It is sometimes meant that loss is round-trip. Round-trip loss is
the probability of losing said packet or a reply sent to
it. (This is loss as measured by the ping
program.)
From the outside point of view, loss is just loss. However, from the network perspective loss can be further categorized as random loss or loss due to congestion. Random loss is the result of equipment malfunction, radio signals corruptions, bad cables, etc. It is always bad to have it. On the other hand, since IP doesn't provide any explicit way to indicate that the network is congested (barring ECN), routers use loss for that purpose. Such loss is a way of communicating back to the users the need to scale back offered load on the network.
Assuming steady network state, bulk TCP throughput is inversely proportional to the square root of loss. (Loss here is the unidirectional loss from the sender to the receiver, assuming loss is the other direction is "reasonable".) For that reason, even small random loss can kill TCP performance.
Loss is an interesting parameter from QoS perspective.