Automatic Signature Verification Software spark Controversy amid US Elections
Experts view lack of Transparency and discrepancies in the voting system with the software.
The world is going to witness the most peculiar and interesting elections in the first week of November. With only a week left, all the eyes have turned towards the US elections 2020. The past couple of weeks were back to back lined with sweltering debates, between the incumbent President Donald Trump, and the Presidential Candidate Joe Biden. With both the candidates actively campaigning, physically and through social media, the fairness of the elections is still under the radar of unambiguity.
Certainly, the USA citizens did not forget the involvement of Cambridge Analytica in the USA elections 2016. That’s why the USA elections commission is trying everything to make this election successful. From Blockchain to the integration of IBM Watson for helping voters with their queries, the USA elections has become one of the most technological advanced elections.
Despite this, as of early October, more than 84.2 million absentee ballots had been either requested or sent to all the 47 states including the District of Columbia for the US presidential elections. Given the current situation of COVID 19, and the constantly rising cases with a potential second wave, slowness in the election tally is inevitable. The tasks like verifying voters, and segregating their information would take considerable time than anticipated in the Pre-COVID era.
The authorities have swung Automatic Signature Software, to expedite the verification process of the voters. And though the technology advances are verifying voters in many aspects, its algorithm is scrutinized under the lens of uncertainty.
Understanding Signature Verification Software
Commonly known as the Offline Signature verification, it relies on the scanned image of a signature. A paper named “An offline Signature Verification System” described the certain geometric features such as the baseline slant angle, aspect Ratio, Normalised Area, the centre of Gravity and slope of the joining lines, on which the dataset works. Offline signature algorithms are trained on the dataset of those signatures, which are authorised by the system. The paper cites that for each subject a mean signature is obtained integrating the above features derived from a set genuine sample signature. This mean signature acts as the template for verification against a claimed test signature. Euclidian distance in the feature space between the claimed signature and the template serves as a measure of similarity between the two. If this distance is less than a pre-defined threshold (corresponding to the minimum acceptable degree of similarity), the test signature is verified to be that of the claimed subject else detected as a forgery.
Moreover, a recent study by the Central Police University’s Department of Forensic Science in Taiwan has found that the datasets trained on national and international signatures display 89.5% to 99.6% accuracy rate. The study is based on the deep-learning method i.e. Deep Convolutional Neural Network and unique local feature attraction approach.
But just like any technology, the automatic verification software relies heavily on data. Coherent and reliable data will give a more precise and accurate outcome. As Big Data is heavily involved while retrieving information about the voters, it becomes imperative that the datasets must be trained with structured data. A study published by EURASIP Journal on Advances in Signal Processing concludes that the accuracy of algorithms varies depending upon the data that is used.
Uncertainty about the Software’s Database and Algorithm
The USA elections are complex and so is the technology. One major drawback using any technology is that most of the algorithm and datasets are trained using only one language i.e. English, with every stroke and feature of the word precisely analyzed. This means that the software has a higher chance of rejecting those ballots where the signature is not done in English.
Portia Allen Kyle, who also leads the American Civil Liberties Union(ACLU), emphasized, that certain voters with mental and physical illness, stress-related ailments and those who are at a disadvantage of being illegible at English, would be at a higher risk of getting their ballots rejected. Moreover, those whose names are short and includes hyphen in their name will have their ballots deemed unfit during US elections 2020.
Over the years, the electoral council has tried to lessen the number of Absentee ballots. Undoubtedly, the USA elections are extremely complex, with each state having its own rule and law regarding the voting process. For example, in certain states, people charged with a felony, or immigrants are not allowed to vote.
And despite the many efforts, the dearth in regularized conduction of votes has accumulated a surge in absentee ballots. For example, an NBC report states that between the year 2016 and 2018, more than 7, 50,000 absentees’ ballots were discounted due to signature discrepancies. Another survey by ACLU states that almost 47% ballot of coloured voters were rejected citing discrepancies.
A study by Stanford University’s Law and Policy lab states that the automated signature verification has 74% chances of rejection in counties which lacks human review.
Software Vendors such as Parascript, a Colorado Develop of Document Capture and Recognition solutions have been positive for conducting fair elections. But experts say otherwise, questioning the transparency of such software.
Liz O’Sullivan, the co-founder of Allen-Kyle and Surveillance Technology Oversight Project stated that “Even from a non-technical standpoint, signature verification powered by AI or any form of automation is more likely to flag folks who have undergone a name change. This means that married women, trans people, or domestic abuse survivors will all be disproportionately likely to have their vote cast out.”