We go along with Seth’s claim that computational functionalism is neither necessary nor sufficient for consciousness as we know it. We also agree that in living organisms (the only entities currently known to be conscious), consciousness is dependent on autopoietic teleo-functional processes, and that similar processes may be necessary for any form of consciousness. This second claim opens several cardinal questions. Since Seth does not assume that all living organisms are conscious, what renders only some living organisms conscious? Seth hints that multimodal integration is at least part of the answer, but there is no discussion of this crucial point. It is not clear, for example, whether and why a nervous system is necessary for consciousness, what kind of neural-bodily interactions are necessary for its realization, and what the distribution of consciousness is in the living world.
We suggest that in order to answer such questions, we need to address two main issues. First, we must clarify what we mean by the term consciousness, since we find Nagel’s “what-it-is-like-to-be” suggestion, which is adopted by Seth, too vague to be helpful. Second, we need to better understand the fundamental affordances of biological systems, which are the foundation on which the newly evolved novel structures and processes that render a living being conscious are built.
Biological consciousness: Connected and partially disconnected
We define consciousness as-we-know-it (biological consciousness) as a mode of being grounded in perceived and felt private bodily experiences that share a common perspective and have physiological/behavioral (and therefore “reportable”) aspects.
According to this definition, none of the current AI algorithms is conscious since they have neither bodily grounded sensory experiences nor bodily grounded feelings. One may, however, argue that our focus on bodily grounded sensory perceptions and feelings is too limited. There are forms of consciousness, such as those that occur during sleep-dreaming, day-dreaming, deep meditation, and deep focused thinking, which are partially disconnected from current sensory inputs and current voluntary motor responses. However, these partially disconnected states depend evolutionarily and ontogenetically on connected, biologically grounded states. Assuming that disconnected, shallow, and solipsistic AI operations (such as those of LLMs) are similar enough to human conscious thinking is therefore unwarranted.
Even if one agrees that connectedness is essential for consciousness-as-we-know-it, it can be argued that embodied, silicon-made AI robots, with sensors, effectors, and embedded value systems, may provide the evaluated world and body inputs, which, on the embodied notion of consciousness we hold, are essential. Hence, the question: “Can non-biological systems be conscious?” resurfaces. This leads us to living-specific attributes, which are involved in all biological processes, including those underlying consciousness.
General biological affordances
Biological agents are very different from non-biological human-made machines (Jablonka & Ginsburg, Reference Jablonka and Ginsburg2022): they are dynamic, partially autonomous, autopoietic (self-generating and self-maintaining), carbon- and water-based, with a default inner turbulent materiality, which is necessary for the open-ended plasticity that allows their persistence. The material affordances of living systems include elasticity, diffusibility, sensitivity to external conditions like temperature and pH, and reactive excitability. Additional major affordances are storage of past effects and feedback interactions that can be actualized in a single molecule, in a molecular complex with a particular spatial organization, or in organized structures made up of complexes of complexes. These material affordances enable the adaptive, open-ended, autopoietic plasticity of living systems and their living components, which is essential for their long-term existence.
In addition to the affordances of biological materials, living organisms have ontogenetic affordances such as gradual increase in size and complexity, allowing flexible step-by-step sensory-motor and cognitive coordination; memory affordances based on multi-level memory systems that enable effective temporal and spatial control of embodied representations; and the affordances of hierarchically nested value systems that enable value prioritization conforming to the ultimate organismal goals of survival and reproduction. Together, these affordances are the basis of biological life, which has evolved in some lineages into the singular mode of being we call consciousness.
Minimal biological consciousness
The general biological affordances discussed above were necessary but not sufficient for biological consciousness. In previous publications, we explored the evolutionary transition from living non-conscious organisms to minimally conscious ones. We suggested that the evolution of open-ended, domain-general, generative, and representational associative learning, which we called UAL – Unlimited Associative Learning – led to the evolution of minimal consciousness and entailed it (in the part-whole, causal-constitutive sense of entailment). We showed that organisms that exhibit UAL instantiate the fundamental features of subjective experiencing, such as multimodal selective integration at different levels of organization, temporal thickness, flexible value systems, self-world distinction, and global accessibility and broadcast of information (Ginsburg & Jablonka, Reference Ginsburg and Jablonka2019). These dynamics of minimal consciousness, which first emerged in the Cambrian, required a central and differentiated nervous system, with specified multidirectional predictive interactions among a hierarchy of sensory, motor, memory, and value subsystems. Although we could not exclude the possibility that the perceptual-affective consciousness that UAL entailed in living organisms can be, in the future, instantiated in an AI silicon body, this seems, with current technology, implausible. In a previous paper we argued that an AI robot that realizes UAL-based consciousness will need to have some analogues of general as well as UAL-specific biological affordances (Bronfman, Ginsburg & Jablonka, Reference Bronfman, Ginsburg and Jablonka2021).
Disconnected consciousness in AI?
Since conscious thinking in humans is based, both ontogenetically and evolutionarily, on minimal consciousness, we suggest that the greatly impoverished material plasticity and connectedness in cognitively sophisticated AI robots may be only superficially similar to human conscious thinking. Open-ended cognitive plasticity based on computational functionalism involving penurious bodily connectedness may render AIs’ ways of knowing so different from what we recognize as conscious thinking that new concepts would be necessary to describe the inner states of such AIs. Consciousness-as-we-know-it is probably not a universal like energy or mass; it seems more like a mode of being that can be instantiated only in biological or biological-like organisms.
You have
Access
We go along with Seth’s claim that computational functionalism is neither necessary nor sufficient for consciousness as we know it. We also agree that in living organisms (the only entities currently known to be conscious), consciousness is dependent on autopoietic teleo-functional processes, and that similar processes may be necessary for any form of consciousness. This second claim opens several cardinal questions. Since Seth does not assume that all living organisms are conscious, what renders only some living organisms conscious? Seth hints that multimodal integration is at least part of the answer, but there is no discussion of this crucial point. It is not clear, for example, whether and why a nervous system is necessary for consciousness, what kind of neural-bodily interactions are necessary for its realization, and what the distribution of consciousness is in the living world.
We suggest that in order to answer such questions, we need to address two main issues. First, we must clarify what we mean by the term consciousness, since we find Nagel’s “what-it-is-like-to-be” suggestion, which is adopted by Seth, too vague to be helpful. Second, we need to better understand the fundamental affordances of biological systems, which are the foundation on which the newly evolved novel structures and processes that render a living being conscious are built.
Biological consciousness: Connected and partially disconnected
We define consciousness as-we-know-it (biological consciousness) as a mode of being grounded in perceived and felt private bodily experiences that share a common perspective and have physiological/behavioral (and therefore “reportable”) aspects.
According to this definition, none of the current AI algorithms is conscious since they have neither bodily grounded sensory experiences nor bodily grounded feelings. One may, however, argue that our focus on bodily grounded sensory perceptions and feelings is too limited. There are forms of consciousness, such as those that occur during sleep-dreaming, day-dreaming, deep meditation, and deep focused thinking, which are partially disconnected from current sensory inputs and current voluntary motor responses. However, these partially disconnected states depend evolutionarily and ontogenetically on connected, biologically grounded states. Assuming that disconnected, shallow, and solipsistic AI operations (such as those of LLMs) are similar enough to human conscious thinking is therefore unwarranted.
Even if one agrees that connectedness is essential for consciousness-as-we-know-it, it can be argued that embodied, silicon-made AI robots, with sensors, effectors, and embedded value systems, may provide the evaluated world and body inputs, which, on the embodied notion of consciousness we hold, are essential. Hence, the question: “Can non-biological systems be conscious?” resurfaces. This leads us to living-specific attributes, which are involved in all biological processes, including those underlying consciousness.
General biological affordances
Biological agents are very different from non-biological human-made machines (Jablonka & Ginsburg, Reference Jablonka and Ginsburg2022): they are dynamic, partially autonomous, autopoietic (self-generating and self-maintaining), carbon- and water-based, with a default inner turbulent materiality, which is necessary for the open-ended plasticity that allows their persistence. The material affordances of living systems include elasticity, diffusibility, sensitivity to external conditions like temperature and pH, and reactive excitability. Additional major affordances are storage of past effects and feedback interactions that can be actualized in a single molecule, in a molecular complex with a particular spatial organization, or in organized structures made up of complexes of complexes. These material affordances enable the adaptive, open-ended, autopoietic plasticity of living systems and their living components, which is essential for their long-term existence.
In addition to the affordances of biological materials, living organisms have ontogenetic affordances such as gradual increase in size and complexity, allowing flexible step-by-step sensory-motor and cognitive coordination; memory affordances based on multi-level memory systems that enable effective temporal and spatial control of embodied representations; and the affordances of hierarchically nested value systems that enable value prioritization conforming to the ultimate organismal goals of survival and reproduction. Together, these affordances are the basis of biological life, which has evolved in some lineages into the singular mode of being we call consciousness.
Minimal biological consciousness
The general biological affordances discussed above were necessary but not sufficient for biological consciousness. In previous publications, we explored the evolutionary transition from living non-conscious organisms to minimally conscious ones. We suggested that the evolution of open-ended, domain-general, generative, and representational associative learning, which we called UAL – Unlimited Associative Learning – led to the evolution of minimal consciousness and entailed it (in the part-whole, causal-constitutive sense of entailment). We showed that organisms that exhibit UAL instantiate the fundamental features of subjective experiencing, such as multimodal selective integration at different levels of organization, temporal thickness, flexible value systems, self-world distinction, and global accessibility and broadcast of information (Ginsburg & Jablonka, Reference Ginsburg and Jablonka2019). These dynamics of minimal consciousness, which first emerged in the Cambrian, required a central and differentiated nervous system, with specified multidirectional predictive interactions among a hierarchy of sensory, motor, memory, and value subsystems. Although we could not exclude the possibility that the perceptual-affective consciousness that UAL entailed in living organisms can be, in the future, instantiated in an AI silicon body, this seems, with current technology, implausible. In a previous paper we argued that an AI robot that realizes UAL-based consciousness will need to have some analogues of general as well as UAL-specific biological affordances (Bronfman, Ginsburg & Jablonka, Reference Bronfman, Ginsburg and Jablonka2021).
Disconnected consciousness in AI?
Since conscious thinking in humans is based, both ontogenetically and evolutionarily, on minimal consciousness, we suggest that the greatly impoverished material plasticity and connectedness in cognitively sophisticated AI robots may be only superficially similar to human conscious thinking. Open-ended cognitive plasticity based on computational functionalism involving penurious bodily connectedness may render AIs’ ways of knowing so different from what we recognize as conscious thinking that new concepts would be necessary to describe the inner states of such AIs. Consciousness-as-we-know-it is probably not a universal like energy or mass; it seems more like a mode of being that can be instantiated only in biological or biological-like organisms.
Financial support
Not funded.
Competing interests
None.