The Quest for a Universal Theory of Intelligence

The Mind, the Machine, and Singularity Hypotheses

Christian Hugo Hoffmann

• Éditeur: De Gruyter
• Année: 2022
• ISBN:
  9783110756197
• Édition: 1
• Sujet: Philosophy
• Sous-sujet: Epistemology in Philosophy

Recent findings about the capabilities of smart animals such as corvids or octopi and novel types of artificial intelligence (AI), from social robots to cognitive assistants, are provoking the demand for new answers for meaningful comparison with other kinds of intelligence. This book fills this need by proposing a universal theory of intelligence which is based on causal learning as the central theme of intelligence. The goal is not just to describe, but mainly to explain queries like why one kind of intelligence is more intelligent than another, whatsoever the intelligence. Shiny terms like "strong AI, " "superintelligence, " "singularity" or "artificial general intelligence" that have been coined by a Babylonian confusion of tongues are clarified on the way.

Part I: The different localizations, facets and forms of intelligence

Instead of fruitless attempts to divide the world into things with and things without the essence of mind, [or intelligence, C.H.], we should examine the many detailed similarities and differences between systems.

Aaron Sloman, 1984

It is crucial to isolate concepts from what they are concepts of, a common misleading assimilation.³ The entities or objects concepts refer or apply to are said to be within their extension. This Part I deals first and foremost with the extension or scope of the concept of intelligence and, eo ipso, investigates in what creatures we find which forms and facets of intelligence. It comes, however, directly with a restriction about its own scope: extraterrestrial intelligence is explicitly factored out from the analysis, and plants are merely touched upon briefly and thus practically out of scope. By contrast, we start with humans and animals, and move towards the speculative, i.e., AI as the field devoted to assembling artificial animals – or at least artificial creatures that, in tailored contexts, appear to be animals (Bringsjord & Govindarajulu, 2018).⁴ In fact, we close this section in Chapter 3 by refashioning the open and hotly debated question of whether AI should be considered intelligent which is illustrated by the juxtaposition of the philosophy of AI, examining “strong” versus “weak” AI (Bringsjord & Govindarajulu, 2018), and the simple economics of AI which refers to AI as mere prediction machines whereby predictive abilities are treated as a necessary, but not sufficient condition for human-level intelligence (Agrawal et al., 2018; Hoffmann, 2018a).

Moreover, we identify the gap that albeit much has been explored in the philosophy of AI, the general key term of intelligence remains under-researched in this connection – e.g., there is only a Stanford Encyclopedia entry on “AI” (Bringsjord & Govindarajulu, 2018), but none on “intelligence” as such. This might be somewhat perplexing as we witness that many different disciplines have tackled and joined forces to clarifying “intelligence”, from psychology to neuroscience and evolutionary biology. Therefore, we account for, but also mix these different perspectives in this Part I. Less astonishingly perhaps, each of them argues that their perspective is the one that truly matters. Here is, in a slightly exaggerated form, the sort of thing in a nutshell you might envisage in the following:

• What we need to care about is psychology, how a mind works in people and how that compares and contrasts with other people or animals. Cognition, including but not limited to learning, attention, memory, perception, but also affect or consciousness, and, of course, intelligence are all psychological categories and together make up the mental life of people and other animals.
• What we need to care about is evolution, the blind watchmaker (Dawkins, 1986), how the human species came to be equipped with heightened intelligence under evolutionary pressure. The human mind is a Swiss army knife of adaptations, and one cannot penetrate it without appreciating how it unfolded, how it is tethered to reproductive success and how the human faculty (cor-)relates to other species.
• What we need to care about is neuroscience, how a brain orchestrates itself to make us intelligent. Thereby, the mind seems to be entrenched by means of states and transitions of a large assembly of highly interweaved electrochemical components.
• What we need to care about is AI. Instead of getting caught up in the details of how human brains do the job of invoking intelligence, we shift our attention to the job itself, and ask how it, i.e., the design of intelligence or minds (Haugeland, 1997a), might get done at all. This is what Daniel Dennett calls taking a design stance (Dennett, 1981) and makes AI a subfield of intellectology (Hernández-Orallo, 2017: 21).
• (And there are others like linguistics: What we need to care about is human language as the “signature of human uniqueness”. Or a historical or religious view, …)

To avoid walking into that trap of the one-size-fits-all fallacy, i.e., that a monolithic view covers the whole spectrum of facets, forms and subtleties of intelligence, we believe a certain amount of humility and a holistic approach, respectively, is called for when we inspect a complex system like intelligence.⁵ Doing philosophy is largely a matter of trying to put things together, to synthesize. Good philosophy is opportunistic; it uses whatever information and whatever tools or whatever angles which look useful and which, playing in concert, are more useful. For example, psychology in its own right is handicapped by its behavioristic or cognitivist stance, by the fact that it gets to look at subjects and their actions only from the outside. External stimuli can be presented, and external responses can be monitored, but that is it.

A tricky question neuroscience has to cope with is whether we will be able to recover the regularities in a person’s behavior in the properties of that person’s components and their interconnections alone. And so on for other limitations of these and the other approaches. Hence, we harness and mingle the different perspectives in this Part I in lieu of jumping on just a single one.

Reinforced by the alignment of Part II, dealing with what philosophy (of science) teaches us on intelligence, the primary goal of this treatise is to shed more light on this salient concept in order to esteem divergent sorts of intelligent systems beyond anthropocentrism and AI mania and to gain meaningful insights on how they can be catalogued, evaluated, and contrasted. Let us launch the study with the concept “intelligence”’s scope, and a little disclaimer, respectively. We relegate speculation on possible extraterrestrial forms of intelligence (for that cf. Edmondson, 2012; Shuch, 2011) or “astrocognition” (Vakoch, 2011) and, guided by Figure 1, focus on intelligence on Earth instead.

Figure 1: Intelligence on Earth map. Source: Moravec, 1988: 18.

The evolution of intelligent life form is quite rare and could not be achieved without billions and billions of trials and errors (Hiraiwa-Hasegawa, 2019: 176). Life was unleashed on Earth about 3.8 billion years ago, but the genesis of animals with simple nervous systems, which has appeared to be a sine qua non for any “intellectual” species to evolve on Earth (see excursus 1 for a critical reply though), had to wait until about 580 million years ago (Hiraiwa-Hasegawa, 2019: 167f.). Hence, in terms of the sheer time of human history, it constitutes only an insignificant portion of the history of intelligent life (Diamond, 1992: 33; see also Figure 4). When we contemplate and walk around the tree of life today, we are confronted with some awkward cases of adaptation, with an arguably minimal cognition ranging from the very small (bacteria) to the very big (ecosystems; Hernández-Orallo, 2017: 93). As of 2004, we have about 1.5 million scientifically named species, however 99% of the species that have appeared on the Earth so far have already become extinct (Hiraiwa-Hasegawa, 2019). One more recent cause of this is that with regard to other animals, humans have since the last few thousands of years (since the agricultural revolution accelerated by the industrial revolution 250 years ago) become gods, Homo Deus (Harari, 2017), turning biodiversity of fauna into masses of domesticated animals of few kinds – 40,000 lions vs. 600 million house cats, 900,000 African buffalos vs. 1.5 billion domesticated cows, 50 million penguins vs. 20 billion chickens (Harari, 2017: 84; Barnosky, 2008). Or in the immortal words of Teilhard de Chardin (1946/1959: 30):

Lorsque Platon agissait, il n’avait probablement conscience d’engager par sa liberté qu’une parcelle du Monde, étroitement circonscrite dans l’espace et dans la durée. Quand un homme d’aujourd’hui opère en pleine conscience, il sait que son choix a retentissement sur des myriades de siècles et de vivants. Il sent en soi les responsabilités et la force d’un Univers tout entier.

Even though there are examples of ecological suicide during the Anthropocene where animal populations actually have eaten themselves into extermination, they often involve humans too that override the natural checks and balances regulating a species’ numbers (e.g., by accidentally or intentionally transferring the animal species from one part of the world to another; Diamond, 1992: 312f.). This nurtures the common, but naïve belief that non-human animal species live in balance with each other and with their milieu unless humans intervene. It is naïve because it disregards the dynamics on Earth, the impact of all kinds of changes since almost 4 billion years that let life forms and species come and go (not all can adapt to new situations; for mechanisms of evolution and the adaptationist program, cf. Millstein, 2002). Still, there is something to this view, an analogy maybe to climate change vs. anthropological climate change, insofar as species do not go extinct under natural conditions as rapidly as we are exterminating them now (except under very rare circumstances like the mass die-off 65 million years ago; Diamond, 1992: 312).

Since nobody knows how many species are actually left on Earth – there are many places where scientific exploration is still far from complete, such as the deep ocean –, we cannot assert to canvass the full scope of the concept of intelligence by this fact alone. And since this restriction already manages our expectations from what can be reasonably accomplished in Part I, we wish to add a second disclaimer. We will also rule out plants from our probes, only for the sake of not moving too far from the subject matter or our research questions outlin...

Table des matières

• Title Page
• Copyright
• Contents
• Acknowledgments
• Introduction
• Part I: The different localizations, facets and forms of intelligence
• Part II: Scaffolding intelligence
• Part III: Evaluating machine intelligence in current and past AI
• Part IV: Singularity hypotheses
• References
• Index