Emergence: From Chaos to Order

I finished Emergence: From Chaos to Order a couple weeks ago, snippets that I want to remember include:

On model building: “For most of us model building starts at an early age. As children we use building blocks to generate concrete realizations of our imagination — castles and space stations. This facility for recombining standard objects to make new items carries over into late occupations. A watchmaker uses familiar mechanisms — gear wheels, spring, pinions and so on — to generate marvels of timekeeping, and a scientist does the same thing at a more abstract level, generating complex objects, such as molecules, from simpler objects, atoms. By selecting the building blocks and the ways of recombining them, we set up the rules that make rule-governed systems comprehensible. A well conceived model will exhibit the complexity and emergent phenonmena, of the system being modeled, but with much of the detail sheared away.” [pg 12]

In relation to the checker playing model, the emergent consequences of weight changing: “Subgoals: Though it seems that the evaluation function makes no provisions for subgoals, in fact it does, providing subtle direction when there is no clear path to a win or an obvious advantage. Anticipating the opponent: The checkersplayer must impute a strategy to the opponent if it is to anticipate the opponent’s actions; the valuation function can serve as a guide to the opponent’s likely response. Toward minimax: The valuation function only indirectly minimizes the maximum damage the opponent can inflict (minimax), yet it captures important elements of this idea. Bootstrapping: The checkersplayer can improve its performance by playing against itself. Lookahead: Knowing the rules of the game, the checkersplayer can look ahead several moves using its model of the other player, changing weights on the basis of the anticipated outcomes.” [pg 68-69]

On anticipating the opponent: “An undesirable outcome can only occur because a) the checkersplayer has made a bad play, or b) the opponent has made a good play. In either case, the checkersplayer is well advised to assign a low (possibly negative) value to configurations that lie in that direction. On the other hand, a desirable outcome presents an ambiguous situation. The desirable outcome may occur because of good play on the part of the checkersplayer, but it can also occur because of poor play on the part of the opponent. If the outcome is the result of an opponent’s poor play, then it is unwise to make any adjustments. That line of play is unlikely to recur, either because the opponent has learned or because a different player plays a better game.” [pg 70-71]

On neural nets and transmitter molecules: “Transmitter molecules diffuse across the synaptic gap to the surface of the receiving neuron; if enough transmitter molecules from enough synapses accumulate at the surface of a neuron, that neuron fires. In doing so, if effectively removes the molecules from the synaptic gap. If this happens repeatedly, the synapse increases its ability to produce the transmitter molecule…. The synapse has increased its effectiveness (weight)…” [pg 95]

On the mechanisms of a neural net: “Variable threshold. A neuron’s threshold decreases as the time since it last fired increases. This decreasing threshold makes the neuron increasingly sensitive to incoming pulses when it remains quiescent over an extended period. The variable threshold allows the neuron to act as a frequency modulator, firing at a rate that reflects the average synapse-weighted strength of the impulses firing on its surface. Fatigue. A neuron that fires at a high rate over an extended period has its threshold steadily incremented — in effect the whole variable threshold curve is translated upward. Contrariwise, a neuron that fires at a low rate over an extended period of time has its threshold steadily decremented. Fatigue eventually forces a neuron’s firing rate back to a ‘normal’ or ‘set-point’ level: no neuron can continue to fire at a rate above or below this set-point. Hebb’s rule. If neuron x fires at time t and neuron y fires at time t+1, then any synapses that x’s axon make at y are strengthened. Contrariwise, if x fires at time t and y does not fire at time t+1, the same synapses are weakened.” [pg 108]

On emergence: “… The behavior of an ant colony is not the simple sum of the behaviors of a group of average ants. The coupled interactions of the ants provide a coherence to the nest that far exceeds anything predictable in terms of simple summations. Emergence is above all a product of coupled, context-dependent interactions. Technically these interactions, and the resulting system, are nonlinear. The behavior of the overall system cannot be obtained by summing the behavior of its constituent parts. We can no more truly understand strategies in a board game by compiling statistics of the movements of its pieces than we can understand the behavior of any ant colony in terms of averages.” [pg 121-122]

On cultivating innovation: “Practice. A part of the answer to all who strive within a discipline, be it tennis, piano playing, writing poetry, or building a scientific model. The answer lies within the word itself: discipline. Only when you are so familiar with the elements (building blocks) of your discipline that you no longer have to think about how they are combined, do you enter the creative phase. If you are a tennis player and you have to concentrate on the elements of each stroke, you will have little appreciation of the flow of the game — your opponent’s strengths, weaknesses and strategy. If you play the piano and have to concentrate on fingering, you will not hear the flow of the music, the ‘long line’. Local concerns drive out global perceptions, and so it is with other disciplines.” [pg 211-212]

More on innovation: “Once a set of building blocks has been chosen, innovation depends on selection from among the plethora of potential combinations. The possibilities are so numerous that the same building blocks can be used over and over again without seriously impairing the chances for original discoveries. Think of the standard building blocks provided by words in a language, or folk themes in music. The key to handling this complexity is the discovery of salient patterns in the tree of combinations. Creative individuals exhibit talent for such selection, but the mechanisms they employ are largely unknown.” [pg 217-218]

Leave a Reply

Your email address will not be published. Required fields are marked *