The Anthropocentrism of Intelligence: Rooted Assumptions that Hinder the Study of General Intelligence

Abstract

We postulate that Artificial General Intelligence remains elusive because of numerous undisputed assumptions that are deeply rooted into the traditional understanding of intelligence. We claim that these assumptions shape an anthropocentric bias that prevents the development of a general theory of intelligence capable of explaining the behavior of not only human and machine intelligence, but also any other entity that exhibits intelligent behavior. The most important of these assumptions is the failure to recognize darwinian evolution as an intelligent entity despite the growing consensus about its superior capabilities to develop biological contrivances. In order to avoid underrating and neglecting evolution as intelligent, other assumptions must be dropped. Such is the case for the requirement of language, which is only relevant in social contexts. Moreover, the boundary of evolution as an agent distinguished from the environment is not well-defined, which suggests that agent boundaries are redundant in General Intelligence and results in an equal treatment of polymorphic robots and multi-agents, to name a few. By revealing these and other assumptions, we propose that human intelligence should be relieved from standing at the center of studies about General Intelligence.

Full text

JSAI Technical Report SIG-AGI-011-07
The Anthropocentrism of Intelligence: Rooted
Assumptions that Hinder the Study of General
Intelligence
Francisco J. (Paco) Arjonilla1∗Yuichi Kobayashi1
1Shizuoka University, Graduate School of Science and Technology
Abstract: We postulate that Artificial General Intelligence remains elusive because of numerous
undisputed assumptions that are deeply rooted into the traditional understanding of intelligence.
We claim that these assumptions shape an anthropocentric bias that prevents the development of
a general theory of intelligence capable of explaining the behavior of not only human and machine
intelligence, but also any other entity that exhibits intelligent behavior. The most important of
these assumptions is the failure to recognize darwinian evolution as an intelligent entity despite the
growing consensus about its superior capabilities to develop biological contrivances. In order to
avoid underrating and neglecting evolution as intelligent, other assumptions must be dropped. Such
is the case for the requirement of language, which is only relevant in social contexts. Moreover,
the boundary of evolution as an agent distinguished from the environment is not well-defined,
which suggests that agent boundaries are redundant in General Intelligence and results in an equal
treatment of polymorphic robots and multi-agents, to name a few. By revealing these and other
assumptions, we propose that human intelligence should be relieved from standing at the center of
studies about General Intelligence.
1 Introduction
Artificial General Intelligence is a much pursued
achievement since the birth of Artificial Intelligence
in the mid 20th century. General Intelligence goes a
step further by dealing with neuroscience and cogni-
tive science as well. Despite all the efforts, there is no
successful theory of intelligence that explains human
intelligence nor any field that attains human-level in-
telligence. The reasons are many, and here we propose
that the assumptions about what intelligence should
be is hindering the research on what intelligence really
is. We believe that this bias is an important factor in
neglecting evidence that might hold the key to un-
derstand General Intelligence. A related bias might
appear if we are confronted with an intelligent system
and reject it if the intelligent processes do not match
our expectations of elaborateness and intricacy.
On the other hand, it is generally accepted that hu-
man intelligence is the reference to study General In-
telligence, yet this approach has only led to narrow in-
telligence. Artificial Intelligence is a fragmented field
with many different subareas that have been devel-
∗E-mail: pacoarjonilla@yahoo.es
oped independently [1]. Some researches attempt to
combine several technologies [2, 3], but mostly are di-
rected towards making different technologies work to-
gether rather than developing the common grounds of
different approaches to Artificial Intelligence. General
Intelligence requires all these subfields to be unified,
or generalized, into a single framework that displays
the advantages and disadvantages of these technolo-
gies with respect to each other. Likely, a combined
approach that replicates the advantages of each tech-
nology, but not the disadvantages, would become a
huge step towards General Intelligence.
Pfeifer and Scheier [4] gave a detailed account on
behavior-based intelligence to argue for embodied in-
telligence. Langley [5] claimed that developing cog-
nitive architectures is an important path to the de-
velopment of generally intelligent systems. Laird &
Wray [6] proposed a list of eight features that human-
level intelligences should have. Pfeifer and Gomez [7]
suggested the neccessity of interaction between the in-
telligent agent and a physical and social environment.
Arakawa et al. [8] gave a broad overview of the current
state of the art in Artificial General Intelligence. The
common denominator is that all these researchers fol-

JSAI Technical Report SIG-AGI-011-07
low the orthodox view of intelligence, which suggests
that human intelligence is the target to replicate, but
does not question that human intelligence has evolved
to accommodate to its environment and may not be
the best ambassador of General Intelligence.
Objective
The objective of this paper is to study the obstacles
that may hinder the study of General Intelligence by
challenging the core assumptions that drive Artificial
Intelligence. With that in mind, we do not intend to
propose a theory of General Intelligence, but rather
pave the way to facilitate the development of such a
theory.
We hope that our viewpoint of Intelligence will help
understand the fundaments of Intelligence by strip-
ping away circumstantial features of human intelli-
gence. In that respect, we see it as a particular real-
ization of a theory of General Intelligence. These cir-
cumstantial features are easier to identify if we com-
pare human intelligence with other systems that ex-
hibit intelligent behavior. Specifically, we claim that
darwinian evolution exhibits intelligent behavior and
thus it must be considered an intelligent system. We
will analyse human intelligence, darwinian evolution
and artificial intelligence to find assumptions incom-
patible with any of these three systems and justify
the rejection of those assumptions as part of a gen-
eral theory of Intelligence, where possible.
2 Assumptions
We will start by defining intelligence and exploring
the problems related to any definition of intelligence.
Afterwards, we will show that some assumptions in
Artificial Intelligence have not been properly justified
and how they lead to revealing the rest of the as-
sumptions. Most of these assumptions come from an
anthropocentric view on intelligence that places repre-
sentations of the world and speed at solving problems
in the main focus of General Intelligence.
2.1 Intelligence requires a definition
Before delving into the intricacies of intelligence,
we need to explain what exactly we are referring to.
This is generally attained by an ambiguous definition
of intelligence or by developing intelligence tests that
measure some of the many abilities of human intellect,
e.g. the g factor [9]. Legg & Hutter [10] collected and
analysed numerous definitions of intelligence. The
problem with these approaches is that once the defi-
nition of intelligence is set, we lose the capability to
evaluate intelligence in a general way. In order to un-
derstand the problem, consider the following analogy.
In quantum physics, quantum entities are described
with wave functions. However, the complete state of
a wave function cannot be directly measured because
the measurement collapses the wave function into a
particle. Similarly, the study of intelligence collapses
into a biased view of intelligence as soon as it is de-
fined, but the definition is still required for objective
research.
This paper is no exception. We define intelligence
as fulfillment of goals, deliberately leaving out any
specification of goals. This way the definition only
assumes intentionality of intelligence while remaining
as general as possible, and without implying that this
is the correct definition of intelligence. For example,
problem solving is included in this definition by defin-
ing the goal as such. The risk of biased assumptions is
still present though, since goals might also introduce
tacit assumptions. Let us take a look into an example
of an assumption in goals.
2.2 Efficiency is important
For the next assumption, let us take a look to our
limitations. We live in a competitive environment
with a limited lifespan, which means that efficiency
and reliability to attain our goals is of utmost im-
portance. If we were to consider only human-related
goals, then being able to predict the future to better
accomodate to it is important, but here we consider
more general goals: goals that may or may not imply
deadlines or resource scarcity.
Benchmarking
Consider two methods that we will call Faster and
Slower. They are challenged with the same cognitive
task and both solve the problem, yet Faster does it
quicker. Does it mean Faster is more intelligent than
Slower? Well, not necessarily. If we assign the goal to
be to solve the cognitive task, then both methods are
equally successful. Moreover, if the goal is to solve
the cognitive task as fast as possible, then again both
methods are successful because each methods arrives

2.3 Evolution is not intelligent JSAI Technical Report SIG-AGI-011-07
at the solution as fast as they can possibly be. It
is only when the goal is specified such as solve the
problem faster than Slower that Faster fulfils the goal
but Slower fails because it cannot be faster than it-
self. Thus, claming that Faster is more intelligent
than Slower implies a goal that compares both Faster
and Slower.
Substrate
Human intelligence and artificial intelligence are sup-
ported by different physical substrates with very dif-
ferent characteristics. This is directly related to, for
example, the speed of cognitive processing. Whereas
the frequency of neuronal interspike intervals (ISI) go
up to 10KHz [11], modern processors achieve easily
4GHz. This rises the question of whether even slower
intelligent entitities, yet even more parallel in nature
than the human brain, exist or are even possible. We
believe this question to be positive: evolutionary pro-
cesses rely on approximately 5 ·1030 prokaryote cells
[12].
Accepting this generalization of processing speeds
in intelligence is crucial to understand the next as-
sumption.
2.3 Evolution is not intelligent
Darwinian evolution is increasingly regarded as in-
telligent in the scientific community. Here we support
this view, but first we justify that avoiding mistakes
is not necessary for General Intelligence.
Aversion to mistakes
Mistake avoidance is another assumption commonly
found implicitly in the definition of goals. Mistakes
are associated to wasting resources, such that it is un-
desirable to have a system that makes mistakes. How-
ever, a more general view on intelligence is to allow
for systems that make mistakes and waste resources,
as long as the goal is fulfilled. If mistake avoidance
is necessary, then it should be explicitly stated in the
goal. For example, scientific research often implies
a large number of experimentation, trial and error.
Upon interesting results, previous mistakes no longer
weigh with the importance of the discovery.
Darwinian evolution is a process that continually
makes replication mistakes, yet the genetic code con-
tinues to evolve successfully. As opposed to human
intelligence, which has a limited lifespan and a lim-
ited patience, evolution does not have deadlines nor
needs to save resources by avoiding mistakes. As long
as there are no constraints or conflicting goals related
to repeatability or resource limits, making mistakes is
not relevant for attaining goals. Consequently, mis-
takes should not factor into General Intelligence.
Free will
Our next step is to make similarities between dar-
winian evolution and human intelligence to strengthen
our thesis on General Intelligence. We now show that
human free will, as theorized with two-stage models,
has strong similarities to natural selection, and by do-
ing so we conclude that either human free will does
not follow a two stage model or that evolution is en-
dowed with free will similar to that of human intellect.
Let us show evidence against the former and evidence
in support of the latter.
James [13] proposed a two stage model of free will
where choice is preceded by the production of alter-
natives to choose from. This model has become a ref-
erence position in free will, with many other authors
revolving around the idea of conceiving alternatives
and choosing among them. Many authors have pro-
posed similar two stage models for free will [14] after
James, and are all related to the process of chance
and choice.
Interestingly, darwinian evolution also follows a two
stage model. Mayr [15] claimed that evolution is a
two-step process that involves the production of new
individuals followed by the selection of the next gen-
eration. Genetic algorithms replicate natural evolu-
tion by assigning random mutations to a population
of candidate solutions followed by the selection of the
more successful ones. This similarity between free will
and evolution suggests that evolution is endowed with
free will, as would be expected from intelligent sys-
tems.
More recently, Simonton [16] suggested a relation
between free will and creativity in his review of ad-
vances in two-stage theories of creative problem solv-
ing that consist of blind variation followed by selective
retention. Creativity in biology is certainly present,
whilst genetic algorithms, which imitate the processes
of darwinian evolution, have been found to be more
creative than original thought by providing unexpected
and impredictable solutions to the problems they are

2.4 Intelligence runs on models JSAI Technical Report SIG-AGI-011-07
confronted to, sometimes taking advantage of unknown
bugs in simulation software [17].
In view of these arguments, human intelligence and
darwinian evolution appear to be deeply related by
two stage models.
Reinterpreting evolution
There are many more examples that defend that
evolution is intelligent. The products of darwinian
evolution have been explained throughout history by
recurring to an intelligent entity as Hume [18] de-
scribed in the Argument from Design on behalf of
the philosopher Cleanthes. Additionally, Orgel [19]
asserted that evolution is cleverer than you are in his
two Rules of Orgel. Lately, Fogel [20] argued that
evolution accomplishes learning by some form of ran-
dom search and the retention of these “ideas” in the
genetic code and wrote that intelligence is not the end
result of evolution; rather, it is woven into the process
itself.
There is enough evidence for regarding evolution as
intelligence that it becomes difficult to reject it as a
realization of General Intelligence. More importantly,
if evolution is reinterpreted as intelligent, General In-
telligence already has at its disposal the mechanisms
that drive darwinian evolution in the form of biology.
General Intelligence then becomes a matter of rec-
onciling evolution and human intellect (and possibly
Artificial Intelligence too) under a single framework,
e.g. [21].
2.4 Intelligence runs on models
The requirement of models for intelligence is one of
the most undisputed assumptions in artificial intelli-
gence. Models are convenient to make representations
of the world and speed up the process of finding the
correct path to a goal by avoiding mistakes and re-
ducing the resources required. Planning, reasoning,
language, knowledge representation, etc. are all abil-
ities that require the use of models. However, this
requirement poses too many restrictions on which sys-
tems are allowed to be intelligent. As we have seen
previously, it is possible to achieve goals without ef-
ficiency as long as efficiency is not part of the goal.
Moreover, darwinian evolution is capable of intelligent
design without the need of models. The conclusion is
that intelligent systems capable of handling represen-
tations of the world are only a subset of all systems
capable of showing purposeful behavior.
Human intelligence does have some limitations de-
rived from the use of models. Complex systems, chaotic
systems and basically any phenomenon that cannot
be decomposed in smaller parts for cognitive process-
ing are outside the limits of human understanding.
We are referring to complex systems, chaotic systems,
and all those systems whose internal entanglement is
so strong that cannot be conceived as interrelated in-
dependent subsystems. Typically, these phenomena
are treated like black boxes where features such as
intelligence is said to emerge somehow, avoiding the
need to justify the mechanisms further.
In contrast, darwinian evolution does not have these
limitations. Life is indeed a complex system of bio-
chemical processes where small variations can affect
many biological processes simultaneously, but that
does not stop evolution from improving life.
2.5 Intelligence emerges from agents
There is an important difference between human
intelligence and evolution that reveals yet another
rooted assumption. As opposed to human intelligence,
the processes that drive evolution are not tied to phys-
ical substrates, i.e. neither mutations nor natural se-
lection can be attributed to any embodied process.
There are also some situations in cognitive robotics
where the boundary of the agent is ambiguous and
subjective: polymorphic robots, swarm intelligence,
chaotic systems, damaged robots, etc. Our under-
standing of General Intelligence is that it should cope
with all these situations under a single framework.
We propose to remove any agent boundaries and
treat agent and environment as a single indivisible
unit when it comes to evaluate intelligence. This way,
a system is not intelligent by itself, but by all the
interactions of the system with itself and with the en-
vironment. For example, emergent intelligence and
synergetic systems are better understood by consid-
ering the whole, rather than the parts. Not only does
this method remove the necessity of defining agent
boundaries, thus removing ambiguities, but it also
supports swarm intelligence, embodied cognition and
evolutionary processes under a single framework.

2.6 Language is necessary JSAI Technical Report SIG-AGI-011-07
2.6 Language is necessary
For the final assumption, we rely on the previous
ones to suggests that natural language stands as a pil-
lar of General Intelligence because of the over-impor-
tance that human intelligence receives. Again, once
we accept that human intelligence does not have ex-
clusivity over General Intelligence, the relevance of
natural language lessens. Indeed, natural language is
necessary in social contexts to understand the state
and intentions of other people and it is based on as-
signing meaning to representations of the world. As
was shown in previous sections, not only the use of
models and representations is not required for General
Intelligence, but human intelligence is not necessarily
the best realization to imitate. Furthermore, if we
consider agents and environment as a single undivisi-
ble unit, natural language falls back to a complex way
of transmitting representations between internal com-
ponents of that single unit, in the same way that other
communication events not involving natural language
function. From computer systems using bitstreams,
to biology by means of hormones, neurotransmitters
and other molecules, there is a pletora of systems
transmitting information. These transmissions may
also be regarded as transmitting representations and
meaning.
This perspective unifies communication events whether
they use natural language or not, such as non-verbal
communication, which also conveys states of others,
and data transfer between computers. The bottom
line is that language in any form enables internal com-
ponents of a system to collaborate and attain com-
mon goals that otherwise would be imposible to reach.
These communication events ought to support intel-
ligence in any form, including intelligent entities that
do not make use of internal representations.
3 Discussion
This paper challenges some rooted assumptions in
the study of Intelligence that are barely questioned in
the scientific community:
1. Intelligence cannot be defined, but it requires a
definition to discuss it.
2. Efficiency in speed and optimization of resource
utilization are dispensable in General Intelligence.
Rather, they are assumed in the definition of the
goals.
3. Darwinian evolution is an intelligent entity. We
ought to learn from evolutionary processes and
generalize human intelligence and evolution into
a single theory of Intelligence.
4. Models and representations of the world are an
alternative approach to solve problems, but not
a universal method to solve all goals.
5. The distinction between agent and environment
is an adequate feature for human intelligence,
but not necessarily a general feature.
The key to advance in General Intelligence is to un-
derstand that the scientific community assigns human
intelligence a central role for General Intelligence and
that the capabilities of evolution are underrated. We
need to revise the requirements for General Intelli-
gence to avoid relying too much on human intelli-
gence, because it may be diverting research efforts
from an underlying theory of General Intelligence.
In analogy with the scientific revolution that took
place in astronomy when the geocentric model was re-
placed by the Copernican heliocentrism [22, spec. ch. 8],
General Intelligence needs a change of paradigm from
an anthropocentric model to a more integrative model
that does not underrate darwinian evolution. A change
of paradigm is justifiable because of the long history
of failures at achieving General Intelligence in Arti-
ficial Intelligence. It is time we start rethinking our
view on intelligence, even if that means that we walk
away from the intuitive understanding of intelligence.
As much as other sciences have proven that many ad-
vancements come from counterintuitive ideas, Arti-
ficial Intelligence still remains at its infancy with a
deep anthropocentric bias. Our mental abilities seem
to be modeled after the environment we live in and the
phenomena we are exposed to. This makes convenient
to think about separating the environment in differ-
ent entities, including ourselves and other agents, but
is not necessarily adapted to strictly follow the direc-
tions of a plausible theory of General Intelligence. Our
stake is that human optimizations for dealing with the
environment makes for a too complex realization of
General Intelligence to study it as a canonical repre-
sentative.
We have a great opportunity to boost General Intel-
ligence if we credit evolution with intelligence because
of the simplicity of the process, based on random mu-
tations and natural selection. This simplicity is much
explained by the lack of models, which does not im-
pede evolution to reach complex goals.

References JSAI Technical Report SIG-AGI-011-07
An important criticism to this paper is that accep-
tance of the assumptions presented requires having
an already biased view towards regarding evolution as
intelligent. Also, there are many areas that we have
not tackled. Consciousness and human intelligence
are intertwined, but it is not known if consciousness
is necessary for General Intelligence, which in that
case evolution might be related to consciousness as
well.
We hope that our perspective in General Intelli-
gence will help broaden the view on Intelligence.
References
[1] Marek Rosa, Jan Feyereisl, and The GoodAI Col-
lective. A Framework for Searching for General
Artificial Intelligence. 2016.
[2] Kenneth O Stanley and Risto Miikkulainen.
Evolving neural networks through augmenting
topologies. Evolutionary computation, 10(2):99–
127, jan 2002.
[3] David Silver, Julian Schrittwieser, and Others.
Mastering the game of Go without human knowl-
edge. Nature, 550(7676):354–359, 2017.
[4] R Pfeifer and C Scheier. Understanding intelli-
gence. MIT Press, Cambridge, 1999.
[5] Pat Langley. Cognitive Architectures and Gen-
eral Intelligent Systems. AI Magazine, 27(2),
2006.
[6] John E. Laird and Robert E. Wray. Cognitive
architecture requirements for achieving AGI. In
Proc. of the Third Conference on Artificial Gen-
eral Intelligence, pages 79–84, 2010.
[7] Rolf Pfeifer and Gabriel G´omez. Interacting with
the real world: Design principles for intelligent
systems. Artificial Life and Robotics, 9(1):1–6,
2005.
[8] Naoya Arakawa, Hiroshi Yamakwa, and Ryutaro
Ichise. An Overview of Artificial General Intel-
ligence as of 2014. Proceedings of the 7th In-
ternational Conference, AGI 2014, (August):1–6,
2014.
[9] Charles Spearman. ”General Intelligence,” Ob-
jectively Determined and Measured. The Amer-
ican Journal of Psychology, 15(2):201–292, 1904.
[10] Shane Legg and Marcus Hutter. Universal In-
telligence: A Definition of Machine Intelligence.
Minds and Machines, 17(4):391–444, 2007.
[11] D S Reich, F Mechler, K P Purpura, and J D Vic-
tor. Interspike intervals, receptive fields, and in-
formation encoding in primary visual cortex. The
Journal of Neuroscience, 20(5):1964–74, 2000.
[12] W. B. Whitman, D. C. Coleman, and W. J.
Wiebe. Prokaryotes: The unseen majority. Pro-
ceedings of the National Academy of Sciences,
95(12):6578–6583, 1998.
[13] William James. The Dilemma of Determinism.
Unit. Rev., 22:192–224, 1884.
[14] Lawrence LeShan and Henry Margenau. Ein-
stein’s space and Van Gogh’s sky. Harvester,
1983.
[15] Ernst Mayr. Toward a New Philosophy of Biol-
ogy. Belknap Press of Harvard University Press,
1988.
[16] Dean Keith Simonton. Creative thoughts as acts
of free will: A two-stage formal integration. Re-
view of General Psychology, 17(4):374–383, dec
2013.
[17] Joel Lehman, Jeff Clune, and Others. The Sur-
prising Creativity of Digital Evolution: A Collec-
tion of Anecdotes from the Evolutionary Compu-
tation and Artificial Life Research Communities.
pages 1–32, 2018.
[18] David Hume. Dialogues Concerning Natural Re-
ligion. 1779.
[19] Leslie E. Orgel. The origins of life: molecules
and natural selection. 1973.
[20] David B. Fogel. Evolutionary Computation - To-
ward a New Philosophy of Machine Intelligence.
2006.
[21] Francisco J. Arjonilla. A three-component cog-
nitive theory. Msc. thesis, Universiteit Utrecht,
2015.
[22] A. F. Chalmers. What Is This Thing Called Sci-
ence?, volume 73. 3rd edition, 1976.