Let's face it, facial recognition will make lives much easier
The technology is gaining popularity but care must be taken to use it ethically
By Shalini Verma
Published: Mon 14 Jan 2019, 5:00 PM
Last updated: Mon 14 Jan 2019, 7:39 PM
If there was such a thing as an age of enlightenment for machines, this is such a time. Scientists are at work to fulfil humanity's quest for Artificial Intelligence (AI). Our need for attaining digital self-actualisation has developed an entire industry to train machines to see, think and reason like humans.
Arguably, one of the most advanced AI technologies is facial recognition. It could be as mundane as Facebook automatically tagging you in a college reunion photo, or as critical as surveillance cameras identifying an absconding criminal. Not so long ago, recognising faces was the domain of law enforcement agencies, wherein experts poured over books of mugshots to find a match. Today, all it takes is a fraction of a second for a well-trained computer algorithm to detect a face in an image, mathematically extract a distinct feature set, and match it with verified data from its database.
The earliest attempt at computer-based facial recognition was made in 1960s when Woodrow Wilson Bledsoe formulated a system for measuring key facial landmarkvs on each picture. Later, scientists devised a method of plotting the face just as a map. The key inflection point for facial recognition occurred when the Internet bred massive databases of images thanks to photo sharing apps. But machines still had to make sense of these pictures. So, online crowdsourcing services made it possible to recruit people who charged a few pennies to label images, much like nursery picture books. This gave birth to databases like Labeled Faces in the Wild.
Algorithms advanced into artificial neural networks that uses a network layered approach to mimic the human brain. This together with four million pictures of a user helped Facebook build a more accurate deep leaning facial recognition system. Facebook had little difficulty in finding users with as many pictures!
The algorithm works on the basis of a match confidence level or the probability that the match is correct. When the confidence threshold is set low, the algorithm gives you a higher number of matches, but with a higher possibility of errors. With improved technologies that can pick up on skin texture, the algorithm can even tell apart identical twins.
After we made giant strides with recognition of mugshots, the next frontier to conquer was facial recognition in a crowd. The Chinese police department can now find a wanted fugitive in a crowd of 60,000. Its platform has more than 1,000 network layers that can train up to 2 billion facial images simultaneously. Its facial recognition technology can detect up to 240 major facial characteristics in a millisecond, and can recognise over 10 facial attributes, such as gender, age, expression, and facial hair.
There is an unsettling feeling about facial recognition because of concerns for mistaken identities and invasion of privacy. However, these are not new worries. The public concern for mistaken identities goes back to Victorian times when prison photography was introduced to track down criminals. The subsequent rise of portable cameras in 1900s also raised privacy concerns leading to several lawsuits against photographers.
Additionally, facial recognition programmes can reflect our conscious and unconscious biases. For example, AI is less likely to identify a dark-skinned female than a white-skinned male, which is ironical because we want to enlist the help of machines expected to be more neutral than humans. Nevertheless, scientists are working on models to remove these human biases from machines.
The demand for an ethical use of AI technologies has been gathering steam. In the UAE, Smart Dubai has released an Ethical AI Toolkit in an effort to put some guidelines in place for organisations to evaluate the ethical level of their AI solution. As the industry matures, we will see government agencies establish clear regulations to reign in the misuse of technologies like facial recognition.
It certainly opens up widespread opportunities to improve our quality of life such as faster access to a large event such as Expo 2020, or to a cinema hall. Facial recognition will save time for everyone involved at a department store checkout counter. We can have safer Uber rides as the algorithm protects us from unverified drivers. Apple Face ID lets us unlock our phones and make payments. It lets us create 3D digital images of our faces, using infrared dots detected by a camera that are measured against a stored image. It's hard for me to type out my password, after getting used to Face ID.
We will eventually look beyond the face to adopt more sophisticated techniques for machines to 'see' objects in a crowd. For example, use of movements in a large crowd. The human brain's cortex perceives the visual landscape as a map divided into grids. Cell groups overlap grids making it easy to recognise an object when it moves from one grid to another.
For now, our face will continue to be a major source of convenience. We may not all possess Helen's face (of Troy fame) that launched a thousand ships, but our face is quickly becoming our biggest asset as we go about performing our daily tasks. Taking people at face value isn't such a bad thing after all.
Shalini Verma is the CEO of PIVOT technologies