Neural Networks: The New Connectionism
Psychology of Learning
Computer Simulation and Artificial Intelligence
Artificial intelligence (AI) tries to understand the ways in which brains are like computers.
based on older research
Computer simulation attempts to mimic the functioning of the human, whereas AI is concerned with devising computer hardware and programs that can accomplish the same things as humans can.
AI is a branch of computer science that tries to make computers smarter.
Myths of AI
Computers are nothing more than computational machines
Computers are slaves
Some are programmed to play chess and can beat their programmers
Some don’t operate in a sequential fashion, but their processing is distributed over a large number of connections simultaneously leading to parallel distributed processing (PDP).
Making Computers Smarter
Why make computers smarter?
Frees people to move on to other bigger and better things.
Learn more about human cognitive processes.
Can machines think? The Turing Test
Computers cannot think, but they can be programmed to respond as if they were thinking.
If people can think, and if a machine can be developed to do everything that a person can do when the person is thinking, then the machine must be able to think.
Does a computer need to think?
The Computer and the Brain
Complex arrangements in computer vs. brain:
electronic components, called hardware: chips, disks, drives, switches, etc.
neural material, called wetware: neurons, various other cells, amino acids, chemical transmitter substances, etc.
Flow of ideas in computer vs. brain
computer: input and output
human: stimuli and responses
computer programs, or software = human cognitive processes, or information processing
A computer that responds like an intelligent human being might function as does a human…or…might use very different processes in its responses.
Computers are fast, brains are slow (100,000x slower)
Brains can multi-task on many levels, computers cannot.
Human ability to store information in memory is virtually unlimited; no computer even comes close.
Human ability to perceive and to recognize complex, changing patterns cannot be matched by computers.
Modern robots cannot compete with humans’ ability in locomoting, recognizing people and objects, discriminating shapes and smells, etc.
Computers have ability to retrieve flawlessly from memory and perform arithmetical computations rapidly and accurately, much more so than humans.
Parallel Distributed Processing (PDP)
The brain is more like an analog computer rather than a serial processing computer.
Analog (PDP) computer does a whole bunch of things at the same time.
The brain doesn’t do things one after another with lightening rapidity but multi-tasks.
Initially, all simulations of human thought were digital; now PDP computers are available
Symbolic models (also called production system or declarative knowledge) are:
Early models of human thought processes reflect characteristics of digital, serial-processing machines, such as the common personal computer.
Models that can carry out numerous operations in a very short time; operate with the strength of brute force.
Models that basically assume that all meaning, and thus all thought-processes, can be represented by symbols such as language.
Logic Theorist and the General Problem Solver
Logic Theorist and GPS are both programs to solve problems; Logic Theorist could only solve mathematical problems, so GPS was developed.
GPS was like humans in several ways
solved some problems, but not all
did better if information was presented systematically
performed better with instructions that provided direction
used concepts in solving problems (assuming axioms can be considered concepts)
Neither Logic Theorist nor GPS revealed anything new about human problem solving; both did only what they were programmed to do.
SOAR is a theory that summarizes the processes and components of major symbol-based information-processing models.
SOAR describes human cognitive system, or cognitive architecture, using 10 components.
cognitive architecture is “description of the cognitive system at an abstract, usually symbolic, level
SOAR is based on assumption that all knowledge can be represented by symbol system.
At any given point there are a limited number of possible moves (approx. 10 to the 120th power), a finite number of possible countermoves, etc.
algorithm = problem-solving procedure in which all alternatives are systematically considered
heuristic approach = makes use of various strategies that eliminate and select from among alternatives.
Modern chess programs can mechanically compute millions of moves and countermoves within a few seconds, relying on brute force.
Deep Blue, IBM chess computer described as a “massively parallel processing” machine won against world champion Garry Kasparov in a rematch in May of 1997.
Humans cannot rely on brute force but can recognize patterns on the chessboard based on previous similar experiences.
Humans can select the best move from experience, synthesizing the effects of previous experience without following explicit rules.
Fuzzy logic is relativistic, considers a variety of factors, and does not have predictable probability of being correct.
Previously, cognitive scientists assumed that logic governing human cognitive activities would be precise, predictable, and machine-like.
If computer scientists are to investigate the human form of thinking, they need something other than the algorithm-driven, symbol-manipulating, calculating machine that is the serial digital computer.
To investigate human thinking, they need a fuzzy-logic driven device.
To emulate human thinking, computer scientists need the PDP computer which uses the connectionist model of cognitive architecture.
PDP models consist of a set of processing units whose structure (architecture) is designed to mimic that of the brain’s neurons.
Thus these separate units can be considered to represent words, letters, sounds, etc.
Neural Network Models
Thorndike used the term connectionism in terms of stimuli and responses; new connectionists are concerned with connections among neural units.
Connectionist model assumes that brain’s collection of neurons is like the processing units in a PDP computer, forming a neural network.
No central organizer governs their activities, but instead these units are simultaneously active.
Learning occurs in 3 ways: new connections might develop, old connections might be lost, or the probability that one unit will activate another might change.
Computer-based tutoring systems
Computer systems that mimic other phenomena
Computer-based simulations of various environments, labeled virtual reality (VR)
Intelligent tutor systems
Teach students programming skills and computer literacy
Through fuzzy logic in programs, computers might truly become intelligent.
Given the right series of experiences, the neural network computer might reach insight; may be better models of human cognitive processes than symbol-based models
suggest that people don’t always think rationally
allow for fuzzier logic
A neural network that adjusts its own connections is highly compatible with Hebb’s notion that neurons that repeatedly activate each other become increasingly more likely to do so.
Connectionist, or neural network approaches, now dominate study of human cognitive processes.
Computers don’t simulate human emotions all that well.
Computer simulations don’t reveal the insight of which human problem solvers are capable
Computer programs tell us very little about how the human nervous system works
Neural network models may generate results that are unpredictable and have never been observed in a laboratory
The model may teach psychologists little about the processes involved in learning.
Problem of interference