, CtmR goes beyond that. There is nothing to preclude CtmR from having processors that function as the ERTAS.
In (Blum & Blum, 2021), we discuss pleasure and pain in the CTM (known here as CtmR). Our discussion of pleasure aligns with Solms, and also with (Berridge & Kringelbach, 2015). Pain is more complex. In (Blum & Blum, 2021) we discuss in more detail how the feeling of pain might be generated.
Predictive dynamics (cycles of prediction, testing, feedback and correcting/learning) in CtmR works to reduce prediction errors, an analogue to minimizing free energy.
And the incorporated Model-of-the-World in CtmR is the basis upon which CtmR acts.
We have taken one of the well-known Friston diagrams (Parr, Da Costa, & Friston, 2019) with Markov blanket separating internal and external states, rotated it clockwise 90 degrees, and then superimposed it on CtmR (with a little stretching and shrinking). And voilà, a perfect fit!
This paper is available on arxiv under CC BY 4.0 DEED license.
[22] Additional references for GNW include: (Dehaene & Naccache, 2001), (Sergent & Dehaene, 2005), (Dehaene & Changeux, 2011), and (Mashour, Roelfsema, Changeux, & Dehaene, 2020).
[23] Other references include: (McClelland & Rumelhart, 1981), (Lee & Mumford, 2003), (Friston K. , 2005), (Clark A. , 2015), (Seth, 2015), (Miller, Clark, & Schlicht, 2022).
[24] We note that here Shanahan views the global workspace as key to access consciousness, but that phenomenal consciousness requires, in addition, embodiment.
[25] We claim Humphrey actually gives a road map how an entity, warm blooded or not, might create world models and sense of self. As an exercise, we have re-written part of Chapter 12, “The Road Taken”, from the perspective of CtmR and sent a copy to Humphrey. His reply, “It would be great if we could meld these theories.” (Personal communication with Nick Humphrey, Oct 9, 2023.)
[26] Italics here ours.