Transcript How to Create a Mind The Secret of Human Thought Revealed

How to Create a Mind
The Secret of Human Thought Revealed
by Ray Kurzweil
Slides by Prof. Tappert
1. Thought Experiments on the World
• Einstein’s thought (Gedanken) experiments led to the special theory
of relativity
• Can we take the same approach with respect to the human brain to
determine how it works?
2. Thought Experiments on Thinking
• Memories are sequential and in order – sequences of patterns
• From observing, for example, that reciting the alphabet backwards is not easy
• We can recognize a pattern even if only part of it is perceived
• E.g., images of portions of faces
• Experience of our perceptions is changed by our interpretations
• We are more likely to perceive what we expect
• E.g., complete the sentence: Consider that we see what we expect to __
• We remember routine procedures as a hierarchy of nested activities
• Like preparing for sleep
• And hierarchies are involved in recognizing objects and situations
3. A Model of the Neocortex
The Pattern Recognition Theory of Mind
• The cortical column and the mini-column
• The neocortex has extraordinary uniformity of column-like structure
• In 1978, Mountcastle hypothesized the neocortex composed of the cortical
column as the basic unit, each containing many mini-columns
• Kurzweil contends the cortical mini-column is a pattern recognizer
• The human neocortex contains a half million cortical columns
• Each containing 60k neurons (the neocortex has about 30 billion neurons)
• A mini-column contains ~100 neurons, so neocortex has ~300 million pattern recognizers
• Compare with human experts
• Chess master Kasparov learned about 100k board positions, Shakespeare
used 100k word senses, medical physicians master about 100k concepts
• With redundancy of 100 to 1 and requirements for general knowledge, we
need over 100 million pattern processors
3. A Model of the Neocortex
The Pattern Recognition Theory of Mind
• A model of the neocortex
• Kurzweil favors the hierarchical hidden Markov model (HHMM)
• Jeff Hawkins and Dileep George proposed similar model
• Another model (not from this book)
• Deep learning neural network models, basically multi-layer perceptrons
(MLP), also demonstrate many characteristics of the human neocortex
• Especially the deep-learning convolutional layers
4. The Biological Neocortex
• https://en.wikipedia.org/wiki/Neocortex
•
http://www.brainwaves.com/
5. The Old Brain
• A portion of the human brain is pre-mammalian
• Located underneath the neocortex
• Provides motivation for seeking gratification and avoiding danger
6. Transcendent Abilities
• Our emotional thoughts occur in the neocortex
• Influenced by regions of the old brain, like amygdala
• And some evolutionary new brain structures, such as spindle neurons
• Aptitude
• Creativity
• Love
7. The Biologically Inspired Digital Neocortex
• Brain simulations –
• Markham’s Blue Brain Project (HBP) – full brain simulation expected 2023
• HBP Problem Christopher deCharms
•
•
•
•
•
Neural nets – Rosenblatt’s perceptron, AI winter, backprop’s resurgence
Sparse coding – vector quantization (basically clustering technique)
Hidden Markov model (HMM) & Kurzweil’s hierarchical hidden Markov model (HHMM)
Genetic algorithms (GA)
Hierarchical memory systems – Numenta’s Hierarchical Temporal Model (HTM)
• Jeff Hawkins
Dileep George Dissertation
• The moving frontier of AI – IBM’s Deep & Watson, etc.
• A strategy for creating a mind
8. The Mind as Computer
• Computers are sometimes regarded as thinking machines
• This chapter compares the computer and the human brain
9. Thought Experiments on the Mind
• This chapter explores consciousness
• Usually considered the capacity for self-reflection, the ability to understand
one’s thoughts and explain them
• Some therefore believe a baby and a dog are not conscious because
they cannot describe their thinking process
• IBM’s Watson can be put into a mode where it explains how it comes
up with a certain answer. Is Watson therefore conscious?
10. The Law of Accelerating Returns
Applied to the Mind
• Law of accelerating returns – growth of information technology
follows predictable exponential trajectories
• The best example of this is the double exponential growth of the
price/performance of computation, steady for 110 years
• Through paradigms so far: electromechanical relays, electronic switches, vacuum tubes,
individual transistors, integrated circuits (Moore’s Law)
• Applied to the mind
• Brain scanning technology resolution improving at exponential rate
• Advances in AI technology, also progressing exponentially, influence our
understanding of brain operation principles
11. Objections
• Human intuition is that advances are linear rather than exponential.
Many critics, therefore, have difficulty accepting the exponential
projections.