The Intersection of AI and Mind/Brain Sciences

[0rthodontist]

AI in "mind or brain sciences"

Why is AI part of the description of the mind or brain sciences forum? Most work in AI has nothing directly to do with the human mind or brain. It's a division of computer science.

 

[Evo]

Why is AI part of the description of the mind or brain sciences forum? Most work in AI has nothing directly to do with the human mind or brain. It's a division of computer science.

That was my question.

 

[0rthodontist]

It would be like saying humanoid robotics is a biological science because parts of it are based on the human body.

 

[Moonbear]

The mind and brain forum actually crosses a number of disciplines and is not purely biology. AI fits there because it is generally attempting to model human intelligence/learning, which means one must understand how human intelligence/learning is functioning in the first place. But, of course it would also depend on the particular question or topic being discussed. AI might also fit within the computers forum, or the engineering forum, depending on the particulars of the discussion being started and the types of replies being sought.

 

[neurocomp2003]

people in cs view AI as something completely different to the rest of the world. To the majority of the world AI is considered the attempt to model human intelligence. In cs it incorporates many other things like fuzzy logic, game trees, A* and other forms of problem solving.

Seeing as this forums is not 100% a cs forum, it would seem suited to belong with other fields that inspired its existence. As moonbear said(btw what happen to that journal club??) Mind and Brain science isn't just biology...its all disciplines related to it from physics(modelling neurons) Chemistry(modelling neurons, NT pathways,ion channels), Neuropsych & cogsci(the entire system,patterns & behaviour), Biology(socio, ecology, predator&prey), Math(Adaptive Learning tools like spiking neurons, Neural Nets, Reinforcement Learning, Markov Models or stochastics) and Rendering Virtual Worlds(testbeds),ALife and AI(simpler means of modelling)

 

[0rthodontist]

I wasn't saying that mind and brain sciences was necessarily biological, just making a comparison with robotics for illustration. Mind and brain sciences could also be psychological, but they are still only marginally relevant to AI, probably on the order of relevance that the study of a human body has to the construction of ASIMO.

The majority of work done in AI is on instance-specific problems such as
--automating a logical inference system (expert system) which was inspired by the human mind but is not a connectionist model. Work in expert systems is based in mathematical logic, not psychology.
--algorithms for computer vision that are sometimes very loosely based on human vision. I was at a commencement at Mount Holyoke College and one student winning an academic award did a project in computer image recognition. The announcer gave a little speech about how making a computer recognize a tiger on the internet might help us understand ourselves. The student was so stricken by this that she had to step up to the microphone and deny it in front of the large audience, saying she only worked on how to make a computer see, not on the human mind.
--genetic algorithms to solve optimization problems
--connectionist models-neural networks-that do appear similar to the nerves in the brain, but work on these is done in trying different algorithms with no basis in biology. Neural networks are used for very specific approximation problems.

Any questions on AI that are suitable for someone with a background in mind and brain sciences cannot reach very deeply into the subject. People who work in AI are computer scientists and talk about AI as a computer science discipline. Some may use analogies to the human mind, but the work done has far more to do with mathematics and computer science. Only a small fraction of AI work even aims to imitate humans; most of it is designed to solve narrow computational problems like analyzing Pap slides.

It would fit in the computers forum, maybe as an addendum to the Programming subforum, probably not the engineering forum. AI has become like an engineering discipline but it's not traditional, physical engineering. It's more like software engineering.

people in cs view AI as something completely different to the rest of the world. To the majority of the world AI is considered the attempt to model human intelligence. In cs it incorporates many other things like fuzzy logic, game trees, A* and other forms of problem solving.

So we should accommodate the layman's perspective instead of grouping it as it is actually done? Just because a science is commonly viewed in a certain way doesn't mean it should be.

 

[Moonbear]

So we should accommodate the layman's perspective instead of grouping it as it is actually done? Just because a science is commonly viewed in a certain way doesn't mean it should be.

Just post the question where you think it fits best. If your question is purely computer science and deals with nothing biological, post it in the computers subforum.

btw what happen to that journal club??

Lack of people willing to present articles. I'd love to start it up again. Do you have a paper you'd like to present?

 

[0rthodontist]

Just post the question where you think it fits best. If your question is purely computer science and deals with nothing biological, post it in the computers subforum.

Well, truth be told, I don't have any questions on AI right now, and anyway who would answer them. It just bothers me in principle that people view AI as some kind of psychology.

 

[neurocomp2003]

Moonbear: I'll be entering grad school(Masters) this fall so hopefully i'll be able to scrounge up a paper...problem is what I'm going to be doing my project on doesn't have anything to do with M&B...hopefully when it comes time to do my Phd...i'll be albe to...cuz this journal club seems like a good opportunity for me to start reading papers and expressing my views..without actually having to talk to people.

 

[Moonbear]

Well, truth be told, I don't have any questions on AI right now, and anyway who would answer them. It just bothers me in principle that people view AI as some kind of psychology.

Why? Some aspects of it are.

CBCL studies the problem of learning within a multidisciplinary approach. Its main goal is to nurture serious research on the mathematics, the engineering and the neuroscience of learning. Established in 1992 with support from the National Science Foundation, CBCL is based in the Department of Brain & Cognitive Sciences at MIT (click here for map information) and is associated with The McGovern Institute for Brain Research and with the Computer Science & Artificial Intelligence Laboratory (CSAIL).

http://cbcl.mit.edu/cbcl/index.html
Also see http://www.csail.mit.edu/events/eventcalendar/calendar.php?show=event&id=1204 for an example of a research project within the Computer Science and Artificial Intelligence Laboratory at MIT that is clearly within the realm of neuroscience.

"AI can have two purposes. One is to use the power of computers to augment human thinking, just as we use motors to augment human or horse power. Robotics and expert systems are major branches of that. The other is to use a computer's artificial intelligence to understand how humans think. In a humanoid way. If you test your programs not merely by what they can accomplish, but how they accomplish it, they you're really doing cognitive science; you're using AI to understand the human mind."

- Herbert Simon

http://www.aaai.org/AITopics/html/cogsci.html

and

Q. How is AI research done?

A. AI research has both theoretical and experimental sides. The experimental side has both basic and applied aspects.

There are two main lines of research. One is biological, based on the idea that since humans are intelligent, AI should study humans and imitate their psychology or physiology. The other is phenomenal, based on studying and formalizing common sense facts about the world and the problems that the world presents to the achievement of goals. The two approaches interact to some extent, and both should eventually succeed. It is a race, but both racers seem to be walking.

http://www-formal.stanford.edu/jmc/whatisai/node4.html

 

[Moonbear]

Moonbear: I'll be entering grad school(Masters) this fall so hopefully i'll be able to scrounge up a paper...problem is what I'm going to be doing my project on doesn't have anything to do with M&B...hopefully when it comes time to do my Phd...i'll be albe to...cuz this journal club seems like a good opportunity for me to start reading papers and expressing my views..without actually having to talk to people.

Just let me or hypnagogue know when/if you're ready to present something and we'll get your sticky thread ready! I agree, it's a great way to practice critical reading and reviewing of a paper without the pressure of having to explain it to someone you know understands it better than you do...you get to be the "expert" on your topic here. And the rest of us get to learn about another area of neuroscience (or other topics related to the Mind & Brain sciences forum) that we might not otherwise find time to read about.

 

[neurocomp2003]

Orthodontist:
M&B isn't just bio or psych, but that's besides the point. The AI questoins that have been posted in on these forums that I've read have been (if not all questions) are on Artificial Human INtelligence.

Also M&B isn't about simulating the entire brain, its also about simulating individual processes and about attempting to create artificial systems that simulate the holistic behaviour for industrial use (ARTIFICIAL and SIMULATE being the key words). Computer vision, the goal is for a computer to do the act of seeing, and not necessarily emulate the subprocesses that the brain takes. But still the act of seeing is a sensation that belongs in M&B.
Likewise with Computer Audition, Speech Rec. To your tiger example it'd be a wonder what else the "agent" actually sees or if she can claim it as the act of seeing if it can't recognize anything else.

Neural Networks is based in simulating brain processes inorder to either mimic the brain or help some application where adaptive learning is required. Its foundations or history however lies in M&B. Likewise Genetic Algorithms...thats why a lot of GA packages have names that relate to the actual biology terms. like Fitness, or Genome.

I'll agree with you that most AI people will only graze certain topics in M&B. But for those who wish to learn Adaptive Learning, Most courses and textbooks will atleast give some introduction 1-2 chapters in these biological fields. Or at least the many that ihave read. have chapters in Neuronal properties, Gene Expression, Pavlov's experiments

I will have to disagree with this statement though
"Any questions on AI that are suitable for someone with a background in mind and brain sciences cannot reach very deeply into the subject. People who work in AI are computer scientists and talk about AI as a computer science discipline".

Most Researchers that I have read up on that deal with computational M&B have formal training in CS, MATH, and then found passion in M&B which became their research.

CMU&Upitts being the primary group of researchers that I have looked at
that run the Graduate Programme called CNBC(i think) have faculty members from robotics, stats,math,psych,bio,eng, linguistics, and i think phys(can't remmeber this one). Other groups include redwood(berkeley i believe) and the groups at Mac, UFT and York, MIT obviously, and i ibelieve arizona.

I also doubt that certain AI profs could help on questions in ALife or Adaptive Learning.

 

[0rthodontist]

AI has links to psychology. But it is not a psychological discipline. The study you referred to doesn't look like it concerns itself much with AI per se, even though it uses a computer model to look at biology. From the summary, the neurological model is only a model, not intended to function as any kind of artificial intelligence.

From what I've read of AI, the biological or psychological side of it only loosely influences actual practice. It may be important, but take the example of expert systems. They were originally designed to mimic human thought processes--in the early days of AI they were thought to do so sufficiently well that soon people would develop a strong AI using expert systems techniques. So you might say that expert systems have a philosophical basis in psychology. But in actual practice, expert systems are mathematical and logical constructs that take their cues from inventive programmers and mathematicians and not from nature. This is true of every field within applied AI.

Apparently from the links you have posted, cognitive science has a connection with AI. How strong this connection is I don't know--I would suspect it is not very strong except to a philosopher. Everybody at my school doing AI research is a computer scientist. AI courses are computer science courses. To get a master's or PhD in CS, AI composes many of your core courses. Do cognitive psychologists actually produce models that are useful for real-world computational problems?


Ah, neurocomp you have a habit of posting just as I'm starting a long post. I think part of what you're saying is that adaptive learning, neural networks, genetic algorithms are not properly part of AI. While it is true that the most widely successful--practical--AI so far has been expert systems, neural networks and genetic algorithms also are AI. Genetic algorithms are biologically inspired, but not inspired by brains but by natural competition. Neural networks are LOOSELY based on nerve cells. It's not such a close comparison. What someone actually using a neural network has to consider are things like: --how many layers to use --how many nodes to use --whether to use batch learning or online learning --the specific learning algorithm to use (backpropagation, RBF) --whether to let some nodes sleep --whether to check if the data is linearly separable --etc. These are all based on practical, empirical or mathematical concerns, not the historical inspiration in biology.

 

[Evo]

Coming form the computer side of it, it doesn't fit in biology or brain sciences. So I disagree with it belonging in mind and brain sciences from the computer standpoint, which is truly what AI is.

 

[Moonbear]

AI has links to psychology. But it is not a psychological discipline.

And while psychology would be included within the M&B forum, the M&B forum is not only about psychology.

The study you referred to doesn't look like it concerns itself much with AI per se, even though it uses a computer model to look at biology. From the summary, the neurological model is only a model, not intended to function as any kind of artificial intelligence.

It's a thesis project, it doesn't have to be the end-all-be-all, just novel information to help build the knowledge base.

Apparently from the links you have posted, cognitive science has a connection with AI. How strong this connection is I don't know--I would suspect it is not very strong except to a philosopher. Everybody at my school doing AI research is a computer scientist. AI courses are computer science courses. To get a master's or PhD in CS, AI composes many of your core courses. Do cognitive psychologists actually produce models that are useful for real-world computational problems?

Again, you're limiting your focus to psychology. While cognitive psychology is a subdiscipline of cognitive neuroscience, there is a broader field represented within cognitive neuroscience. Cognitive neuroscience includes things such as neural architecture and computational neuroscience.

You might find this interesting:

Theories of cognition, perception, problem solving, and so on must be implementable if they are to be explanatory in nature rather than simply descriptive. This requirement is often neglected, but can place important constraints on the kinds of theories that should be entertained...
Observations and theories of human intelligence become guiding principles for artificial intelligence research helping to refine the cognitive theories. Interactions between this group and the Cognitive Science group are ongoing and fruitful.

http://www.beckman.uiuc.edu/research/hcii/ai.html

 

[neurocomp2003]

of course current teaching methods sway away from the depth in biology and psychology because for industrial application they are unnecessary but the foundations or inspirations of these fields come from the biopsych side. Remember many years ago 1950, cs was rather new.

You mentioned that the close comparison is "not so close" between NNets and Psych. Have you actually studied neuroscience or rather mathematical /comptuational neuroscience? Neuropsych or at least N-body Computational Neuropsych studies the # of neurons/synapses, how they fire, time of fire, learning types, cell formation, neuronal formation(Hebb's Rule which is one of the fundamentals of NNets). Batch Learning & Online Learning are both Child development phenomena. Specific Learning Rules also come from Child development...Unsupervised Learning, Reinforcment Learning, Supervised Learning..There's also Competitive Learning vs Cooperative Learning in cells. Neural Nets began wiht the study of the Neuron and that's where a lot of the neural nets come from(...too lazy to go get my textbook). You should look up spiking neurons though.
Studying the brain, is the study of modules which in turn lead to Layers..
The visual Cortex is all Layers, the Neocortex likewise and Auditory Layers...infact most of the brain is layered or columned. And that's where the layer concept comes from. G.Hinton(UFT) one of the fathers of Backprop has a lot of Phd students who eventually move into teh M&B realm after doing Phds in CS, he himself has published somewhat in teh field.
His algorithms tend to have Biological Counterparts. RBF i believe comes from retinal studies( but this one I'm probably wrong on)

I will agree with you that most AI students, have supervisors who are in Math, CS or CEng. But that's probably because (IMO) no old psychologist or biologist is going to be able to teach you the programming or mathematics needed to perform research in AI/ALife/Computational or Mathematical Neuroscience. Your only hope is for some co-supervision. NOwadays there are interdisciplinary programmes like CNBC,MIT,Redwood that teaches you both the Computational and Science part(CNBC i think is only 5-10 years old)...and only within these recent years have courses in Computational ### been offered for M&B.

Sorry about the GA mix up...but to me its all Adaptive Learning where I associate it with its biological Field and thus sometimes just call it M&B. GA design mimics its biological Counter Part...not just the the concept of "Survival of the Fitness" but also the processes that occur in Mutation, Reproduction etc. GAs are also used in M&B studies. LIke you said GAs are mostly for optimization like TSP problem, which is to the advantage of a M&B modeller who needs to manipulate multliple structures. Hopefully my Phd will be related to GA & NNets, if my laziness doesn't kick in first.

You speak of Mathematical Concerns. I ask you how do these mathematical studies first form.

I'm actually quite interested to know how much Neuropsych or Ecological Reading you have done?

Here are some text that I have:
Spiking Neurons: Gerstner
Reinforcment Learning: Barto&Sutton
Various Neural Nets Text: Haykin; Mehta; Anderson;
Computational Beauty of Nature: Gary Flake.
All are AI books, with the first few chapters in each listing biological concepts.

Finally I guess it depends on how you define AI to.
[0] there's the AI researcher who takes scientific concepts( our case M&B and Bio) and applies them to AI and eventually Industry.
[1] there's the AI/Science Researcher who takes AI techniques and uses them in scientific research.

If your the former, there is know need to learn and memorize all the complex biological processes, you would only require the basic knowledges.
Why would i Care to know ACGT when i just need to know that there exists this variety in creating the Helix. Or the compositions of the Neuronal Cell when I just need to know that it fires, passes info to other cells, has a time delay or refractory period and has perhaps a time to fire. Also the potential for firing would just become a threshold in a math fx.

however if your either you would require foundations in both math and cs. Which is why you would require a supervisor with Math & CS skills.

PS: You mentioned that Adaptive Learning may not properly be apart of AI, but from what I last read(mmm 4yrs ago) Adaptive Learning is the New Revolution or Age of AI,its history comes from inspirations in biology especially the Brain, Evolution, Predator/Prey, Child Development

Now I must go back to reading tensors and Nbody management before I fall behind.