Before the covid19 pandemic, human beings were traveling more than ever- and demanding both convenience and security in their transactions. Gone are the days when all you had to was buy a ticket and roll into the airport- now you need multiple IDs and multiple ID checkpoints at airports to ensure that travelers are whom they say they are. This need for security increases the labor costs that facilities e.g., airports incur and pass on to consumers and taxpayers. Because of new global threats, we need more robust security systems.
Idemia is a $3 billion revenue company born of a series of mergers and acquisitions involving OT (Oberthur Technologies) and Safran Identity & Security (Morpho) in 2017. The company had 14000 employees in 2018 and served clients in over 180 countries. Its claim to fame is the ability to provide an “augmented identity,” which is, according to the company website, “an identity that ensures privacy and trust and guarantees secure, authenticated and verifiable transactions.” Idemia provides its security products to telecommunication companies, finance companies, law enforcement, local and national governments. Chances are, if you’re reading this blog, you have interacted with technology from this company.
Idemia’s value creation occurs in the following ways:
- More robust user identification with fewer errors than humans and identifying multiple faces at the same time.
- Rapid identity verification at multiple use cases- e.g., stores, voting centers & immigration centers with a smaller human intervention which lowers labor costs
- Leveraging multiple forms of biometrics simultaneously to verify identity e.g., iris scans from a distance, can complement fingerprint data to ascertain that you are you.
- With covid19, the company’s contactless payment uses biometrics to provide a high-throughput, hygienic way to transact.
The company makes revenue through sales of devices, and installation and support charges. It has multiyear contracts with its B2B customers as well as governments.
On the AI front, we generate an increasingly vast repository of personal data. The company has to find ways to learn from volumes of unstructured and unlabeled data in some parts of the world while compensating for regions where there is little data available. This recognition is essential, especially as Idemia begins to sign contracts with governments in the global south. Thus, the company must put in extra effort to acquire and secure the data to make sure its product features work as well for individuals in the global south as they do for the rest of the world.
Global data still skewed towards users in the global north, where the bulk of it originates. (The volume of user data on the internet is often a function of how many users are online.) While this data improves the predictive power of the security & facial recognition models, there are concerns about the in-built biases.
Since the Idemia platform is used so much in immigration functions, false positives, and false negatives could limit the ability of travelers to reach their destinations. In the case of financial transactions, false positives could result in fraud, while false negatives would breed inconvenience for customers. There are also concerns about racial bias where technologies similar to this are applied for criminal identification. Part of the problem is that AIs perform better at identifying faces of white, male individuals and worse at identifying the faces of people of color and particularly black women. The reason for this discrepancy includes the training data used on these AI, as well as the optimization of camera settings to favor lighter skin.
In addition to the regular challenges that face any AI-supported multinational company, there are some challenges that Idemia faces that are a result of its unique business model. Do individuals have the right to opt into giving data to Idemia as the company’s technology becomes ubiquitous? Will future citizens and consumers have the option to protect their privacy away for Idemia? In the case of Kenya, registering biometric data through the Huduma Number campaign was mandated by the government and had to be challenged in court. The catch 22 is that to improve predictive power, you need data- and you can get that via mass registration- but should companies be allowed to interfere with citizen’s freedom to get data for their models?
AI and data sciences challenges
Our world in data https://ourworldindata.org/internet