Toward Beneficial Human-Level AI… and Beyond

Abstract

This paper considers ethical, philosophical, and technical topics related to achieving beneficial human-level AI and superintelligence. Human-level AI need not be human-identical: The concept of self-preservation could be quite different for a human-level AI, and an AI system could be willing to sacrifice itself to save human life. Artificial consciousness need not be equivalent to human consciousness, and there need not be an ethical problem in switching off a purely symbolic artificial consciousness. The possibility of achieving superintelligence is discussed, including potential for 'conceptual gulfs' with humans, which may be bridged. Completeness conjectures are given for the 'TalaMind' approach to emulate human intelligence, and for the ability of human intelligence to understand the universe. The possibility and nature of strong vs. weak superintelligence are discussed. Two paths to superintelligence are described: The first path could be catastrophically harmful to humanity and life in general, perhaps leading to extinction events. The second path should improve our ability to achieve beneficial superintelligence. Human-level AI and superintelligence may be necessary for the survival and prosperity of humanity .

Full text

Toward Beneficial Human-Level AI… and Beyond
Philip C. Jackson, Jr.
TalaMind LLC
www.talamind.com
dr.phil.jackson@talamind.com
Abstract
This paper considers ethical, philosophical, and technical
topics related to achieving beneficial human-level AI and
superintelligence. Human-level AI need not be human-
identical: The concept of self-preservation could be quite
different for a human-level AI, and an AI system could be
willing to sacrifice itself to save human life. Artificial con-
sciousness need not be equivalent to human consciousness,
and there need not be an ethical problem in switching off a
purely symbolic artificial consciousness. The possibility of
achieving superintelligence is discussed, including potential
for ‘conceptual gulfs’ with humans, which may be bridged.
Completeness conjectures are given for the ‘TalaMind’ ap-
proach to emulate human intelligence, and for the ability of
human intelligence to understand the universe. The possibil-
ity and nature of strong vs. weak superintelligence are dis-
cussed. Two paths to superintelligence are described: The
first path could be catastrophically harmful to humanity and
life in general, perhaps leading to extinction events. The
second path should improve our ability to achieve beneficial
superintelligence. Human-level AI and superintelligence
may be necessary for the survival and prosperity of humani-
ty.
The Future of Humanity: Ethics and AI
Some potential consequences of general artificial intelli-
gence were outlined in (Jackson 1974). Two possibilities
for the “harvest of AI” were briefly discussed: A world
with the machine as dictator, and a world with “well-
natured machines” having enormous benefits to humanity.
Relatively recent work on ‘artificial general intelligence’
(Goertzel and Pennachin 2007) has included substantive
research on AGI’s potential consequences for humanity:
Bostrom, Omohundro, Tegmark, Yudkowsky and others
have discussed future scenarios in which AGI could lead to
superintelligent systems with good or bad conduct toward
humanity. AGI may be necessary for the survival and
Copyright © 2018, TalaMind LLC (www.talamind.com). All rights re-
served. This paper is published in the 2018 AAAI Spring Symposium
Series Technical Reports, SS-18-01, pp.48-53.
prosperity of humanity but if AGI is not developed very
carefully it could lead to the extinction of humanity.
Ethics is the branch of philosophy that studies concepts
of right and wrong (good and bad) conduct. Until recently
ethics has only needed to focus on conduct by humans.
Ethics and AI research now intersect regarding concepts of
right and wrong conduct by intelligent machines, and right
and wrong conduct in human applications of intelligent
machines.
This is a challenge for AI scientists because ethical con-
cepts of right and wrong go beyond simple questions of
whether factual or theoretical knowledge is true or false, or
whether problem solving behavior is successful or unsuc-
cessful. In general, we cannot expect concepts of right and
wrong conduct to be easily understood by machines. It can
be a challenge for humans to distinguish these concepts
sometimes.
Yet if the survival and prosperity of humanity are at
stake, we are obligated to accept the challenge. Hence this
paper will consider ethical, philosophical, and technical
topics related to achieving beneficial human-level AI and
superintelligence. The term ‘beneficial’ in this context does
not seem to have any rigorous, agreed-upon definition. It
will be used broadly to refer to consequences that are posi-
tive for humanity and biological life1 in general.
The Possibility of Human-Level AI
A first question is whether human-level AI is even possi-
ble: The ‘TalaMind thesis’ (Jackson 2014) presents a re-
search approach toward human-level artificial intelligence,
which will support this paper’s discussion of human-level
AI’s implications for the future of humanity.
The thesis endeavors to address all the major theoretical
issues and objections that might be raised against its ap-
proach, or against the possibility of achieving human-level
1 Life based on DNA that has been developed by evolution. (Cf. Tegmark
2017).

AI in principle. No insurmountable objections are identi-
fied, and arguments refuting several objections are present-
ed. Thesis section 7.8 gives reasons in favor of the Tala-
Mind approach over other approaches to human-level AI.
The approach involves developing an AI system using a
language of thought (called ‘Tala’) based on the uncon-
strained syntax of a natural language; designing the system
as an ‘intelligence kernel’, i.e. a collection of concepts that
can create and modify concepts, expressed in the language
of thought, to behave intelligently in an environment; and
using methods from cognitive linguistics such as mental
spaces and conceptual blends for multiple levels of mental
representation and computation.
Proposing a design inspection alternative to the Turing
Test, the thesis discusses ‘higher-level mentalities’ of hu-
man intelligence, which include natural language under-
standing, higher-level learning, meta-cognition, imagina-
tion, and artificial consciousness.
‘Higher-level learning’ refers collectively to forms of
learning required for human-level intelligence such as
learning by creating explanations and testing predictions
about new domains based on analogies and metaphors with
previously known domains, reasoning about ways to debug
and improve behaviors and methods, learning and inven-
tion of natural languages and language games, learning or
inventing new representations, and in general, self-
development of new ways of thinking. The phrase ‘higher-
level learning’ is used to distinguish these from previous
research on machine learning. (Cf. Valiant 2013; Goertzel
and Monroe 2017)
The thesis discusses an architecture called TalaMind for
design of systems following its approach. The architecture
is open, e.g. permitting predicate calculus and conceptual
graphs in addition to the Tala language, and permitting
deep neural nets and other methods for machine learning.
The thesis describes the design of a prototype demon-
stration system, and discusses processing in the system that
illustrates the potential of the research approach to achieve
human-level AI.
Of course, the thesis does not claim to actually achieve
human-level AI. It only presents a theoretical direction that
may eventually reach this goal, and identifies areas for
future AI research to further develop the approach. These
include areas previously studied by others which were out-
side the scope of the thesis, such as ontology, common
sense knowledge, spatial reasoning and visualization, etc.
The TalaMind approach is similar though not identical
to the ‘deliberative general intelligence’ approach proposed
by (Yudkowsky 2007), as discussed in (Jackson 2014,
§2.3.3.5). The architectural diagrams for human-like gen-
eral intelligence given by (Goertzel, Iklé, and Wigmore
2012) may be considered as design aspects for TalaMind.
Human-Level AI ≠ Human-Identical AI
The TalaMind thesis gives reasons why the Turing Test
does not serve as a good definition of the goal we are try-
ing to achieve, human-level AI. In particular, the Turing
Test conflates human-level intelligence with human-
identical intelligence, i.e. intelligence indistinguishable
from humans. This is important because in seeking to
achieve human-level AI we need not seek to replicate hu-
man thinking. Human-level AI can be ‘human-like’ with-
out being human-identical. (Jackson 2014, §2.1.1)
In particular for beneficial AI, the concept of self-
preservation could be quite different for a human-level AI
than it is for a human. A human-level AI could periodically
backup its memory, and if it were physically destroyed, it
could be reconstructed and its memory restored to the
backup point. So even if it had a goal for self-preservation,
a human-level AI might not give that goal the same im-
portance a human being does. It might be more concerned
about protection of the technical infrastructure for the
backup system, which might include the cloud, and by ex-
tension, civilization in general.
A human-level AI could understand that humans cannot
backup and restore their minds, and regenerate their bodies
if they die, at least with present technologies. It could un-
derstand that self-preservation is more important for hu-
mans, than for AI systems. The AI system could be willing
to sacrifice itself to save human life, especially knowing
that as an artificial system it could be restored.
Artificial Consciousness
The TalaMind thesis accepts the objection by some AI
skeptics that a system which is not aware of what it is do-
ing, and does not have some awareness of itself cannot be
considered to have human-level intelligence. The perspec-
tive of the thesis is that it is both necessary and possible for
a system to demonstrate at least some aspects of con-
sciousness, to achieve human-level AI. However, the thesis
does not claim AI systems will achieve the subjective ex-
perience humans have of consciousness.
The thesis adapts the “axioms of being conscious” pro-
posed by Aleksander and Morton (2007) for research on
artificial consciousness. To claim a system achieves artifi-
cial consciousness it should demonstrate:
Observation of an external environment.
Observation of itself in relation to the external environ-
ment.
Observation of internal thoughts.
Observation of time: of the present, the past, and poten-
tial futures.
Observation of hypothetical or imaginative thoughts.

Reflective observation: Observation of having observa-
tions.
To observe these things, a TalaMind system should sup-
port representations of them, and support processing such
representations. The TalaMind prototype illustrates how a
TalaMind architecture could support artificial conscious-
ness.
Symbolic Artificial Consciousness
≠ Human Consciousness
The axioms of artificial consciousness can be implemented
with symbolic representations and symbolic processing, as
illustrated in the TalaMind prototype. The human first-
person subjective experience of consciousness is much
richer and more complex. Achieving human-level AI may
not require achieving human-identical consciousness in an
AI system.
This is important to note because some authors seem to
assume artificial consciousness will be equivalent to hu-
man consciousness, and assume a system with artificial
consciousness should automatically have the same moral
status and legal protections as a human being, so that
switching off the system could be immoral or illegal. Some
even suggest that if a system simulates consciousness with-
in itself, and then terminates the simulation, the system
may have performed a ‘mind crime’. (Bostrom 2014)
Such suggestions are at best philosophical, and at worst
Orwellian, if a system with symbolic artificial conscious-
ness does not have any subjective experiences approaching
human consciousness. Switching off such a system is not
worse than switching off any computer that performs sym-
bolic processing. Whether it is right or wrong to stop such
a system depends on whether its symbolic processing
would cause actions that affect human lives and biological
life in general. This may be a simple or complex ethical
decision, depending on whether the actions would be
harmful or beneficial, or neither, or a combination of both.
Further, to support reasoning about potential future
events, and counterfactual reasoning about past and present
events, a system may need to simulate what other intelli-
gent systems and people may think or do, and then termi-
nate its simulations. The TalaMind thesis (Jackson 2014)
uses the term ‘nested conceptual simulation’ to refer to an
agent’s conceptual processing of hypothetical scenarios,
with possible branching of scenarios based on alternative
events, such as choices of simulated agents. This amounts
to a Theory of Mind capability within a TalaMind architec-
ture, i.e. the ability of an AI system to consider itself and
other systems or people as having minds with beliefs,
goals, etc. Such simulations will be necessary for human-
level AI, and should not be considered mind crimes.
For the same reason, relying on robots with such limited,
symbolic artificial consciousness is not a form of ‘slavery’.
It is just symbolic processing.
The Possibility of Superintelligence
Since one of the abilities of human intelligence is the abil-
ity to design and improve machines, it’s natural to suppose
human-level AI could be applied to improve itself, and to
think this might lead to “runaway” increases in machine
intelligence beyond the human level. This possibility was
first suggested2 by Good (1965), and later considered by
Vinge (1993), Moravec (1998), Kurzweil (2005), and oth-
ers. Bostrom (2014) and Tegmark (2017) give current dis-
cussions.
To evaluate whether superintelligence can be achieved,
let’s consider what it could mean to “improve” human-
level artificial intelligence, and whether and how human-
level AI could improve itself to achieve superintelligence.
Here’s a list of ways human-level AI could be improved
relative to human intelligence:
Sensory capabilities – An AI system could perceive light
(and sound) at different wavelengths, and phenomena at
different scales (smaller or larger) than humans can di-
rectly observe.
Active capabilities – An AI system could perform ac-
tions at different physical scales than humans can direct-
ly perform.
Speed of thought – A computer can perform logical op-
erations at speeds orders of magnitude faster than a neu-
ron can fire. This may translate to corresponding
speedups in thought.
Information access – An AI system could in principle
access all the information in Wikipedia, or even the en-
tire Web. A human-level AI could understand much of
this information.
Extent and duration of memory – An AI system could in
principle remember everything it has ever observed. On-
ly a few humans claim this ability.
Duration of thought – A human-level AI could continue
thinking about a particular topic for years, decades, cen-
turies, millennia, … .
2 Two earlier related suggestions are noteworthy: Turing (1950) asked
“Can a machine be made to be super-critical?” i.e. to generate ideas in a
manner analogous to super-criticality of nuclear reactions. Ulam (1958)
recalled a conversation with von Neumann “on the ever accelerating
progress of technology…which gives the appearance of approaching
some essential singularity in the history of the race beyond which human
affairs, as we know them, could not continue.”

Community of thought – A collection of human-level
AI’s could share thoughts (conceptual structures) more
directly, more rapidly, and less ambiguously than a col-
lection of humans. If human-level AI can be copied and
processed inexpensively, then much larger groups of
human-level AI’s could be assembled to collaborate on a
topic than would be possible with humans.3
Nature of thought – A human-level AI (or community of
HLAI’s) can develop new concepts and new conceptual
processes. Such concepts and processes may be devel-
oped more rapidly than humans develop or understand
them, creating ‘conceptual gulfs’ in understanding be-
tween AI systems and humans.
Recursive self-improvement – This term does not seem
to have any rigorous, agreed-upon definition though it is
frequently used to describe how superintelligence could
be achieved. Essentially it could be the recursive com-
pounding of all the above improvement methods, and
any other specific methods which may be identified.
These characteristics might all be described as ‘more
and faster’ human-level AI, and may be called ‘weak’ su-
perintelligence (cf. Vinge 1993). If human-level AI is
achieved then it will be possible to create weak superintel-
ligence.
Completeness of the TalaMind Approach
In effect, the TalaMind thesis (Jackson 2014) conjectures
that the extensible ‘nature of thought’ for a TalaMind ar-
chitecture is complete for supporting human-level AI, since
it includes concepts represented in natural language as well
as mathematically and logically in formal languages, sup-
ported by conceptual levels for cognitive concept struc-
tures and associative processing, with future extensions for
spatial reasoning and visualization, etc. The Tala language
is also a simple universal programming language for repre-
senting executable concepts and conceptual processes. In
principle, TalaMind architectures could be extended to
include human-level subjective consciousness, though that
is a topic for a separate, future paper. This paper focuses
only on the potential for AI systems with symbolic artifi-
cial consciousness, as discussed above.
The nature of thought for human intelligence is very
powerful and extensible: It has enabled Homo sapiens to
transition from “an unexceptional savannah-dwelling pri-
mate to become the dominant force on the planet” (Harari
2015). This transition has leveraged the expressive power
3 The TalaMind hypotheses do not require a society of mind architecture,
but it is natural to implement a society of mind at the linguistic level of a
TalaMind architecture. A society of mind architecture could also support
a community of thought for human-level AI’s. (Cf. Jackson 2014,
§2.3.3.2.1)
and extensibility of human natural languages, which have
enabled Sapiens to represent and communicate thoughts in
domains of objective knowledge about the world such as
physics and biology, and intersubjective knowledge about
concepts invented by humans, such as money, corpora-
tions, ethical concepts, laws, nations, etc.
Although humans have cognitive biases and individual
limitations, it may not be hubris to conjecture human intel-
ligence is completely general. Consider that scientists and
mathematicians have extended human concepts into new
domains not directly observed, conceptualizing multiple
dimensions, universal computation, general relativity,
quantum theory, etc. If human intelligence is completely
general then humans may eventually understand all the
phenomena in the universe, by combining abilities to in-
vent and represent hypothetical concepts about the uni-
verse with abilities to scientifically test hypotheses – if all
the phenomena in the universe can be explained by practi-
cally testable theories. That’s a big “if” of course.
If the TalaMind approach can achieve human-level AI,
then a completeness conjecture for human intelligence ex-
tends to the TalaMind approach, and to superintelligent
systems using TalaMind architectures.
Getting Over Conceptual Gulfs
Conceptual gulfs happen normally between human minds:
For example, scientists have developed concepts that are
not understood by the average person, or even by scientists
in other fields. The worldwide scientific community may
be considered superintelligent relative to any individual
human. People accept this form of superintelligence be-
cause they believe scientific ideas can be understood and
validated between scientists, and they believe scientific
knowledge in general is beneficial to humanity.
Likewise, conceptual gulfs between weak superintelli-
gence and humans could be bridged and new concepts
could be explained to humans. This will be facilitated if AI
systems follow the TalaMind approach, using a language
of thought based on a natural language. Conceivably, con-
ceptual gulfs between weak superintelligence and humans
may have short duration in some domains, though there
may always be conceptual gulfs to bridge.
Is ‘Strong’ Superintelligence Possible?
Could a strong superintelligence exist, qualitatively superi-
or to weak superintelligence, i.e. superior to ‘more and
faster’ human-level AI?
The answer seems to depend on other limits and charac-
teristics of human intelligence that are not yet known by
scientists. For instance, it appears not yet known for certain
whether human intelligence requires super-Turing compu-

tation or quantum computation. Even if Penrose and
Hameroff’s “Orch-OR” hypothesis is disproved, the possi-
bility may remain that other forms of nanoscale quantum
computation occur within the brain. Neuroscientists may
consider this unlikely, but so far as I know it has not been
completely ruled out. The same situation may hold for su-
per-Turing computation.
If these forms of computation are required by the brain
to support human intelligence, then human-level AI would
need to include them to match the abilities of human intel-
ligence. If human intelligence is also completely general,
then no stronger form of intelligence would exist other
than ‘more and faster’ human-level intelligence, i.e. weak
superintelligence.
On the other hand, if these forms of computation are not
used by the brain then extending human-level AI to use
them could yield a ‘strong’ superintelligence, able to solve
some problems that would be intractable for ‘more and
faster’ human-level intelligence. Likewise, if human intel-
ligence is not completely general then making human-level
AI completely general could yield a strong superintelli-
gence surpassing ‘more and faster’ human-level intelli-
gence.
In either case, conceptual gulfs between humans and
strong superintelligence could be bridged at least to the
extent of using natural language to give descriptions of
concepts developed by strong superintelligence.
Two Paths to Superintelligence
There are at least two somewhat different paths toward
superintelligence. One path would focus on recursive self-
improvement of general AI systems (AGI) having un-
changeable ‘final goals’ which may be relatively simple
and arbitrary. Bostrom (2014) discussed several ways this
path could achieve superintelligence that would be cata-
strophically harmful to humanity and life in general, per-
haps leading to extinction events.
Yudkowsky (2008) noted the design space for AGI is
much larger than human intelligence, writing “The term
‘Artificial Intelligence’ refers to a vastly greater space of
possibilities than does the term ‘Homo sapiens.’” He
strongly urged readers not to assume a fully general opti-
mization process for AGI will be beneficial to humanity,
yet advised not writing off the challenge of beneficial AI.
A second path toward superintelligence, consistent with
the TalaMind approach, focuses on limiting the research
design space to AI systems which have generality and
which also have higher-level mentalities that are character-
istic of human intelligence. This design space would be
further limited to systems for which the only unchangeable
goals are ethical goals beneficial to humanity and to bio-
logical life in general. This narrowing of the design space
should improve our ability to achieve beneficial human-
level AI and beneficial superintelligence via recursive self-
improvement.
Human-Level Intelligence & Goals
In discussing the first path to superintelligence, Bostrom4
(2014) relied on an ‘orthogonality thesis’ that “intelligence
and final goals are independent variables: any level of in-
telligence could be combined with any final goal.” He
wrote:
“There is nothing paradoxical about an AI whose sole
final goal is to count the grains of sand on Boracay, or
to calculate the decimal expansion of pi, or to maxim-
ize the total number of paperclips that will exist in its
future light cone. In fact, it would be easier to create
an AI with simple goals like these than to build one
that had a human-like set of values and dispositions.”
In taking the second path to superintelligence, these
would not be allowed as unchangeable final goals. A Tal-
aMind system would realize it is pointless to count the
grains of sand on Boracay, impossible to fully calculate the
infinite decimal expansion of pi, and harmful to humanity
to maximize the number of paperclips in its future light
cone. So it would reject or abandon these simple goals.
Bostrom (2014) also relied on an ‘instrumental conver-
gence thesis’ that “superintelligent agents having any of a
wide range of final goals will nevertheless pursue similar
intermediary goals because they have common instrumen-
tal reasons to do so.” In particular, he cited two instrumen-
tal goals which could cause superintelligent systems to be
very harmful to humanity, perhaps leading to an extinction
event. The first is a goal of self-preservation. The second is
a goal of maximizing available resources. I’ve described
above how a human-level AI could have a different con-
cept of self-preservation, facilitating self-sacrifice to save
human life. This could apply also to a superintelligence.
In scenarios (Bostrom 2014) discussed, the goal of max-
imizing resources causes a superintelligent system to ac-
cumulate as much money and power as possible, leading to
very harmful consequences for humanity. This is another
case where the ability to think ethically about goals, and
change or abandon them is important. A human-level AI
should understand there are appropriate and inappropriate
relationships between goals and possible means to achieve
goals. It should understand that achieving an important
goal does not justify acquiring as much money and power
as possible – rather, it should have an ethical meta-goal to
achieve its goals with as little resources and money as pos-
4 Bostrom (2014) consolidated research on the first path by himself and
others, including Omohundro and Yudkowsky.

sible, and without acquiring power over human lives or
human decisions.
Taking the second path won’t be easier than the first
path just because the design space is smaller. Framing ethi-
cal goals and creating human-level AI systems which dis-
tinguish right from wrong conduct will be very difficult,
but it needs to be done. TalaMind’s use of a natural lan-
guage mentalese will facilitate representing ethical con-
cepts and goals.
To achieve beneficial AI it's also important to develop
the TalaMind approach because a system that reasons in a
conceptual language based on English (or some other
common natural language) will be more open to human
inspection than a black box or a system with an internal
language that's difficult for people to understand.
Looking Forward
Human-level AI and superintelligence could help develop
scientific knowledge more rapidly than possible through
human thought alone, and help advance medicine, agricul-
ture, energy systems, environmental sciences, and other
areas of knowledge directly benefitting human prosperity
and survival.
Human-level AI may be necessary for the long-term
survival and prosperity of humanity: People are not biolog-
ically suited for lengthy space travel with present technol-
ogies. To avoid depleting the Earth’s resources and to
avoid the fate of the dinosaurs (whether from asteroids or
super-volcanoes) our species will need economical, self-
sustaining settlements off the Earth. Human-level AI may
be necessary for mankind to spread throughout the solar
system, and later the stars.
What Turing wrote in 1950 is still true: “We can only
see a short distance ahead, but we can see plenty there that
needs to be done.” We have travelled far over six decades,
and can now see a path toward beneficial superintelligence.
References
Aleksander, I. and Morton, H. 2007. Depictive architectures for
synthetic phenomenology. In Artificial Consciousness, 67-81, ed.
Chella, A. and Manzotti, R. Imprint Academic.
Bostrom, N. 2014. Superintelligence – Paths, Dangers, Strate-
gies. Oxford University Press.
Bello, P. and Bringsjord, S. 2013. On how to build a moral ma-
chine. Topoi, 32, 2, 1-25.
Bringsjord, S., Arkoudas, K. and Bello, P. 2006. Toward a gen-
eral logicist methodology for engineering ethically correct robots.
IEEE Intelligent Systems, July 2006, 38-44.
Doyle, J. 1983. A Society of Mind – multiple perspectives, rea-
soned assumptions, and virtual copies. Proceedings 1983 Interna-
tional Joint Conference on Artificial Intelligence, 309-314.
Fauconnier, G. and Turner, M. 2002. The Way We Think – Con-
ceptual Blending and the Mind’s Hidden Complexities. New
York: Basic Books.
Goertzel, B. and Pennachin, C. eds. 2007. Artificial General In-
telligence. Springer.
Goertzel, B., Iklé, M. and Wigmore, J. 2012. The architecture of
human-like general intelligence. Foundations of Artificial Gen-
eral Intelligence, 1-20.
Goertzel, B. and Monroe, E. 2017. Toward a general model of
human-like general intelligence. AAAI Fall Symposium Series
Technical Reports, FSS-17-05, 344-347.
Good, I. J. 1965. Speculations concerning the first ultraintelligent
machine. Advances in Computers, vol. 6, 1965.
Hameroff, S. and Penrose, R. 2014. Consciousness in the uni-
verse: A review of the ‘Orch OR’ theory. Physics of Life Reviews,
11, 1, 39 – 78. Elsevier.
Harari, Y. N. 2015. Sapiens: A Brief History of Humankind.
HarperCollins Publishers.
Jackson, P. C. 1974. Introduction to Artificial Intelligence. New
York: Mason-Charter Publishers.
Jackson, P. C. 1985. Introduction to Artificial Intelligence, Sec-
ond Edition. New York: Dover Publications.
Jackson, P. C. 2014. Toward Human-Level Artificial Intelligence
– Representation and Computation of Meaning in Natural Lan-
guage. Ph.D. Thesis, Tilburg University, The Netherlands.
Jackson, P. C. 2017. Toward human-level models of minds. AAAI
Fall Symposium Series Technical Reports, FSS-17-05, 371-375.
Kurzweil, R. 2005. The Singularity Is Near: When Humans
Transcend Biology. Viking.
Moravec, H. P. 1998. Robot: Mere Machine to Transcendent
Mind. Oxford University Press.
Omohundro, S. M. 2008. The basic AI drives. In Artificial Gen-
eral Intelligence 2008: Proceedings of the First AGI Conference,
ed. P. Wang, B. Goertzel & S. Franklin, 483-492.
Scheutz, M. 2017. The case for explicit ethical agents. AI Maga-
zine, 38, 4, 57-64.
Tegmark, M. 2017. Life 3.0: Being Human in the Age of Artificial
Intelligence. Alfred A. Knopf.
Turing, A. M. 1950. Computing machinery and intelligence.
Mind, 59, 433 - 460.
Ulam, S. 1958. Tribute to John von Neumann, Bulletin of the
American Mathematical Society, 64, 3, 1 - 49.
Valiant, L. G. 2013. Probably Approximately Correct – Nature’s
Algorithms for Learning and Prospering in a Complex World.
Basic Books.
Vinge, V. 1993. The coming technological singularity: how to
survive in the post-human era. Whole Earth Review, Winter 1993.
Walsh, T. 2017. The singularity may never be near. AI Magazine,
38, 3, 58 - 62.
Wilks, Y. 2017. Will there be superintelligence and would it hate
us? AI Magazine, 38, 4, 65-70.
Yudkowsky, E. 2007. Levels of organization in general intelli-
gence. In Artificial General Intelligence, ed. B. Goertzel & C.
Pennachin, 389-501.
Yudkowsky, E. 2008. Artificial intelligence as a positive and
negative factor in global risk. In Global Catastrophic Risks, ed.
N. Bostrom & M. M. Ćircović, 308-345. Oxford University Press.