A UK tech startup has developed an artificial intelligence (AI), tool that can determine if users are too young for apps like Instagram or TikTok.
Yoti’s ‘Age Estimation’ system — which may well soon be rolled out across social media — can tell how old users between 6–18 are to a 1.5-year margin of error.
The software compares the user’s facial features, as captured via their device camera, against millions of other images from Yoti Digital ID app users of known ages.
Social media firms such as Facebook have long struggled with how to handle minimum age verification without requests for passport details, which many see as intrusive.
Scroll down to see the video
A UK tech start-up has developed an artificial intelligence tool that can tell whether users of social media are too young to use apps such as Instagram or TikTok. Pictured: A child uses Yoti’s “Age Estimation” tool to verify his ages
Yoti’s ‘Age Estimation’ system — which may well soon be rolled out across social media — can tell how old users between 6–18 are to a 1.5-year margin of error (right). The software compares the user’s facial features captured by their device camera (left) with millions of images from Yoti digital ID apps users of known age (right).
‘The threats that children face online sadly continue to grow, so I’m proud to introduce Yoti’s Age Estimation for under 13s,’ said Yoti CEO Robin Tombs.
“This technology will allow businesses and regulators to better protect young people with low friction, while still protecting their privacy.
‘We’ve now made it easy for platforms around the world to design services “age appropriately”.
‘Yoti’s facial Age Estimation will help many different businesses comply with changing age regulation such as the new Children’s Codes.’
Businesses using the software — which previously only worked on adults — can set an age threshold for the AI to compare each user to.
In this way, the system can be used to see if children meet the legal threshold of 13 to join apps like Facebook and Twitter — or to determine whether an individual has, say, passed the drinking age.
In fact, the system is already being employed in supermarkets in Estonia for age verification at automated checkouts, and by the German version of the adult entertainment platform Fan Centro — and has already made more than 550 million age checks.
Yoti engineers used images taken by millions of users who had downloaded the firm’s digital ID application to train the AI to recognize adult faces.
This is a stand-alone method of proving age and identity. It was launched in 2014 using a combination ID documents and facial recognition.
For training the AI in age estimation of younger people, photographs of children were used — with parental consent — as part of a programme organised by the Information Commissioner’s Office, a UK data watchdog.
Yoti, a London-based company claims they have improved the accuracy with which the system estimates the ages of younger individuals over the past three years.
In 2018, the AI was accurate to within a margin of 1.5 years for those aged 13–24 and to within a year for those aged 16–17 — but the firm now reports an accuracy of within 1.3 years for users aged 6–12 and to 1.5 years for those aged 13–18.
For users on the cusp of the legal limit to access social media sites — whose ages might conceivably be incorrectly assessed — platforms could then request further forms of verification before granting access.
MailOnline was told by the company that the AI has been built with ‘proprietary Anti-spoofing Technology and Passive Liveness Detection’ to ensure that an authentic, live face is being presented for analysis.
In this way, the company explained, the system should not be able to be tricked by underage users presenting, say, a photograph or video of an older individual.
Yoti engineers used images taken by millions of users who downloaded its digital ID app to train the AI on adult faces. For training the AI in age estimation of younger people, photographs of children were used — with parental consent — as part of a programme organised by the Information Commissioner’s Office, a UK data watchdog. Pictured: An artist’s impression showing a facial recognition program and its analysis.
‘The threats that children face online sadly continue to grow, so I’m proud to introduce Yoti’s Age Estimation for under 13s,’ said Yoti’s Robin Tombs. ‘This technology will help regulators and businesses better protect young people with low friction and privacy.
The Children’s Commissioner for England conducted research and found that 60% of eight-year olds and 90% of twelve-year-olds use private messaging apps, despite them being restricted to 13-year-olds.
Similarly, a 2018 investigation by the market intelligence firm Kids Insights found that nine-in-ten 12-years-old are illegally on social media sites likes Facebook, Instagram and Twitter.
This is why MPs proposed legislation in the Draft Online Safety Bill that would require social media companies to take’reasonable measures’ to prevent children from accessing inappropriate content.
HOW ARTIFICIALINTELLIGENCES LEARN Using NEURAL NETWORKS
Artificial neural networks (ANNs), used in AI systems, attempt to mimic the brain’s workings in order for them to learn.
ANNs can be trained in order to recognize patterns in information, including speech and text data. They are the foundation for many of the recent developments in AI.
Conventional AI uses inputs to ‘teach” an algorithm about a specific subject by feeding it large amounts of information.
Artificial neural networks (ANNs), used in AI systems, attempt to replicate the brain’s learning process. ANNs can learn to recognize patterns in information, including speech and text data.
Practical applications include Google’s language translation services, Facebook’s facial recognition software, Snapchat’s image altering live filter and Snapchat’s image recognition software.
This process can take a lot of time and only one type of knowledge is allowed.
Adversarial Neural networks is a new breed ANN that pits two AI bots against each another, allowing them to learn from each others.
This approach is intended to accelerate the learning process and refine the output of AI systems.