Original article was published by Nick Nielsen on Artificial Intelligence on Medium
Bird Brains: Better than We Thought
Alternative Neural Architectures for Consciousness
The issue of Science for 25 September 2020 features a crow on its cover with the headline “Avian Awareness: Carrion crows display sensory consciousness.” There are three articles in the journal on this theme, “A neural correlate of sensory consciousness in a corvid bird” by Andreas Nieder, Lysann Wagener, and Paul Rinnert, “A cortex-like canonical circuit in the avian forebrain” by Martin Stacho, Christina Herold, Noemi Rook, Hermann Wagner, Markus Axer, Katrin Amunts, and Onur Güntürkün, and “Birds do have a brain cortex — and think” by Suzana Herculano-Houzel.
Everyone who has watched crows carefully knows that they are intelligent birds. A friend once told me that if he went outside and pretended to target crows with a broom handle as though it were a gun, the birds would not move, but if he went outside with an actual gun, the birds would scatter. There is a video of a crow repeatedly sliding down a snowy roof, as well as another video of two crows sliding and rolling on a snow-covered car, which looks like the kind of intentional play behavior we associate with mammals (there are many similar videos of crows playing). I’m sure everyone has their own anecdotal account of avian intelligence.
Now we have something more than anecdotal evidence for corvid intelligence. The articles in Science report, respectively, an experiment that implies sensory consciousness and anatomical features of the corvid brain that are analogous, but not identical, to the mammalian brain. Herculano-Houzel notes that it has long been said that birds have no cerebral cortex, but she goes on to explain that the avian pallium derives from the same embryonic developmental structures from which the mammalian cerebral cortex derives. (She cites “A developmental ontology for the mammalian brain based on the prosomeric model” by Luis Puelles, Megan Harrison, George Paxinos, and Charles Watson, in which the authors argue, “Because genomic control of neural morphogenesis is remarkably conservative, this ontology should prove essentially valid for all vertebrates…” which would include both birds and mammals.)
Similarly, the conventional view has been that the limbic system is unique to mammals, but there may be structures in the avian brain that are homologous to the limbic system. A re-assessment of the avian brain is evident from papers such as Avian brains and a new understanding of vertebrate brain evolution by The Avian Brain Nomenclature Consortium, and Cell-type homologies and the origins of the neocortex by Jennifer Dugas-Ford, Joanna J. Rowell, and Clifton W. Ragsdale, and this re-assessment has been carried back to common ancestors of mammals and birds, as we find in the paper The Limbic System of Tetrapods: A Comparative Analysis of Cortical and Amygdalar Populations by Laura L. Bruce and Timothy J. Neary. All of this points to the increasing complexity and detailed articulation of evo-devo conceptions and the idea of deep homology, such that highly conserved genes produce similar structures — eyes, brains, and perhaps consciousness too — across many different species, even when there isn’t a direct line of descent; we should take this as a memo to similarly examine behavioral evolution from an evo-devo standpoint, but leave that aside for now.
Given the earlier research in the papers cited above, we would not be surprised to learn of further homologies being recognized to hold between avian and mammalian brains, but while there may be unrecognized neural homologies between birds and mammals, the bird brain is quite different from a mammalian brain. The Stacho, et al., paper addresses these different neuronal structures, but they conclude, “Our study reveals a hitherto unknown neuroarchitecture of the avian sensory forebrain that is composed of iteratively organized canonical circuits within tangentially organized lamina-like and orthogonally positioned column-like entities.” In other words, the avian pallium exhibits an architecture of layered neurons, and columns connecting the layers, which is a structure than has long been understood to characterize the mammalian cerebral cortex. The two structures are distinct in detail, but display overall similarities in the way in which iterated and interconnected neural circuits are arranged.
The Nieder, et al., paper approaches avian intelligence through behavioral research rather than through anatomy, although the stimulus response experiments are traced to a single neuron, so that there is an anatomical component to this research as well. The authors write:
“We trained two carrion crows (Corvus corone) to report the presence or absence of visual stimuli around perceptual threshold in a rule-based delayed detection task. At perceptual threshold, the internal state of the crows determined whether stimuli of identical intensity would be seen or not perceived. After a delay, a rule cue informed the crow about which motor action was required to report its percept. Thus, the crows could not prepare motor responses prior to the rule cues, which enabled the investigation of neuronal activity related to subjective sensory experience and its lasting accessibility.”
Nieder, et al., recognize the philosophical problems involved here by citing the famous paper by Thomas Nagel, “What is it like to be a bat?” They add, “…whether pure subjective experience itself (‘phenomenal consciousness’) can and should be dissociated from its report (‘access consciousness’) remains intensely debated.” And so it is.
The Nieder, et al., paper, though it appears in the same issue of Science as the Stacho, et al., paper, is entirely independent of the Stacho, et al., paper, and the former repeats many of the traditional assumptions about the absence of a cerebral cortext in the avian brain. However, knowing what is now shown in the Stacho, et al., paper, and its earlier anticipations, we should not be at all surprised to find both empirical evidence of consciousness and mechanisms of sensory consciousness in birds that are apparently parallel to those of mammals. Our common terrestrial ancestry, and the DNA all life in the terrestrial biosphere shares, seems to count quite significantly toward cognitive similarity, and points to the possibility of an evo-devo cognitive science.
Both Nieder, et al, and Herculano-Houzel discuss the phylogenesis of consciousness: since mammals and birds have a common ancestor about 320 million years ago, this raises the question of whether the common ancestor to both birds and mammals had some rudimentary form of consciousness, or whether consciousness appeared later, independently emerging in both birds and mammals. (I just discussed what I call the phylogenesis of mind in my newsletter 101.)
On the one hand, accounting for consciousness by the deep homology of highly conserved genes closely ties consciousness to the terrestrial biosphere and its contingent processes; on the other hand, multiple distinct biological mechanisms that realize consciousness suggest that consciousness as an emergent complexity is not exclusively reliant upon the specific biological mechanisms and neuronal architecture of highly developed mammal brains, which is the way in which we ourselves are familiar with consciousness. This in turn suggests that other intelligence in the universe could also be conscious intelligence something like we know from our own experience, and a mind constrained by the reality of consciousness as we know it would be at least partially understandable by us — and we would be at least partially understandable by an alien conscious intelligence — in virtue of shared consciousness, even if the biological underpinnings of consciousness were distinct in each case.
We cannot communicate via (grammatically structured) language with other forms of life on Earth, but we can and do communicate with them in terms of conscious interaction with other conscious beings. Even a biological relationship as adversarial as predation, for example, is mediated by consciousness — both beings seeking to survive, while one listens and watches in order to detect a threat, while the other waits and watches for a moment to pounce. (I earlier made a similar point in A Sentience-Rich Biosphere.) This ecological relationship is mediated by a conscious relationship between predator and prey, i.e., the shared consciousness of both predator and prey. Similarly, communicative relationships between ourselves and other beings that evolved in other biospheres, such as is postulated as the basis of SETI, could have a similar communicative structure based in shared consciousness that mediates an ecological relationship (with “ecological” here understood in a cosmological sense), even if it should turn out to be the case that human and alien minds are incommensurable and communication in the sense of shared information content is not possible (i.e., if what Freeman Dyson criticized as the “philosophical discourse dogma” is, in fact, an unsupported dogma).
These findings regarding avian consciousness should also be of great interest to artificial intelligence researchers, in so far as artificial intelligence can be conceived (even if it is not always conceived) as machine consciousness. Machine intelligence that is not conscious would be alien from human intelligence in a fundamental way (in the same way that an extraterrestrial intelligence what was not conscious would be more alien to us than a conscious mind). Artificial intelligence that was the result of machine consciousness, like an alien consciousness, would have at least something in common with us, increasing the possibility of our having aligned interests (i.e., the constructed AI more likely to be friendly AI).
Knowing that consciousness in both avian and mammalian brains may be associated with layered neural structures, engineers of computer hardware involved in artificial intelligence might consider constructing an iterated architecture of layered neural pathways — that is to say, layered neural networks — connected every so often by columns, and so producing a different kind of hardware more specifically suited to the emergence of consciousness.
The economic motive for artificial intelligence research is simply to extend automation beyond what automation has accomplished to date, and this is certainly where the most significant economic gains are likely to be found; this research will pay for itself. But the epochal breakthrough in computer science will not appear from the incremental improvement of increasingly “intelligent” expert systems, but from the appearance of machine consciousness, which is something entirely different from what is today understood by “artificial intelligence.” Since artificial intelligence researchers seem to be mostly content writing and re-writing software that runs on more or less the same hardware, artificial consciousness is not likely to emerge from these efforts; machine consciousness will probably require distinctive hardware that imitates the neuronal architecture of biological brains from which consciousness is an emergent.
But suppose that we can isolate the neuronal circuits of consciousness, and reproduce them in hardware form: once we can do this, we can do this at a larger and at a more complex scale than exists in any biological brain. If consciousness is an emergent from iterated layers and columns of neurons, hardware mimicking layers and columns of neurons could be constructed that also serves as an emergent basis for consciousness, and a consciousness that could be far more sophisticated than any biological consciousness, insofar as the technological basis of consciousness could be rapidly streamlined and miniaturized. From such a research program an optimized consciousness could emerge. And not only optimized consciousness, but it might also be possible to engineer qualitatively distinct forms of consciousness that are variously optimized for specific tasks.