Mind vs. Machine

Mind vs. Machine

A computer can multiply large numbers much faster than a person doing mental arithmetic. That comparison does not rank all of the work performed by a brain against all of the work performed by a computer. To compare them meaningfully, we need to specify the task, what counts as an operation, and the resources being measured.

The adult brain weighs roughly three pounds. A frequently cited estimate puts fat at about 60% of its dry weight, rather than of the whole water-containing organ. It contains an estimated 80–100 billion neurons. Gray matter includes neuronal cell bodies, dendrites, and synapses; white matter consists largely of myelinated axons connecting regions. The author cites an estimate of 160+ trillion synaptic connections. Axons, rather than exons, carry signals between neurons. This interconnected tissue supports memory, perception, movement, and reasoning; the ends-and-means heuristic is one description of how some reasoning tasks are approached.

The network’s estimated 160+ trillion synaptic connections do not translate directly into computer storage units. Memories depend on distributed activity and changes across neurons, rather than each neuron storing a single item. The author cites an estimate of around 2.5 petabytes, on the order of a million gigabytes, and compares that with roughly 300 years of television or about 3 million shows. These are scale analogies, not evidence that the brain records and retains every second of life. Storage, forgetting, and retrieval are distinct processes, and difficulty remembering cannot always be reduced to weak retrieval. Computer storage is expandable but finite, and a useful comparison must account for how the information is represented.

Counting machine operations and timing conscious human thought measures different things. A computer may perform many operations in parallel, while a person may use learned patterns without consciously examining every step. The Kasparov–Deep Blue comparison is useful only if we specify whether we are comparing search, evaluation, pattern recognition, elapsed time, or energy. Like comparing writing a novel with retyping it, a shared output does not make the underlying tasks equivalent.

Deep Blue could devote close to 100% of its computational resources to chess. Kasparov’s brain also supported perception, movement, and bodily regulation. The author describes a 2014 experiment on the K Computer, using nearly 83,000 processors, that took 40 min to simulate a second of neural-network activity. The machine is described as weighing roughly 100 tons 8. Those 83,000 processors and 40 min of computation do not provide a direct comparison with an entire functioning brain: the simulation and a biological brain are different systems. The author also cites an estimate of a quintillion, or 10¹⁸, calculations per second for the brain.9 Such estimates depend on what is counted as a calculation; they do not establish a general brain clock speed or a ranking against current supercomputers.

Energy offers a separate comparison. Neural tissue combines storage and processing in ways that differ from conventional computer architectures, but that alone does not specify how many operations a task requires. Comparing the energy used to simulate 1 second of network activity with a brain’s power consumption can illustrate the cost of simulation. It does not mean the two systems performed an identical task, and computers also vary in how they organize memory and computation.

For chess, performance depends on the combination of search, evaluation, memory, and computing resources. We cannot infer from a rough brain-capacity analogy that speed and storage do not matter. Nor does a machine’s win establish that its evaluations are unbiased.

During his career,10 Kasparov acquired a large repertoire of positions and strategies. Learning changes neural connections, but neuroplasticity is not simply a process that taints once-objective memories. The Deep Blue team studied Kasparov’s games and designed a search and evaluation system to play against him. That system reflected choices made by its designers. Its consistency differed from human judgment, but consistency and objectivity are not the same thing.

Deep Blue won two of the six games in the match. The score alone cannot tell us how much of Kasparov’s performance came from any particular heuristic. The ends-and-means heuristic offers a useful account of part of human play: recognize a promising position—the end—then look for moves that could bring it about—the means. This can narrow the search to a manageable set of candidates. The position and the opponent’s replies still need to be assessed; recognizing a desirable pattern does not establish that it can be reached.