and gender, their use in the machine learning process could allow artificial intelligence to refine its understanding of the human face and become more accurate in matching samples. Seeing as facial recognition technology is severely compromised by its inherent racial bias, many pioneers and researchers in the industry suggest this is a crucial step before it can be implemented in a public safety application. Early statistics definitely support the basis of the technology and encourage us to pursue the application of it universally.
49 Fig. 6. Plot demonstrating error vs. diversity of reference parameters in a biometric system However as of now more research and resources need to be allocated toward the refinement of the technology for the precision to reach a point where it can be implemented in more crucial fields such as homeland security or law enforcement. As seen in Fig. 6, there is a clear correlation between the diversity of a reference database and the prevalence of classification errors that occur during a facial scan. By increasing the exposure of the system’s reference database to a greater variety of ethnic populations, the accuracy of facial recognition technology has the potential to increase to the point of reliable accuracy. While companies like Microsoft and IBM have already made efforts to improve its software by personally developing new datasets with more racial diversity to retrain its artificial intelligence program, some like Amazon fail to recognize the impact that these accuracy issues in facial recognition technology can have in society and government. 5.1.2 Biometric Data Security In order to combat the security issues mentioned previously, the two most effective ways to avoid a failure in a biometric system are to purposely distort the stored data and employ more advanced
50 encryption strategies. By distorting the data in a highly specified manner that is only understandable by the system in which it is stored, it is much harder for hackers to access and abuse it. The transforms used to distort each image is unique to each person, making theft more difficult because if one person is compromised, others remain safe. Alternatively, if the transform function used is compromised, it is possible to overwrite the compromised data and apply a new transform, effectively creating a new person in the system that can no longer be accessed using the old stolen data. This distortion process ensures that the image a hacker would receive is not the image that will provide them access to the system they are ultimately attempting to infiltrate. This encryption system also makes cross-matching impossible because the original image never existed in the system and therefore cannot be extracted. If applied, this would put the public’s mind at ease concerning the possibility of the government taking this data from consumer applications and abusing it for their own applications in national security and criminal investigation.