These system used for voting means that humans don't hand tally hundred of millions of votes. They tally those in a voting district only. Those them get aggregated with other districts and so on until the whole states and then the country is counted.
The problem with the accuracy assumption of electronic voting is that a) its all coded without errors and b) someone hasn't deliberately but code into manipulate the vote numbers.
That pretty much undermines the entire concept of unit and integration testing.
If you're saying we should be writing voting machine code in ML and keeping the firmware in Fort Knox, I'm going to make the argument that it's a lot cheaper to do sampled hand-counts to check against machine error or tampering... Which we already do.
> That pretty much undermines the entire concept of unit and integration testing.
It doesn't because even with unit and integration testing, software still fails and get hacked/exploited on a regular basis. What's worse with software voting is that a single good exploit could affect all the votes.
It really can't. What is the mechanism by which it could? Different states use different machines and each individual machine is airgapped. A general supply chain attack would show up in statistical sampling audits (i e. hand recounts of small numbers of ballots, which we already do).
The problem with the accuracy assumption of electronic voting is that a) its all coded without errors and b) someone hasn't deliberately but code into manipulate the vote numbers.