Human tallying is a source of errors, but it typically doesn't affect the outcome in major ways. This is more of an argument against large scale winner-takes-it-all election systems, as they have the least resilience against this kind of error.
The main benefit of manual tallying is that election tampering at scale becomes a rather labor-intensive and physical process that is more likely to leave detectable traces. Compare that to the the last US presidential election that has statistical oddities in machine-tallied voting results of kinds that have historically been shown to correlate with election fraud. If this was indeed caused by fraudulent voting software, it happened without leaving any other obvious traces of tampering.
> Compare that to the the last US presidential election that has statistical oddities in machine-tallied voting results of kinds that have historically been shown to correlate with election fraud.
It's being litigated, but in general the answer is there is not yet evidence that machine voting systems were compromised.
- in New York there is statistical anomaly correlated with a couple small-town polling stations. Those towns are small enough that they have a huge population of one religion, and one explanation is that the Democrat party's perceived "soft on Israel" stance tilted 100% of voters in those locations away from supporting the Democrat presidential candidate.
- in Pennsylvania a standard statistical analysis tool used to detect vote disruption suggested disruption occurred. The form of the disruption could be fraud, but it can also be things like voter intimidation (which was observed and reported in Philadelphia) and sudden discontinuity in voter behavior (the aforementioned "soft on Palestine" issue).
Correlation does not imply causation, and the lack of evidence of tampering of the machines in the audit logs is lack of evidence of tampering of the machines, not indication that the audit logs were compromised.
The main benefit of manual tallying is that election tampering at scale becomes a rather labor-intensive and physical process that is more likely to leave detectable traces. Compare that to the the last US presidential election that has statistical oddities in machine-tallied voting results of kinds that have historically been shown to correlate with election fraud. If this was indeed caused by fraudulent voting software, it happened without leaving any other obvious traces of tampering.