Biometric Longread List: What’s Happening in AI and Biometrics (Christmas 2022 Update)

A lot has happened this year in biometric and AI fields. Everything is now (or very soon will be) AI-enhanced, from text editors to painting programs. Let’s dive in into what it means for our near future with our Biometric & AI Longread List, now updated for Christmas 2022!

• For a long time, synthetic faces are a mainstay in biometrics, for both testing and improving algorithms. However, you can now create whole synthetic persons if you need a model for your campaign or an impersonator when catching online predators.
• DALL-E 2, Stable Diffusion and other text-to-image generators are all the rage this year. The computer-generated images have the potential to upend many sectors, as Wired notes. The real artists trying to sell their own original works online are not happy, though. Especially when AI is already winning painting competitions.

• Want to use AI anyway? Microsoft is already introducing the text prompt generation in its Designer, and so does Adobe. Want to use it for free? Turn to Canva, which offers unlimited AI image generation for free.
• Do you want to know how do the text-to-image generators work in more technical terms? Towards Data Science has a more in-depth look.
• ChatGPT is to text what image generators are to pictures. It’s just a few weeks old, but some already predict that it can destroy Google’s business model by threatening its ad business. Google itself is a competitor in the race for a believable AI chatbot, but doesn’t want to risk its reputation.
• Meanwhile, many tools are already implementing AI capabilities for generating text. Marketing may be one of the first impacted fields, because of the ability of ChatGPT to generate believable text to a simple prompt. And note-taking apps are now able to summarize Twitter or e-mail threads into concise notes via AI. Google can already do the same for its chats and e-mail threads.
• EU has adopted a common position to AI Act that should govern the use and deployment of AI-powered solutions in the member countries. The resistance to all things biometric has not seeped into the common position, although there has been a strong push to ban most forms of biometric identification, especially those based on facial recognition.
• Not everything can be made easier simply by introducing a new technology. The elections, for example, are difficult to digitize if you want to keep them both trusted and anonymous. Still, they can still be made faster and more convenient by introducing biometrics-based technologies, as our example from Albania shows.

Christmas 2021 Edition

• Do you want your facial recognition or computer vision algorithm fast or accurate? Google tried to offer both benefits at the same time and devised new families of neural networks. You can read about them here.
• Neural networks used in biometrics are hungry for new and faster processing, but making chips takes some ingenuity. The newest machine able to etch new chips created for AI is “roughly the size of a bus and costs $150 million. It contains 100,000 parts and 2 kilometers of cabling. Shipping the components requires 40 freight containers, three cargo planes, and 20 trucks”. Read more here.
• Sometimes when you need to use AI extensively, a good idea can be to design your own chips. At least, that’s what Tesla is doing to power computer vision in their autonomous cars.
• While biometrics can be biased, what you see more frequently is how other algorithms make communications on social networks biased and toxic. This, in turn, “poisons” language models created by scraping those networks. Google’s DeepMind lab use AI to try to make the models such as GPT-3 less toxic. The result: AI is unable to help. You can also read the whole research paper about detoxifying language models here.
• You know that the neural networks have to be trained to provide their services in biometrics and beyond. The thing is, even physical things require training. The examples are huge magnets used for particle acceleration in the Large Hadron Collider: if they are not trained, they can fail and burn out. “The coils are heated up to 665 degrees Celsius over an 11-day heat cycle; a process which transforms the ordinary niobium-tin cable into a superconductor, but also makes it incredibly brittle. It becomes as fragile as uncooked spaghetti.” Read more here.
• Do you want to see behind our curtains? Read how we have developed a passive liveness check that has passed iBeta levels 1 and 2 tests. Our expert Peter Molcany explains how we have acquired a specific dataset that was both useful and ethically sourced.
• If you still have some spare time, why not look into the future? AI legend Kai Fu Lee has written a book with Chen Quifan sketching out his vision of how the world will be changed by AI in twenty years. Simply named “AI 2041”, it  is already among the most recommended tech books of this year.

June 2021 Edition

• You might have missed a fascinating Wired story last year about Woody Bledsoe, a guy that invented facial recognition 60 years ago. Along with that, as his wiki now states, he’s rightfully considered one of the founders of AI.
• Our faces are unique and therefore great for identity verification, right? Not so fast, say these researchers (PDF) from Tel Aviv University. They devised a neural network able to produce “master faces” – faces able to pass authentication for a large portion of the population even without access to any user information. They still have to beat the liveness check, though. (An older paper had already made a similar experiment generating master fingerprints.)
• Deepfakes are fast becoming one of the ways of improving traditional social engineering attacks. Last year, this type of basic fraud increased by 820 percent. Here’s an in-depth summary (PDF, presentation is here) of the types of attacks from this year’s Black Hat conference. Biometrics can surely help with them.
• Do you want to know just how COVID-19 has upended business? Professor Scott Galloway will show you what to expect in the coming years. His book, Post Corona, is full of insights into how the pandemic has accelerated changes that will profoundly alter the way we approach business and services.
• One of the most-read stories in our inaugural Trust Report is about the Panama registry. It sets the blueprint on how biometrics can make people’s lives easier.