Identity fraud is a problem that affects over 16.5 million US adults, and this rate is growing at about 8% per year. This is partly due to unprecedented numbers of black-hat hackers, to whom making fraudulent transactions, breaching stored data on large scales and stealing and selling identities makes their bread and butter. Artificial intelligence (AI) technological advances have provided businesses with methods they can use to stand strong against hacker activity and identity fraud. They do so in a few ways:
Compliant machine learning
Official documents that support identities (e.g. passports) are typically scanned by businesses that need to comply with Know Your Customer (KYC) regulations. Knowing KYC compliance obligations is crucial when conducting B2B or even B2C transactions, as they help markets avoid funding terrorism or facilitating money laundering and corruption. KYC obligations require keeping official document copies as well as basic data about customers, but this data can also prove to be independently valuable to the companies utilizing them.
Machine learning can create a more efficient and accurate process, as it relies on automated document checks as opposed to the human eye (upon which many organizations still rely). AI can combat identity fraud with compliant machine learning by managing an anonymous data-collection mechanism that stores metadata about the software’s performance. When this is automated by AI it can improve the quality and reliability of results.
Relying completely on machine learning tended to lead to marking documents that have experienced wear and tear (e.g. an old passport) as untrustworthy or suspicious. That’s why good AI processing involves a feedback loop, where data is fed back into the system (often with semi-supervised learning) and processed with regression analysis to improve its algorithm.
Biometric identification can be with face, voice or fingerprint recognition, though voice recognition is used tentatively as it can be more easily forged. All AI biometric security mechanisms use deep learning to replicate the way that the human brain processes very complex information like face structure. Facial recognition may be the best way for AI to combat identity fraud as it is excellent at quickly finding patterns between the eyes mouth and nose, as well as comparing complete facial features and shapes. The AI software then improves its own processes and speeds up the time it takes to look at and confirm that two pictures are faces that belong to the same person. Facial recognition is also easiest to implement to confirm an identity, as a picture on a driving license or passport is much easier to obtain than government-approved fingerprint records.
Is AI enough?
Artificial intelligence is, at the moment, simply a tool. It is a technology that can massively improve automation and itself automate processes that would require human intervention (as in the case of facial recognition), but trained professionals are still needed to use AI software. This is especially true when dealing with customers, who would no doubt complain if a legitimate ID had been flagged for being fake. Humans are still required to identify why the ID has been flagged, and to teach the computer how to identify the error next time. For now, this is still a vital process in combating identity fraud with AI but watch out for future developments.