Your Brain is a Time Machine: The Neuroscience and Physics of Time

by Dean Buonomano

of cause and effect is built into our synapses. Consider the two neurons shown in Figure 2.2: neuron A is connected to B, and B in turn to A. Thus there are two synapses: A→B and B→A. We would say these neurons are recurrently connected: neuron A is the input to neuron B, and vice versa. Now let’s assume that each neuron is driven by distinct events in the outside world. Perhaps the owner of these two neurons is a baby named Zoe, and neuron A is driven by the sound of the letter z, and neuron B by the sound of the letter o; thus, whenever Mom and Dad say Zoe’s name, neuron A will fire right before neuron B and, for argument’s sake, let’s say that neuron A consistently fires 25 milliseconds before neuron B. The job of a synaptic learning rule is to strengthen or weaken synapses, according to the pattern of activity of the presynaptic and postsynaptic neurons. In this case STDP will preferentially strengthen the A→B synapse and weaken the B→A synapse. It took neuroscientists a surprisingly long time to hit upon this simple learning rule. It was only in the 1990s that STDP was conclusively demonstrated.14