One of the most important objectives any animal faces is the faithful representation of at least some partial aspects of its external environment. Only if faithful, stable and, if possible, validated representations are available, purposeful reactions can be deduced in an ever changing environment.
By itself, the response of a single neuron is not a good carrier of information. There are two main reasons for this: the limited coding range any single neuron has, and the mixing of independent feature dimensions within single neurons.
The range of stimulus values which a single neuron can represent or process is restricted for two different reasons: (1) even though the connection numbers in cortex may be rather high, any single neuron can sample information only from a limited number of sensory receptors or other neurons. So by design, neurons have a limited access to available information, restricting trivially the ``field of view'' of neurons; (2) a further limiting factor is the input-output relationship realizable by a neuron, i.e, the neuronal transfer function. At one end of the scale, resource restrictions lead to saturation effects limiting neural responses in the range of high neural activity. On the other end of the scale, noise and threshold characteristics restrict the range of representable values from below.
To make things even more complicated, most neurons not only respond to variations along a single, fixed stimulus quality, but usually to many, rather different stimulus characteristics . For example, a single neuron might change its firing rate if the orientation of the stimulus, or the texture, or the contrast, or some combination of these stimulus properties is varied.
In contrast to most lab experiments, all these stimulus qualities are available and will vary independently in normal visual scenes. Summarising, no single neuron can give an accurate representation of any specific stimulus aspect of the external world. For this, signals from several neurons must be appropriately combined.