Eliasmith’s perspective on the architecture of minds is a functionalist one (this collection, p. 2, p. 6, pp. 6–7, pp. 9–11, p. 13). The thread running through his paper is his interest in “understanding how the brain functions” and realizing “detailed functional models of the brain” (ibid., p. 9). The basic idea is that if we construct artificial minds and endow them with certain functions (such as natural language and human-like perceptual abilities), we can examine empirically, in a process comparable to reverse engineering, what it is that constitutes so-called mindedness (ibid., p. 11). But in their striving to unearth the nature of mindedness, it is not the task of artificial intelligence research or biology to deliver comprehensive and full-fledged theories on biological cognition in general and human cognition in particular. Rather, a very interesting reciprocal relationship between the two parties, in which one learns from the other, is what will propel forward our understanding of biological cognitive systems. In the following I give an overview of the most relevant points that are presented in the target article. They will be divided up into the original sections (technical, theoretical, and empirical).
First, in the technical area and according to Eliasmith, we are fairly far advanced—although there are certain hindrances to successfully implementing theories on this technology. The main obstacle is the size of artificial neuronal systems and, connected to that, their power consumption. Even though neuromorphic chips are being improved steadily, the number of neurons that can be reproduced artificially is still much lower than the number of neurons a human brain has. Thus, the processing of information is significantly slower than in natural cognitive systems (Eliasmith this collection, p. 14). Consequently, what can be realized in the field is still far from the complexity displayed by natural, biological cognition. However, as Eliasmith argues, since we are already in possession of the theoretical groundwork, the main barrier to overcome are technological advances (ibid., p. 9). Throughout the paper, Eliasmith informs the reader that in case we had the technologies needed, artificial minds would immediately be created (ibid., e.g., p. 7, p. 9, p. 11). However, where Eliasmith emphasizes technological barriers, I would like to point out that theoretical obstacles exist as well. These mainly revolve around the fact that a system of ethics has to be created before we encounter artificial agents. Eliasmith also comments on the consequences for philosophy, arguing that some major positions in the philosophy of mind, such as functionalism, will receive more empirical grounding (ibid., p. 11).
It seems as if the tacit understanding that Eliasmith has of the function of artificial minds is that they serve as shared research objects of biology and artificial research science in order to gain a better understanding of biological cognition (this collection, p. 9). That is of course only true if indeed the functional architecture of the artificial agent produces convincing behavior, similar to that of biological cognitive systems (humans and animals alike). To illustrate possible problems, one can think of the fact that in research, we learn from animal experiments, even though these animals are quite different from us in many ways. They are, however, similar or at least comparable in one epistemically relevant and specific aspect, i.e., the one that is to be examined, for example in certain aspects of metabolism used to test whether a new drug causes liver failure in humans (Shanks et al. 2009, p. 5). It is the same with artificial agents: they are similar to us in their behavior and thus a worthwhile research object. As such, we could formulate the underlying reasoning as a variant of analytical behaviorism. Analytical behaviorists suppose that intrinsic states of a system are mirrored in certain kinds of behavior. Two systems displaying identical behavior on the outside can be investigated in order to detect whether they do so on the inside as well (Graham 2010). This means that we could gain insight on the origin of mental states from a functionally isomorphic system, i.e., an artificially constructed system that is identical in organization and behavior to the natural system copied.
Last, since it seems that it will be possible in the future, given the required hardware, to design artificial agents according to our needs, it does not appear far-fetched to assume that the quality of human life might consequently be improved to a great extent (Eliasmith this collection, p. 11). This requires, however, that we make up our own minds about how to interact with such agents, which rights to grant and which to deny them. And also the opposite case may not be disregarded: it is imaginable that the artificial agents will at some point turn the tables and be the ones to decide on our rights (cf. Metzinger 2012). In highlighting aspects from different areas to be considered, Eliasmith reminds us of the possibilities that lie ahead of us, but also of the challenges that might show up and have to be faced. I want to suggest that we also take into consideration alternative outcomes that are not minds in the biological sense, but rather derivates of minds. I will therefore put the notion of postbiotic systems into play as a way of escaping the dichotomy “human-like” vs. “artificial” (Metzinger 2013). The philosophical point here is that the conceptual distinction between “natural” and “artificial” may well turn out to be non-exhaustive and non-exclusive: there might well, as Metzinger points out, be future systems that are neither artificial nor biological. By no means do I intend to argue against the use of scientific models, since they are what good research needs. Rather, I wish to draw attention to the possible emergence of intermediate systems, rather than only the extremes (i.e., human-like vs. artificial agents), or classes of systems that go beyond our traditional distinctions, but which nevertheless count as “minded”. As mentioned above, this is due to these intermediate or postbiotic systems being possible much earlier—probably preceding full-blown minded agents.
I will end this section by drawing attention to some of the author’s thoughts on the crucial elements of artificially-minded systems. According to Eliasmith, three types of skills are vital in building artificial minds: cognitive, perceptual, and motor skills have to be combined to create a certain behavior of the minded artificial agent. This behavior will then serve as the basis for us humans to judge whether we perceive the artificial agent as “convincing” or not (Eliasmith this collection, p. 9). Unfortunately, no closer specification of what it is to be “convincing” is given in the target article. No theoretical demarcation criterion is offered. What we can say with great certainty, however, is that in the end our subjective perception of the artificial agents will be the decisive criterion. One could speculate on whether it is merely an impression, or even an illusion, that leads us to concluding that we are facing a minded agent. According to Eliasmith, any system that produces a robust social hallucination in human observers will count as possessing a mind.