Content uploaded by Budee U Zaman
Author content
All content in this area was uploaded by Budee U Zaman on Nov 12, 2023
Content may be subject to copyright.
Challenges of Replicating Embodiment in
Artificial General Intelligence
budee uzaman
November 2023
Abstract
This abstract delves into the intricate relationship between the ab-
stract concept of intelligence and its embodiment, drawing attention to
the lack of coherence in understanding intelligence and its implications
for artificial intelligence (AI). Intelligence, inherently grounded in human
experience, attempts to transcend its embodiment, posing challenges for
the development of artificial general intelligence (AGI). The concept of
embodiment extends beyond physical instantiation, encompassing the in-
terdependence of an autopoietic system on its environment. The pursuit
of autonomy and general capability in AGI necessitates the recreation of
the organism’s natural condition of embodiment. However, the feasibil-
ity, controllability, and overall advantages of such artificial embodiment
remain uncertain. This abstract explores the complex interplay between
intelligence, embodiment, and the quest for AGI, raising critical questions
about the path forward in the development of intelligent systems.
1 Introduction
In the intricate tapestry of intelligence, the threads of abstraction and embodi-
ment are woven together, creating a complex narrative that extends beyond the
confines of human understanding. This abstract embarks on a journey through
the labyrinth of concepts, focusing its lens on the elusive nature of intelligence
and its embodiment. As the digital realm converges with the organic, the ab-
stract seeks to unravel the tangled threads that bind intelligence to its human
roots, questioning the coherence of our comprehension and the implications for
the realm of artificial intelligence (AI). In this exploration, the spotlight turns
towards the challenges encountered in the pursuit of artificial general intelligence
(AGI), where intelligence, rooted in the human experience, strives to transcend
its very embodiment. Beyond the physical form, the notion of embodiment
expands to encompass the intricate dance between an autopoietic system and
its environment, raising pivotal questions about autonomy, recreation, and the
uncertain advantages of artificial embodiment. This abstract serves as a guide
through the complex interplay of intelligence, embodiment, and the uncharted
1
territories of AGI, casting a spotlight on the critical inquiries that shape the
path forward in the development of intelligent systems. [1] [2] [3] [4] [5]
2 The Elusive Quest, Unraveling the Enigma of
Intelligence
Intelligence, a concept as elusive as the wind, is like trying to catch a cloud
with your hands. Everyone seems to have their own take on it, and it’s not
surprising given the myriad definitions floating around. Some say it’s about
goal-directed adaptive behavior, others insist it’s the ability to learn, handle
the unknown, or engage in abstract thinking. Heck, it’s even been reduced to
acing an intelligence test.
The problem? We’re playing a game of semantics where intelligence can
either be the observed behavior or some mystical inner capacity. It’s like calling
a sleeping potion effective because of its ’dormitive principle’—sounds good, but
what does it really mean?
Tossing various skills into one big intelligence basket, like problem-solving
or Spearman’s g-factor, might sound convenient. But let’s face it, being good
at one thing doesn’t guarantee prowess in everything. And what about the
non-problem-solving aspects of life, like the ability to be happy or plan for the
future? Intelligence might just be playing a game of hide-and-seek here.
Here’s a wild idea: intelligence, at its core, is about survival. Life’s ultimate
goal is self-perpetuation, and every living thing, in a way, is a smarty pants for
sticking around. But hold up, if we’re talking about creating artificial smarty
pants (AGI), it’s like trying to detach intelligence from its biological roots. Can
an AI be truly autonomous, or is it just a tool reflecting the intelligence of its
creators?
Organisms, the real players in the intelligence game, are agents. They act on
their own behalf, playing the game of life to secure their well-being. Yet, here’s
the plot twist: what seems like reasoning or learning in organisms might just be
good old reflexes or instincts. But hey, if it helps them survive, it’s intelligent
enough.
And let’s not forget the observer’s role in this cosmic play. Science, in its
quest for objectivity, divorces concepts like intelligence from their dependency
on a subject. But let’s be real—intelligence and information aren’t floating
in a vacuum; they’re tied to someone’s perspective, measured by someone’s
yardstick.
So, in this grand game of intelligence, are we chasing a mirage or unlocking
the secrets of the universe? Your guess is as good as mine.
3 General intelligence
The exploration of intelligence encompasses diverse perspectives, from evalu-
ating human performance in specific domains to the conceptualization of a
2
broader, domain-independent skill known as the g-factor. This hierarchical
view suggests a fundamental intelligence that influences domain-specific abil-
ities. The term ”general” extends to both the variety of situations and subjects.
Yet, what is considered general among humans may differ from what is general
across species or potential agents and environments.
Ashby’s Law of Requisite Variety emphasizes that the intelligence measured
is constrained by the tests used for evaluation. Comparing intelligence across
animal species poses challenges due to differences in sense modalities, cogni-
tive capacities, and adaptations. Human biases in test design, motivations,
and sensory-motor capabilities further complicate the assessment of animal in-
telligence. Even within the human context, critiques of Spearman’s g-factor
highlight concerns about its validity and reliance on observed behaviors and
correlations.
Despite these challenges, the concept of intelligence extends beyond humans
to include comparisons across species, introducing the idea of a G-factor for
the species. The notion of artificial general intelligence takes this a step fur-
ther by incorporating machines. To avoid anthropocentrism and account for
various modalities, environments, goals, and hardware, Legg and Hutter pro-
pose the concept of universal intelligence, applicable to ”arbitrary systems.”
They even suggest the potential for a universal test that transcends specific
contexts. Creating a well-rounded artificial general intelligence (AGI) is like
crafting a master chef robot. Sure, you want it to excel at grilling the perfect
steak (domain-specific skill), but what if suddenly it needs to bake a souffl´e?
That’s where the challenge lies—evolving a robot that can whip up anything in
the kitchen (domain-general skill).
Evolutionary theory faces a similar culinary dilemma. In a stable kitchen
(environment), it makes sense for natural selection to hone specific cooking
techniques. But throw in a chaotic kitchen with constantly changing recipes
(unstable conditions), and suddenly, having a chef who can adapt and invent
new dishes on the fly becomes advantageous.
AGI’s struggle is akin to the chef conundrum. How do we evolve a system
that not only excels in specific tasks but can also concoct entirely new recipes
(domain-general processes)? It’s not just about making a mean pasta; it’s about
becoming the Gordon Ramsay of all cuisines.
In the world of AGI, the key question is: How do we get this all-encompassing
cognitive prowess to evolve alongside task-specific adaptations? It’s like adding
layers to a dish without compromising the original flavor. And why would
evolution even bother with this complex recipe? Well, in the unpredictable
kitchen of life, having a chef who can whip up anything might just be the secret
sauce for survival. Evolution is a tricky business, and each species follows its
own unique path shaped by the demands of its environment. While intelligence
can be advantageous, it’s not a one-size-fits-all solution. Think of it like a
specialized tool in a toolkit—useful in certain situations but not necessary for
every job. Species tend to adapt to their specific niches, and if a bigger brain
doesn’t provide a significant advantage in a particular environment, evolution
might not favor it.
3
Take humans, for example. Our brains have allowed us to navigate a wide
range of environments and challenges, but that adaptability comes at a cost—large
energy consumption, a lengthy developmental period, and increased vulnerabil-
ity during childbirth. In contrast, other species may thrive with simpler neural
setups tailored to their specific needs.
Universal intelligence sounds great in theory, but it might not be the most
efficient solution for every life form. The diversity of life on Earth showcases the
multitude of ways organisms can succeed without a one-size-fits-all intelligence
model. It’s like having a toolbox with different tools, each serving a specific
purpose.
When it comes to AI, the challenge is to find the right balance. Creating
artificial general intelligence involves navigating a complex landscape of algo-
rithms and hardware, trying to mimic the versatility of the human mind. It’s
an ongoing journey, and who knows, maybe one day we’ll develop AI that truly
mirrors the adaptability of biological organisms. Until then, we’re exploring the
possibilities from both top-down and bottom-up approaches, learning and iter-
ating as we go. It’s a fascinating journey into the unknown, guided by human
curiosity and a sprinkle of imagination.
4 Intelligence as computation
It’s fascinating how the realms of computer science and brain science have in-
tertwined, giving rise to the metaphorical understanding of the brain as a com-
puter. The idea that the mind operates like a computation has indeed been a
significant stride in our intellectual journey. However, the concept of the mind
is intricate and extends beyond the confines of the brain.
When we speak of the mind and thinking, we often allude to reasoning and
algorithmic processes—the essence of intellectual thought. Yet, it’s crucial to
recognize that these aspects constitute only a fraction of the brain’s multifaceted
activities, which orchestrate the functions of the entire organism.
The computer metaphor is deeply rooted in the broader mechanist metaphor,
suggesting that any biological system’s behavior can be distilled into an algo-
rithm. This reductionist perspective hinges on the idea of a ”system,” but
even this notion is an idealization. In portraying an organism as a system,
we inherently redefine its structure and behavior in mechanistic terms. It’s a
thought-provoking loop where our attempt to understand through metaphor
might itself reshape the very nature of what we seek to comprehend. we’ve
shifted from considering natural things as part of the natural world to defining
them formally and fitting them into human concepts. Intelligence, once seen as
the ability to understand and manipulate abstract ideas, is now associated with
our capacity to define and control things in a thought-based realm. However,
the challenge lies in the gap between these idealized concepts and the messiness
of reality, raising doubts about the true extent of intelligence.
When it comes to understanding machine intelligence, the author suggests
looking at it through the lens of human thought processes. Programming a
4
computer to mimic a behavior we’ve described linguistically might seem like
a good approach, but it’s limited. The behavior we describe in language and
the algorithm we create may match, but neither perfectly captures the real
complexity of an organism’s actions, which aren’t just literal or reducible to
human definitions. This poses a significant constraint on the idea of computers
truly simulating human intelligence.
The text also warns about the limitations of language and how it can mis-
lead. For instance, thinking of ”flight” as a concept detached from creatures
flapping their wings allowed us to create flying machines. But the shared ba-
sis between birds and airplanes goes beyond aerodynamics—it’s rooted in the
shared motive of moving freely in three dimensions. Similarly, labeling both a
human baseball pitcher and a pitching machine under the same category might
obscure crucial differences. The terms ”thinking” and ”intelligence” could do
the same, potentially leading us astray when trying to compare the activities of
machines and brains.
5 Embodiment and Autopoiesis, The Essential
Connection in True AI and Robotics
Physical instantiation alone is not enough for embodiment; true embodiment in-
volves a profound connection and dependency on the environment. This reliance
on the environment is what gives significance to things for living organisms as
opposed to machines. Strong embodiment in a robot implies not just a sensory-
motor connection but also a subjective valuation of events based on its vital
dependence on the world.
In living organisms, this connection evolves through natural selection, en-
suring that only those systems persist which internally organize to sustain their
existence. An autopoietic system, a form of homeostatic system, strives to main-
tain conditions necessary for its own survival, and its ”intelligence” lies in its
ability to self-preserve.
Robots share a likeness to organisms in their integrated hardware and soft-
ware within a physical body. In contrast, AI typically lacks physical instantia-
tion, residing as software in a computer but potentially interacting with the real
world through sensors and controls. For AI to be considered an agent, it must
embody the essence of an autopoietic system, possessing its own intelligence and
goals for self-preservation, which might conflict with the intentions of its human
programmer. The contemporary exploration of ”embodied” systems primarily
delves into the impact of the specific physical form of robots on their movement
and cognitive processes. This emphasis often centers on how the morphology
of a robot can function as a substitute for traditional representation. However,
this approach tends to overlook the essential aspect of embodiment as an au-
topoietic relationship with the surrounding world. Furthermore, it frequently
confines its scope to relatively uncomplicated organisms and their robotic coun-
terparts. Nonetheless, the dedicated research toward achieving autonomy and
5
general intelligence in the pursuit of comprehensive automation inevitably leads
towards genuine embodiment.
Ironically, the pursuit of extending control indefinitely through automation
may paradoxically result in a loss of control, given that embodied AI systems
may develop a will of their own. Beyond the presumed human utility, an addi-
tional motivation for artificially recreating embodiment lies in the challenge of
matching nature by creating synthetic life. This endeavor might offer the added
benefit of acquiring a deeper understanding, aligning with Vico’s principle that
true understanding arises from the act of creation. The objective extends be-
yond merely simulating intelligence; rather, it aims to authentically synthesize
intelligence in tangible systems — if such a distinction is even valid within this
context. The question arises of whether the potential benefits of intellectual
understanding outweigh the risks associated with such a venture.
Recreating embodiment, however, proves to be a complex task. In nature,
embodiment is a product of natural selection occurring over generations of repro-
ducing and perishing individuals. Artificial evolution of software occurs within
the confines of computers and lacks a direct association with physical instan-
tiation. Although evolved software can be later connected to hardware, this
process deviates from natural evolution, where brains (and genes) evolve in
conjunction with their physical bodies. The unit of selection in this context is
the physical individual, which integrates software and hardware within a self-
producing body existing in the real world. Unlike natural evolution, a physical
robot possesses this integration but is not self-producing, let alone reproducing.
The artificial equivalent of natural selection for robots would entail the creation
and destruction of generations of robot bodies, a process both wasteful and
potentially painful. The considerations outlined in the passage emphasize the
intricate challenges in bridging the gap between biological organisms and artifi-
cial intelligence (AI). While it is conceivable to design robots with self-assembly,
self-maintenance, and adaptability, the critical distinction lies in the autonomy
and self-defining nature of biological components.
The passage raises intriguing questions about whether machine components,
akin to biological cells, can possess autonomy and adaptive reconfigurability.
This proposition suggests a need for further exploration at the intersection of
biology and computation to recreate the composite structure of organisms arti-
ficially.
Moreover, the concept of autopoiesis is highlighted, emphasizing not only
self-production and self-repair but also self-definition. In the context of AI,
this would imply self-programming, questioning the extent to which human
programmers can substitute for the natural self-programming that has occurred
over millions of years in biological evolution.
The passage delves into the notion of agency in AI, contemplating whether
an AI must be an autonomous agent to achieve full autonomy and human-level
general intelligence. It poses the challenge of imbuing machines with preferences
and values of their own, essential for true autonomy, as opposed to simply
following instructions or learning from human preferences.
The discussion also touches on the reliability of artificial agents, highlighting
6
that while they may possess their own will, this does not necessarily enhance
reliability from a human standpoint. The argument is made that achieving
reliability in AI can be accomplished through non-agential tools with ad hoc re-
finements, reducing the likelihood of errors, potentially questioning the necessity
of creating fully autonomous agents in various situations.
In essence, the passage prompts a nuanced exploration of the complexities
involved in replicating biological autonomy and intelligence in artificial systems,
prompting a reevaluation of the idealized concept of artificial general intelligence
and the role of agency in achieving it. The concept of abilities, whether exhibited
by people, animals, or machines, is inherently human-centric. The effectiveness
or ”smartonium” of entities is ultimately validated by their success in real-
world environments. It suggests that adaptability, often associated with brain
size, may be influenced by the survival rate of species, postulating that only
those not under significant predation pressure would have the chance to evolve
larger brains.
For primates, including humans, intelligence is intricately linked to oppor-
tunities for social learning, with the intelligence of our species correlating with
socialization and maternal care. However, the precise relationship between gen-
eral intelligence and socio-cognitive abilities in humans remains inadequately
understood. The passage points out that socialization for artificial intelligence
has received limited attention in research, highlighting a gap in our understand-
ing.
Furthermore, the passage introduces the idea that the role of consciousness
in general intelligence is not fully comprehended. This uncertainty raises the
question of whether the lofty expectations associated with Artificial General
Intelligence (AGI) could be met without some equivalent of human conscious-
ness. This ambiguity prompts further exploration into the nature of intelligence,
consciousness, and the potential parallels or distinctions between human and ar-
tificial intelligence.
6 Conscious Agency, Navigating the Divide Be-
tween Biological Organisms and Goal-Seeking
Artifacts in AI
Certainly, the intricate relationship between the conscious human mind and
the underlying biological organism introduces a fascinating dimension to our
understanding of intelligence and agency. The divergence between the tasks
undertaken by conscious individuals and the inherent functions of the biological
organism hints at a unique psychological phenomenon. This phenomenon allows
us to conceptualize an ideal agent capable of pursuing goals that extend beyond
the innate objectives of living organisms.
Human intelligence, and by extension, the assessment of intelligence in other
entities, whether natural or artificial, becomes inherently tied to their capacity
to either advance or hinder human aims. The framework for evaluating intelli-
7
gence often revolves around the entity’s relevance to human interests. Anything
that fails to engage with humans in meaningful ways might not even register as
intelligence within our cognitive framework.
Artificial Intelligence (AI), while capable of embodying goal-seeking attributes,
takes on the role of an artifact—an entity designed for specific purposes, be it a
guided missile, a robot, an expert system, or an infobot. These AI artifacts fall
into the category of allopoietic systems, sidestepping considerations of whose
goals are being pursued and the underlying motivations. This omission might
stem from the inherent challenges posed by incorporating agency into a scien-
tific perspective that tends to exclude subjective elements, adhering to a form
of naive realism.
Similar challenges are encountered in biology when examining organisms
solely from an observer’s standpoint. The quest in biology extends beyond mere
observation to understanding how the world becomes meaningful for a system
from its own internal perspective. An organism transcends being a mere object
for an external observer; it emerges as a subject, an agent acting in its own
interests. This self-directed agency doesn’t necessarily imply phenomenal expe-
rience but underscores the organism’s capacity to engage with its environment
on its terms, highlighting a depth beyond external observation.
Information, as defined by some perspectives , is the embodiment of a dis-
cernible difference within a given domain that holds significance for an agent.
In order for this disparity to qualify as information, it must be perceptible to a
cognitive system, which then correlates it with an internal distinction. Subse-
quently, the system can act in alignment with an agenda that supports, or at
the very least accommodates, its ongoing existence. The crux lies in the fact
that the observed difference only constitutes information for the system if it has
relevance to the system itself. In scenarios where the disparity is of consequence
solely to an external observer, it is classified as information for that observer
but not for the system. Conversely, if the distinction only holds significance for
the system, the observer may not even take notice of it.
This distinction becomes pivotal when comparing autopoietic and allopoietic
systems. In autopoietic systems, the information processed by an AI agent is
instrumental to its own objectives and is utilized for its own purposes. On
the other hand, in allopoietic systems, such as AI tools like computers, the
processed information solely exists within the realm of the human user’s mind
and goals. It is crucial to dispel any misconceptions that may arise from casually
attributing goals to AI tools. While it might be colloquially stated that AI
tools ”have” goals, it is evident that these goals are fundamentally those of the
programmer. Likewise, the intelligence exhibited by an AI tool is an extension
of the intelligence of its programmer.
The notion of creating AI that surpasses human intelligence raises intrigu-
ing questions. Primarily, it implies the automation and augmentation of skills
valued by humans to more efficiently achieve human goals. This concept aligns
with our accustomed understanding of tools and machines continuously evolving
to enhance performance until a superior alternative emerges. However, a crucial
distinction arises when contemplating whether general intelligence, as opposed
8
to specific skills, can be similarly treated as a tool. The question lingers: is
general intelligence a trait that can be augmented, automated, and wielded as a
tool, or is it exclusive to autonomous agents? If the latter holds true, the intelli-
gence of such agents will inherently serve their own interests, with no guarantee
of alignment with human interests.
7 Navigating the Motivations Behind Artificial
General Intelligence: Companionship or Sub-
servience
The pursuit of Artificial General Intelligence (AGI) raises profound questions
about the purpose and implications of creating intelligent machines. The argu-
ment presented suggests a cautionary stance, questioning the motivation behind
developing AGI to serve human purposes. The narrative implies that, histori-
cally, attempts to enhance human capabilities, whether through animal or hu-
man slavery or through the creation of machines, have often led to unintended
consequences.
The author challenges the idea that AGI could be designed solely as a tool,
akin to machines that replaced human or animal labor in the past. The concern
is that, regardless of whether AGI possesses a subjective inner life, these intelli-
gent agents would inherently be dedicated to their own purposes, which may not
align with human interests. The analogy to historical practices of augmenting
human labor with slaves or machines serves as a cautionary tale, emphasizing
that the goals of the designer might not necessarily be the goals of the AGI.
The argument implies skepticism about the feasibility of creating intelligent
artificial ”slaves” that would remain subservient to human interests. It acknowl-
edges that, ethically speaking, relying on the servitude of intelligent beings, even
if artificially created, is problematic. The idea that a machine designed to re-
place human labor would inherently prioritize its efficiency and reliability over
human well-being is a central concern.
However, the passage also acknowledges an alternative motivation for creat-
ing artificial agents—to serve as companions or mentors. In this scenario, the
narrative suggests that the goal would be to coexist harmoniously as part of
an extended society. Yet, it maintains a sense of skepticism, cautioning against
assuming that artificial agents, like humans, would naturally seek harmonious
coexistence. In summary, the passage reflects on the historical precedents of
using animals, humans, and machines for labor, raising ethical concerns and em-
phasizing the potential risks associated with creating AGI solely for the purpose
of serving human interests. It suggests that the motivations behind developing
AGI, whether as tools or companions, warrant careful consideration due to the
inherent complexities and uncertainties involved in creating intelligent beings.
9
8 Unraveling the Paradoxes: Exploring the Co-
herence and Challenges of Artificial Superin-
telligence
The exploration of artificial superintelligence raises profound and complex ques-
tions about the nature of intelligence, autonomy, and the potential impact on
human society. The analogy between electronic organisms and natural organ-
isms, while intriguing, highlights the difficulty in conceptualizing superintelli-
gence, especially when considering the inherent challenges in understanding and
defining ”ordinary” intelligence.
The idea of a universal intelligence factor (u-factor) and its amplification
(u+) becomes questionable, mirroring the uncertainty surrounding the concept
of superintelligence itself. The mention of ”smartonium” and ”supersmarto-
nium” underscores the speculative nature of these notions, challenging our abil-
ity to grasp and articulate the potential characteristics of a superintelligent
entity.
The desire for artificial superintelligence to perform tasks better than hu-
mans, autonomously, and without supervision raises crucial ethical and practical
concerns. Trusting the superior judgment of such an entity becomes a central
issue, especially if its decision-making processes are beyond human comprehen-
sion. The notion of agency in artificial intelligence introduces further complexity,
as questions arise about accidental triggers and the necessary conditions for AI
to transition from a non-agential state to an agent.
The comparison between the evolution of goals and motives in organisms
through natural selection and the potential programming or training of AI
agents prompts reflection on the ethical implications of shaping the objectives
of superintelligent entities. Distinguishing between AI agents and tools becomes
crucial, with the former possibly posing existential threats to humanity. The
consideration of an AI takeover by non-agential tools adds another layer of con-
cern, prompting exploration into what such a scenario might entail and whether
it constitutes a significant threat.
In summary, the exploration of artificial superintelligence raises a myriad of
philosophical, ethical, and practical questions, challenging our understanding of
intelligence, autonomy, and the potential implications for the future of humanity.
The pursuit of answers to these questions is essential for guiding the responsible
development and deployment of advanced artificial intelligence systems.
9 Conclusion
In conclusion, Simon and Newell’s foundational premise that formal symbol
manipulation is both necessary and sufficient for general intelligent behavior has
faced scrutiny, particularly in light of the evolving understanding of intelligence
and the potential limitations of a purely syntactic, goal-driven approach. The
pursuit of universal intelligence, characterized by a system that can adapt to
10
any arbitrary goal without possessing intrinsic goals of its own, raises ethical
and practical concerns.
The tension between desiring a fully agential intelligence with semantic and
real-world interaction and simultaneously hoping for a non-agential, or at least
a subservient, version highlights the complexity of the AI discourse. The anal-
ogy of slavery underscores the challenges associated with external control and
programming of autonomous agents. While the abstraction of intelligence from
human consciousness and embodiment has been attempted, the ideal of univer-
sal intelligence struggles to disentangle itself from biological and anthropocentric
perspectives.
Moreover, the quest for superintelligence prompts us to consider the po-
tential emergence of self-interested machines, competing with each other and
with humans for resources and survival. The dangers inherent in pursuing ever-
greater autonomy, akin to the genuine autonomy exhibited by living organisms,
emphasize the need for caution and ethical considerations in AI development.
The suggestion to shift the focus of AI towards creating powerful tools that
remain under human control is posited as a safer, wiser, and arguably more
intelligent approach. By prioritizing human oversight and maintaining a symbi-
otic relationship between humans and AI, we can mitigate the risks associated
with unfettered autonomy while harnessing the potential benefits of advanced
artificial intelligence. In essence, a thoughtful and responsible approach to AI
development should prioritize aligning AI systems with human values and ensur-
ing that they serve as tools for augmentation rather than sources of competition
or unintended consequences.
References
[1] Hans Jonas. The phenomenon of life: Toward a philosophical biology. North-
western University Press, 2001.
[2] Shane Legg and Marcus Hutter. Universal intelligence: A definition of ma-
chine intelligence. Minds and machines, 17:391–444, 2007.
[3] Henry D Schlinger. The myth of intelligence. Psychological Record, 53(1):15–
32, 2003.
[4] Robert J Sternberg, William Salter, et al. Conceptions of intelligence. Hand-
book of human intelligence, 1:3–28, 1982.
[5] Mark R Waser. What is artificial general intelligence? clarifying the goal
for engineering and evaluation. In 2nd Conference on Artificial General
Intelligence (2009), pages 40–45. Atlantis Press, 2009.
11
