At first thought, the answer seems easy. You have 100 billion neurons; Neuron has far fewer eponymous processing units. You are the product of millions of years of evolution; Neuron is an advanced system, but is the culmination of merely a quarter century of development. You have common sense; Neuron doesn’t know how to boil water. That being stated, the situation is not so simple.
To see why, we need to examine what human beings are good at and how they are different from artificial neural networks. Although we have a large, long-term memory store at our disposal (it has been estimated that we can remember up to 50,000 distinct faces), our short term memory is rather limited – it varies between a few and up to ten distinct items depending upon how they are organized internally. The implications of this for human intelligence are profound. We tend to develop models and theories from a few well-chosen characteristic examples and we are able to examine only the most salient aspects of these examples. This works extremely well when the number of variables at hand is relatively small and there is a great deal of regularity in the data. It has allowed us to develop everything from MRI machines to the theory of relativity.
But consider the situation when thousands or millions of records exist and each of these consists of many hundreds of independent variables. For example, a company may have a record for each financial transaction each with dozens of variables describing the customer’s actions. This may include demographic and past behavior information. Moreover, this data may be noisy or even contradictory, either because the information was recorded incorrectly or simply because of quirks in purchasing behavior. It is well beyond the ken of human processing to take all of this data and develop a coherent model. Of course, it may be possible to develop intuitions regarding why certain people buy certain products, but intuition will break down in all but the most obvious cases.
This is where Neuron comes in. Neuron has no trouble examining hundreds of factors, extracting out the most important of these, and determining the correlation between the input features and the desired output. In an important sense, it is Neuron’s lack of common sense – or preexisting bias – that allows it determine non-obvious relationships in the data. This combines with the ability to entertain thousand of possibilities at once to give it insight that can surpass native human capabilities.
In summary, we can state the following. When working in a relatively confined domain in which the determining factors are well-understood and cleanly relate to the desired outputs, humans will often outperform computational systems because of the flexibility of their model-building behavior. But in the increasingly common case in which vast amounts of data must be pored over to extract meaningful relationships, Neuron can often outshine even the most qualified expert in the field in question.
Tags: ANN, Artificial Neural Network, Bruce Katz, human brain, memory, Neuron