Sketch of an AGI Curriculum
I often get asked questions like: "If I want to work on AGI, what should I study first to get up to speed?"
This post — updated from a similar post I wrote some years ago — gives a rough stab at an answer.
Basically what I’m going to give here is a list of some books that would be useful reading for anyone wanting to seriously get into AGI. Some are more critical than others; and of course, omission of a book from this list does not imply its irrelevance or unimportance.
(By the way, I regret to note that most of these books cost money to legally obtain. I would have liked to assemble a comparable list consisting only of legally freely available materials, but that would have required a lot more effort.)
If I were going to structure a degree program on AGI, I would use these books as part of the core. At a first stab, I might divide the curriculum into six main courses, such as:
History of AI
AI Algorithms, Structures and Methods
Neuroscience & Cognitive Psychology
Philosophy of Mind
AGI Theories & Architectures
Future of AGI
The books listed below would give raw materials for all the above courses (to be supplemented by various papers and assignments, etc.). The division into 6 categories corresponding to the above list is extremely crude as many of the books richly cross-cut multiple categories.
Prelude
As prelude to serious study of a truckload of books on AGI I would recommend a few simple brief readings first:
My slightly technical summary and overview of the Artificial General Intelligence field — plus a much longer version, Artificial General Intelligence: Concept, State of the Art, and Future Prospects that is dated as of 2014
My thoughts in 2020 on what major steps still need to be taken to get from here to AGI: From Here to Human-Level AGI in Four (Not All That) Simple Steps.
My thoughts on getting From Narrow AI to AGI via Narrow AGI
History of AI
These classic books will give you a feeling for the early history of the AI field:
Computers and Thought, an edited book from 1963 that gives a feel of where AI was at back then
What Computers Still Can't Do, by Hubert Dreyfus — a powerful and entertaining counter-argument to the rule-based AI that was popular in the 1970s and 80s. When I met Dreyfus in the mid-90s and asked him if he thought neural network methods might bypass the difficulties he was observing with rule-based AI, he said maybe.
Machines Who Think, by Pamela McCorduck — an actual history of the early AI field, written in 2004. See also her 2019 memoir This Could Be Important
I also wrote a chapter on the history of AI and AGI XXXX, for an AGI textbook that I haven’t yet managed to pull together.
Melanie Mitchell’s 2019 book AI: A Guide for Thinking Humans reviews aspects of the history and current state of AI — and her 2021 article Why AI is Harder Than We Think? follows up by articulating key mistakes she feels the AI field has made:
Narrow intelligence is on a continuum with general intelligence
Easy things are easy and hard things are hard [when actually some stuff that’s easy for current computers is hard for people and vice versa]
The lure of wishful mnemonics [calling a software process “perception” doesn't make it so]
Intelligence is all in the brain [versus the body, the relationships, the society…]
I tend to agree with her that these mistakes have typically been made, but I feel some modern approaches such as “OpenCog Hyperon embedded in SingularityNET” XXX go beyond them.
AI Algorithms, Structures and Methods
The books in this category are not mainly about AGI, but present ideas that are worth knowing about if you're going to work on AGI. (Just as if you want to learn quantum mechanics and build quantum machines, then given the nature of current physics knowledge, you’d better learn classical mechanics and classical electromagnetism first.)
Artificial Intelligence by Russell and Norvig. This is totally, explicitly and proudly not an AGI book, it's a narrow AI book that definitively exalts in being a narrow AI book. But it's excellent for what it is. Many of the ideas and methods described here have a role to play in various AGI architectures.
(When Peter Norvig gave a brief intro talk at the 2011 iteration of the AGI conference that I’ve organized every year since 2006 (skipping 2007), he ended up amused at the optimism I and some other AGI researchers demonstrated that we would achieve human-level AGI before 2030. He hacked Google’s internal scheduling system so that the Jan. 1 2030 calendar entry contained the event “Human-level AGI”, and the calendar entry for a couple months later (March 1 I think) contained the event “Human-level AGI, Mac version”)
The Design of Innovation by David Goldberg -- a deep, wide-ranging and readable discussion on evolutionary learning
Introduction to Evolutionary Computing, by Eiben and Smith -- a competent, current overview of genetic algorithms and genetic programming
Neural Networks and Learning Machines, by Simon Haykin -- a good (though long) review of work on neural nets and related methods for machine learning, including recurrent neural networks
Foundations of Language by Ray Jackendoff -- by far the most thorough and deep treatment of traditional linguistics (as opposed to statistical linguistics) I've seen. A great way to understand the nature of all the various linguistic phenomena that an AGI will have to deal with.
Speech and Language Processing, by Jurafsky and Martin -- an excellent review of the field of "statistical language processing." This is certainly not an AGI-ish approach to linguistics, yet it does teach us a lot about the nature of language, since any AGI that learns language is going to rely on similar statistical phenomena to a certain extent.
Probabilistic Robotics, by Thrun, Burgard and Fox -- a narrow-AI approach to robotics, but it's useful to know how this stuff is done, and what difficulties they run up against
Handbook of Practical Logic and Automated Reasoning, by John Harrison -- a thorough and practical guide to the current state of automated theorem proving, with copious examples in OCAML …
Deep Learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville— a masterful treatment by Bengio — a long-time master of the field (and an inventor of key ideas in the field) — and two fantastic younger colleagues
Neuroscience and Cognitive Psychology
Some solid textbooks...
Neuroscience: Exploring the Brain, Bear, Connors and Paradiso -- a remarkably comprehensible textbook summarizing our current knowledge on the complex system that is the human brain
Fundamentals of Cognitive Psychology by Robert Kellogg -- a straightforward review of cognitive psychology, a bit dry but well worth understanding if you want to build a human-like AGI
Developmental Cognitive Neuroscience, by Johnson and de Hann -- a straightforward text reviewing how the child's mind/brain develops. Useful to know if you want to build an AGI that develops in some vaguely similar way.
Some lighter-weight books presenting individual scientists' relevant ideas:
Constructing a Language, by Michael Tomasello -- a masterful review of language learning as a social and embodied process
Action in Perception by Alva Noe -- on the connection between seeing and acting
The Vision Revolution by Mark Changizi -- explaining vision as prediction from a neuroscience and cognitive science perspective
On Intelligence by Jeff Hawkins -- presents a view of neuroscience and AGI that I have argued is very oversimplified, but still is worth knowing about
Some Philosophy of Mind Relevant to AGI
My book The Hidden Pattern presents a philosophy of mind specifically oriented toward AGI, though also dealing with many other topics. Some more recent ideas along similar lines are given in my long paper on General Theory of General Intelligence and, from a different angle, my long paper on the Euryphysical model of the universe.
Neural Correlates of Consciousness, an edited volume by Thomas Metzinger
Being No One, by Thomas Metzinger -- the best book I know about how the mind creates the self
The Radiance of Being, by Allan Combs -- a fantastic review and analysis of the various states of consciousness humans get into
The Embodied Mind, by Varela, Thompson and Rosch -- a classic on the relation between mind and body
Supersizing the Mind by Andy Clark -- an up-to-date overview of work on the "extended mind"; the way mind extends into environment and body, rather than just residing in brain
Erik Jantsch, The Self-Organizing Universe … a "big picture" systems-theory view of the mind and world, putting human intelligence in perspective. Out of print and hard to find, but a rather good read
Gregory Bateson, Mind & Nature: a Necessary Unity -- describing mind as a cybernetic system among many others
The Neurophilosophy of Free Will, by Henrik Walter -- a take on "Free will" that is scientifically sound and relevant to AGI
Constructing the World by David Chalmers, a long, technical, at times tedious but extremely important work explaining how key “semantic primitives” constituting essential aspects of the mind can be viewed as building up to create all the other aspects. If you like modern analytical philosophy you will love this book, and there are many deep and fascinating implications for AGI throughout. If you don’t like modern analytical philosophy you will not get through it, unless you have masochistic tendencies or this is the only book you’ve been allowed in your jail cell, etc.
Some Readings about AGI Itself
In 2014 I wrote a nontechnical overview book on my own approach to AGI (also surveying a lot of other material), called The AGI Revolution
Way back in 2005 or so,Cassio Pennachin and I edited a book on Artificial General Intelligence, which initially put the term out there to the world
Pei Wang and I edited a book on the Theoretical Foundations of Artificial General Intelligence
Marcus Hutter’s book Universal AI is pretty mathematical but it’s also a work of art and has a great deal of conceptual value. Shane Legg’s book Machine Superintelligence is along similar lines and shorter and easier to read.
Eric Baum's book What Is Thought? reviews a variety of interesting issues related to AGI -- including the intersection of AI and economics, which is critical to the SingularityNET project among other things...
Erik Mueller's book Commonsense Reasoning presents a logic-based approach to AGI in some depth, drawing directly on the ideas of AI pioneer John McCarthy
Pei Wang's book Rigid Flexibility outlines Pei's unique view on AGI and the underlying logical and control mechanisms
The proceedings volumes of the Artificial General Intelligence conference series contain a host of papers related to AGI, and the conference websites contain links to free PDFs of the papers
The AGI Journal contains papers relevant to AGI, primarily at this point of a theoretical nature.
My former colleague Moshe Looks' PhD thesis is excellent and explains how to combine evolutionary learning and probabilistic modeling.
This book summarizes work on SOAR, perhaps the most thorough and successful of the "Good Old Fashiond AI Systems" (and which is still under development, incorporating a number of modern features): The Soar Cognitive Architecture
SOAR-related but in some ways more conceptually advanced, there is more recent work by a closely overlapping community on the Common Model of Cognition
And then of course there’s our tome on OpenCog: Engineering General Intelligence, Volume 1 and Volume 2 … some concepts in which are updated in the recent General Theory of General Iintelligence paper, and some practical engineering notions of which are updated in the OpenCog Hyperon design.
Future of AGI
My own visions here are covered in my book "AGI Revolution" mentioned above, and in a few other books such as
Between Ape and Artilect, a collection of dialogues with other AGI researchers and advanced technologists and scientists on the radical future
A book I edited with my dad, The End of the Beginning,
Kurzweil's Singularity is Near remains a classic ... as does Feinberg's 1969 book on the same themes, The Prometheus Project (fascinating to see how a brilliant physicist thought about superhuman AI, nanotech and superlongevity back in the 1960s...)
Max Tegmark's Life 3.0 is engagingly written and wide ranging. He is much more worried than I am about potential negative futures related to AI; but of all the "worried about AI" folks in vogue today, I find his take the most compelling.