How prevalent is AI in electoral counting in different countries, or why is it not more prevalent ?

I see an ideal electoral counting system as one using both AI and human counters

  • At each human counters position would be a minimal camera ( containing no or minimal software ) connected by a wire to a software-box maybe using only ROM memory for the software ( obviously to avoid illicit changing of the software ), the camera would be looking straight down at the desk
  • The camera would record one image for each ballot, each image would show both the ballot and the paper? on which the human adds that ballot vote.
  • Obviously, the software would determine / read the vote on the ballet, and also determine / read the paper? on which the human adds that ballot vote, to see if it agrees with what the human has determined / read.
  • The software-box would also give an audio or visual alert to the human counter if it detects that they have made an incorrect reading of the ballot paper
  • IT WOULD BE A COMPLETE AUDIT SYSTEM - This would be the most perfect system for real-time-audit and stored-audit
  • Also, a mirror could be placed at angle, on the desk, so that the image of the human-counter could also appear on the stored images of the ballots.
  • Also, perhaps, more than one set of software could be installed in the software-box, and they could both function of the same camera-feed, and a main-checking-program checking if both sets of software agree with each other
  • Also, ideally the camera be a type that contains no or minimal software, if any minimal sofware is required inside the camera, it could be solely on ROM memory
  • 2
    While generally speaking questions which need to push ideas or convince readers something is a good idea are considered inappropriate Stack Exchange questions, if you just ask if AI is being tried the first comment will be "What makes you think that it would be a good idea?" so I can see why you've done so :-)
    – uhoh
    Oct 9, 2022 at 6:38
  • As someone in ICT, I will refer you to xkcd.com/2030
    – Shadur
    Oct 9, 2022 at 9:30
  • Also, from an ICT perspective your specific suggestions are terrible but this isn't security.se so I'll spare you the lengthy rant. I recommend not cross-posting this suggestion there unless you like having several strips torn off of your hide.
    – Shadur
    Oct 9, 2022 at 9:34
  • 2
    Do you have any idea how much software (in the form of firmware) is in a digital camera? Also, where's the AI in this system? Electronic ballot counting systems do not need AI. The preference is to keep it short and simple, aka the KISS principle. Oct 9, 2022 at 12:45
  • 1
    In reference to @Shadur’s XKCD, I will note that aeroplane design safety standards are written in blood and I would not be surprised if elevator design standards are written in blood too. However, most coding is not and likely never will be.
    – Jan
    Oct 11, 2022 at 11:52

2 Answers 2


Counting is not something that requires intelligence, human or artificial. It requires precision. Counting is something that ordinary software is very good at, and automatic vote counting machines are fairly common around the world.

There are a couple of tasks that require intelligence: one is reading numbered ballots, and this is probably a reasonable task for an AI. But it would be much simpler to use a ballot paper that doesn't require handwritten numbers.

Another task that requires intelligence is processing spoilt votes, ambiguous votes and votes which may need to be removed for legal reasons. This, however, requires judgement and quasi-legal interpretation and isn't suitable for an artificial system.

I've looked at your plan, and it seems to be an outrageously complex way of solving a problem that doesn't really exist. If you have decided to use a computer system to count votes, you would just design your ballot paper to make it easy for a computer to count votes and not have some wildly complex (and therefore error prone) camera.


A good role for AI might be in the detection of errors and fraud.

I recall a case from Germany where one precinct reported party A at almost the overall average percentage for party B, and vice versa. District officials made that precinct check their numbers immediately on election night and they found a transciption error in a reporting form. This was an "easy" error to catch. AI might flag less obvious ones for human attention.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .