Hello,

in my opinion, an introductory AI course should introduce the main paradigms: good-old-fashioned-AI (logic, search, planning, etc.), probabilistic reasoning, and machine learning based approaches.

As I learned AI in a mostly GOFAI-focused lecture (using Prolog), I would intuitively say that 'classic AI' should make up most of the course. A good textbook for these topics (and also for probabilistic reasoning) is Russell & Norvig's "Artificial Intelligence, A Modern Approach". This part of the course should definitely cover (1) search, (2) constraint satisfaction problems, (3) inference in first-order logic and automated theorem proving, (4) planning, (5) maybe some game theory for multi-agent systems.

The rest of the course could then be split between probabilistic inference (subtopics (6) fuzzy logic, (7) Bayesian reasoning) and machine learning (as Peter König said, (8) supervised, (9) unsupervised and (10) reinforcement learning, probably with a short discussion of the current deep learning hype).

Maybe outlooks on natural language processing or robotics applications would be nice, especially if the course has a practical part where the students have to program themselves.

Best, Christian

hey rey,

AI is a wide term. what focus have you thought of? the following branches crossed my mind:

- planning and the whole branch about heuristic search

- empirical risk minimization including the whole machine learning zoo

- and maybe data mining

best

christoph

Dear Rey, the set of supervised learning / reinforcement learning / unsupervised learning should be introduced. Best, Peter

Hello,

in my opinion, an introductory AI course should introduce the main paradigms: good-old-fashioned-AI (logic, search, planning, etc.), probabilistic reasoning, and machine learning based approaches.

As I learned AI in a mostly GOFAI-focused lecture (using Prolog), I would intuitively say that 'classic AI' should make up most of the course. A good textbook for these topics (and also for probabilistic reasoning) is Russell & Norvig's "Artificial Intelligence, A Modern Approach". This part of the course should definitely cover (1) search, (2) constraint satisfaction problems, (3) inference in first-order logic and automated theorem proving, (4) planning, (5) maybe some game theory for multi-agent systems.

The rest of the course could then be split between probabilistic inference (subtopics (6) fuzzy logic, (7) Bayesian reasoning) and machine learning (as Peter König said, (8) supervised, (9) unsupervised and (10) reinforcement learning, probably with a short discussion of the current deep learning hype).

Maybe outlooks on natural language processing or robotics applications would be nice, especially if the course has a practical part where the students have to program themselves.

Best, Christian

Dear Christian,

kind of reminds me of the Cognitive Science Studies at University Osnabrück.

I always have a neuro bias :)

Best, Peter

Dear Peter,

in fact, that is where I studied (and I still recommend it) and the course I had in mind was Introduction to Artificial Intelligence by Helmar Gust.

But since I have drifted more towards neural information processing research (also thanks to your Action and Cognition lectures) and seeing the current developments in deep learning and cognitive computing, I definitely agree with your neuro bias: someone who puts together an AI course today should consider neural architectures for learning and decision making.

Best, Christian

Dear Rey Segundo Guerrero-Proenza    

I would say cover the real (actual) "open" capacities of the human -- the memories.  AND: Cover abilities using a very open structure ** -- very much a challenge, since we do not know the ultimate environmental/behavior-pattern-responses involved in the inception of the higher level cognitive abilities [( these could well be discovered with the new eye-tracking technology, etc. )].  Emphasize using definitions of these capacities and abilities INDEPENDENTLY DEFINED and based on good research with strong results.  Minimize anything encouraging the "making up" of capacities or behaviors, rather learn to use what is there in the human FOR AI.  Don't get goals  ahead of foundations for any first-good-start.

** FOOTNOTE: Abandon any ideas of embodiment (somehow trying for some version of an analogue of all understanding in-the-body, in those terms or in similar terms -- that will NEVER be an "open" system, as AI requires).  Largely ignore phenomenon supposed and described by "embodiment theories" OR anything simply analogous -- this area of 'theory', study and research shows NO good results and is tenable ONLY if one uses their much-less-likely assumptions about human behavior (and it seems weirdly-founded and to have no promise: see "The poverty of embodied cognition", Jun 2016, Psychonomic Bulletin & Review).  (Allow no homunculus -- any "man-wthin-the-man":  THUS abandon all notions of "meta" processes OR "executive control".)

RATHER: Emphasize what nature no doubt emphasized: embedded-ness WITH ***  the environment --  and  CONCRETE and AVAILABLE ENVIRONMENTAL ASPECTS FOR ANY NEW behavioral/response DEVELOPMENTS (and have it workable that the major results are used to very much guide learningS (note the plural) -- making aspects of major types of learningS both universal and reliable). (Would you really want to MAKE INFERENCES from anything other than behavior-and-environment, in its specifics?)

You can see my posts (Questions and Answers) on AI Questions to get any more of an idea -- ESPECIALLY CAUTIONING ABOUT ASSUMPTIONS (ethology has an entirely different set of more-likely assumptions, than common Psychology).  KNOW what assumptions about humans YOU are making -- because NO MATTER how much you do other things correctly, these will have an impact.  See my Project: https://www.researchgate.net/project/Developing-a-Usable-Empirically-Based-Outline-of-Human-Behavior-for-FULL-Artificial-Intelligence-and-for-Psychology

*** FOOTNOTE: Note the word "WITH" not "IN" to avoid nature/nurture issues completely.

Article A Human Ethogram: Its Scientific Acceptability and Importanc...

Article The poverty of embodied cognition

Hi,

I think it is better to go with a combination consists,

Introduction to Ai - What is Ai, birth of Ai, Ai in fiction/movies,difference between Ai & Human Intelligence, Turnig Test, & topics like How Ai can shape up our future

Basics - Problem solving, Searching algorithms , game theory,real world planning & uncertainty, probability in Ai

Prolog Programming

Introduction to Expert Systems and Expert Systems Programming (CLIPS + Java )

Machine Learning Concepts

ANN - Supervised / Unsupervised, Back propagation

Support Vector Machines

Introduction to Fuzzy Set Theory & Logics, Genetic Algorithms, Neuro-Fuzzy, Multi Agents

Applications of Ai Techniques ( Software, Robotics etc.)

Regards,

Hiroshan

Dear Hiroshan Gunawardane,

I agree with your version, or at least it is the nearest, I have thought previously, could be an Introduction to AI. Even it has some components, proposed by other participants in this discussion. Just I have a question: May be could be considered in the order of topics, first learning Prolog Programming?; as you know, many Prolog books include implementations of Problem solving, Searching algorithms, game theory, and even primarily Expert Systems (shell I mean) can be developed easily using Prolog.

BTW, there exist some points of view, that consider Prolog (or Declarative Programming in general) as the best option to introduction to Computer Science, using It as the programming tool, in usually complicated themes for students, like Data Structure and Algorithms.

Regards,

Rey.

Dear Brad Jesness,

Very interesting, and, I have to say, complicated (at least for me) your post. I'll need more time to deeper analyze it, read your suggested sources. About embodiment theories, I have a little background, some time ago I took part in a course (MOOC), which partially was related to embodiment learning, but I have not further researched on this topic.

On the other hand, I found (or I think I found) in your paragraph:

"Emphasize what nature no doubt emphasized: embedded-ness WITH *** the environment -- and CONCRETE and AVAILABLE ENVIRONMENTAL ASPECTS FOR ANY NEW behavioral/response DEVELOPMENTS (and have it workable that the major results are used to very much guide learningS (note the plural) -- making aspects of major types of learningS both universal and reliable). (Would you really want to MAKE INFERENCES from anything other than behavior-and-environment, in its specifics?) "

some shared ideas with Cultural Historical Theory and Activity Theory. Is it true?

Regards,

Rey.

Dear Rey Segundo Guerrero-Proenza  

I am not familiar with "Cultural Historical Theory and Activity Theory".  Is that one thing or two?  In any case, regarding "cultural factors": I have never been able to see anything as direct, empirical proximate causes (and effects) of these.  In my perspective the individual organism is THE unit and I think only in those terms. 

As far as details for AI (and for a good psychology) from my theoretical perspective: it is the 'open' (though at the same time limited) CAPACITIES of the memories which is one grand set of things to consider; then also there is the development of ABILITIES, major ones greatly guided by innate action patterns (expressing themselves likely very subtly in the later stages of ontogeny in "perceptual/attentional shifts"). 

You might like to see my Project, https://www.researchgate.net/project/Developing-a-Usable-Empirically-Based-Outline-of-Human-Behavior-for-FULL-Artificial-Intelligence-and-for-Psychology

Also, in particular, the last "letter" (as in 'letter you send'), at the bottom of   https://mynichecomp.com/AImemory.txt .  THAT is my overall view: the needed investigation and needed integration of the open capacities (memories) with abilities that develop.   To have a full overview of the development of major cognitive abilities (and the hypotheses that need to be detailed, then tested) see "A Human Ethogram ..." a Resource under my Ethogram Theory Project: https://www.researchgate.net/project/Human-Ethology-and-Development-Ethogram-Theory

[ The statements in the last letter (last few paragraphs) of https://mynichecomp.com/AImemory.txt give you an idea of what just IS (that is: NOW), phenomenologically, based on given human nature and past developments and learnings _AND _the new aspects of the environment yet to be discovered, which provide the major direction for behavioral development and new learnings -- these are what I believe must be (in effect) simultaneously related to the innate action patterns (referred to above) and are the bases of major cognitive-developmental changes. ]

My entire perspective involves a different (better) set of assumptions about the way humans are.  SEE: https://www.researchgate.net/post/Would_Skinner_support_my_view_He_would_if_he_wanted_to_be_the_ultimate_empiricist_OK 

(especially my recent follow-up Answer under that question ).

[ Emotions and language may be worked into the system later; they are not necessarily present, or primary (respectively) determinants of behavior.  I do not consider "interest" an emotion -- rather that "just IS" , since it is very ubiquitous. ]

P.S. I have not found myself to have a good "interface" with present modern cognitive-developmental psychology theories; it is the do-ability and practicality (concrete nature) of my perspective which I think might be attractive to AI  engineers and, like in the 1980s (the information-processing hayday) , computer models may LEAD psychology.  I believe my system, for AI engineers, is as simple as it gets and is the least subjective. (In a real sense, I define nothing; as should be the case the subject (organism-and-environment) DOES ALL THE DEFINING -- but we do have to do a quite a bit of finding (discovery); yet, at the same time, it makes for doable proof-of-concept robots rather readily.)

Dear Rey Segundo Guerrero-Proenza 

It is my hope that you read my response after I made the final edits.  I have a habit of finding all typos and adding all I want only by a point about an hour after I made the response in the first place.  I apologize to all for my habit of composing on-line

Dear Brad Jesness,

No problem with that, I am more concerned with content.

Regarding Cultural Historical Theory (or approach) and Activity Theory, they are different theories, but the second is derived from the first one, although currently many researchers are using the expression Cultural Historical Activity Theory (CHAT), mostly in West Europe and North America, and even many people, following "the tradition" in Russia, are considering both as one theory. The David Bakhurst article, "Reflections on Activity Theory" is, probably, a good first source for understanding both theories from the western philosophical tradition point of view.