As other answers have pointed, on UNIX systems the numbers represent CPU load averages over 1/5/15 minute periods. But on Linux (and consequently Android), what it represents is something different.
After a kernel patch dating back to 1993 (a great in-depth article on the subject), in Linux the load average numbers no longer strictly represent the CPU load: as the calculation accounts not only for CPU bound processes, but also for processes in uninterruptible wait state - the original goal was to account for I/O bound processes this way, to represent more of a "system load" than just CPU load. The issue is that since 1993 the usage of uninterruptible state has grown in Linux kernel, and it no longer typically represents an I/O bound process. The problem is further exacerbated by some Linux devs using uninterruptible waits as an easy wait to avoid accommodating signals in their implementations. As a result, in Linux (and Android) we can see skewed high load average numbers that do not objectively represent the real load. There are Android user reports about unreasonable high load averages contrasting low CPU utilization. For example, my old Android phone (with 2 CPU cores) normally shown average load of ~12 even when the system and CPUs were idle. Hence, average load numbers in Linux (Android) does not turn out to be a reliable performance metric.