← Artificial Intelligence (demo) · [ PAPER ROOM ]
Stanley P. Frankel · 1955 · Psychometrika
Can machines mimic human behavior, and what does this say about the brain?
THE ITCH THE FIELD HAD, BEFORE THIS PAPER
Imagine building a replica of a grand piano with intricate mechanisms. The more precise and deterministic the replica, the more it resembles a factory-made piano. But can we replicate the unique, human-like imperfections of a Stradivarius?
Consider a river delta, where the flow of water is unpredictable and influenced by countless variables. Stochastic models can capture this complexity, allowing us to understand and replicate the unpredictable nature of the river.
Envision a flock of birds, where individual birds follow simple rules, yet the collective behavior exhibits complex patterns. Stochastic automata with many nodes can exhibit emergent learning patterns, much like the flock's behavior.
The brain's behavior can be understood and replicated using stochastic models, which capture the complexity and unpredictability of human-like behavior.
Reach for it when
It misleads when
What it quietly disagrees with
Rejects deterministic automata as brain-like; favors stochastic models despite their lesser precision.
The bet it implies
Stochastic automata with >100 nodes will exhibit emergent learning patterns in 80% of trials.
Left unanswered
How to quantify 'learning' in stochastic models? What is minimal complexity for emergent behavior?
Oddly specific application
Designing adaptive industrial control systems with 'learning' capabilities via simple stochastic automata.
[ THE ARGUMENT, AS A MAP ]
Premises left, conclusions right. Click any claim to inspect it; drag the lens to fade the weakly-valid links and see which conclusions still stand.
Founder
Develop a minimum viable product (MVP) that captures the essence of the stochastic model, rather than trying to replicate the entire system.
WHY · Claim C3: Stochastic models can more easily be designed or understood by reason of lesser logical co
Use the McCulloch-Pitts 'neuron' as a building block for your stochastic automata, allowing for modularity and flexibility.
WHY · Claim P2: The McCulloch-Pitts 'neuron' is a convenient elementary component for the control mechanis
ProductLeader
Design a user interface that reflects the stochastic nature of the system, using visualizations and animations to convey uncertainty and var
WHY · Claim C3: Stochastic models can more easily be designed or understood by reason of lesser logical co
Implement a feedback loop that allows the system to adapt to changing user behavior, using the emergent patterns exhibited by the stochastic
WHY · Claim P1: In principle it is possible to design automata to display any explicitly described behavio
Researcher
Investigate the neural basis of learning and memory using stochastic models, focusing on the emergent patterns that arise from complex inter
WHY · Claim P5: The brain is more plausibly represented by stochastic models as proposed by Hebb.
Develop new techniques for designing and understanding stochastic models, building on the work of Hebb and others.
WHY · Claim C4: One extremely simple stochastic model is shown to have properties suggestive of learning a
Engineer
Implement a stochastic automaton with more than 100 nodes, using the McCulloch-Pitts 'neuron' as a building block, to exhibit emergent learn
WHY · Claim P1: In principle it is possible to design automata to display any explicitly described behavio
Use the emergent patterns exhibited by the stochastic automata to inform the design of more efficient and adaptive systems.
WHY · Claim P3: Previously described techniques permit the design of an automaton which would arbitrarily