Tags: Technology
In our latest guest blog, Samuel Rowe of Yoti explores the challenge of embedding trustworthiness by focussing on Yoti's age estimation product.
Yoti is a digital identity company. Across our suite of services, the commonality is that we let individuals prove who they are to other individuals and organisations. At the core of Yoti’s mission is trust; we want to become the world’s most trusted digital identity company. But how do we engender trust? One of the key methods is a focus on responsible business practices, as well as embedding trustworthiness throughout our business.
However, embedding ethical business practices is easier said than done. That’s especially true when one takes into account the importance of ensuring the entire product lifecycle is subject to the same oversight. As they say, it takes a village. This blog will explore the challenge of embedding trustworthiness by focussing on one of Yoti’s products: Yoti age estimation.
What is age estimation?
Age estimation uses a neural network to provide a predicted age based on a still or moving image. It has been trained using machine learning. In layperson’s terms, it accurately estimates a person’s age by looking at their face. We’ve conducted around 500 million age estimations since February 2019. We have published a white paper on age estimation in order to be as transparent as possible concerning how the technology was developed, how it works across different demographics, and what mitigating measures we have put in place to protect users from negative outcomes. The white paper is updated regularly to reflect the increasing accuracy of the technology, amongst other things.
When Yoti age estimation returns a result, a margin of error is provided in addition to an estimated age. This means that the system is configurable to set whatever threshold a business or regulator requires, for example requiring those over 18 to be estimated as at least 21 or 23 – a buffer of three to five years.
It is a privacy-friendly approach to age verification, which doesn’t require any personal details or documents. It does not involve facial recognition or enable surveillance more widely. Further, all information is deleted instantly once someone receives their estimated age – nothing is ever viewed by a human. These are some of the conscious choices that have been made by Yoti in order to create a trustworthy technology. However, these decisions have not been made in a vacuum. There are several measures that have influenced and continue to influence, how Yoti’s age estimation is developed and deployed in a trustworthy way.
Yoti’s Ethical Framework
From Yoti’s inception, there have been a set of core principles that form the foundation of the company’s approach to ethics and trustworthiness. The principles can be used as a yardstick by both internal and external stakeholders when evaluating our activities, as well as a steer when making decisions. Over time, those core principles have come to sit as the foundation for our ethical framework. In addition to Yoti’s principles, the ethical framework has three further components.
Similar in function to Yoti’s principles, our external pledges help us determine the correct course of action, as well as providing criteria to help with the decision making process. They also signal publicly our commitment to certain values. For example, before it was subject to a sunset clause, Yoti was one of the first signatories to the Safe Face Pledge, an initiative focused on mitigating the harmful effects of biometric technologies. It helped greatly when developing our approach to age estimation, ensuring that any application of the technology would be in line with the expectations of the Pledge. Similarly, we have made use of the Biometrics Institute’s Ethical Principles for Biometrics to help steer the development and deployment of age estimation.
At Yoti, we are fortunate to have conscientious employees, who are unafraid to voice their views on how Yoti conducts itself. To that end, I founded and now chair an internal Ethics and Trust Committee, made up of employees from across the breadth of Yoti’s staff. The Committee’s role is to make sure that we stick to our core principles in day-to-day work. Anyone in the company can suggest a topic for discussion, and the findings, as well as rationale, of the Committee, is given to the senior management team to help assist their decision making.
The above three measures are all very well and good, but could give rise to accusations of marking our own homework. Fortunately, the final component of our ethical framework is the Guardian Council, an independent ethics board made up of influential individuals from relevant fields such as human rights, data privacy and anti-online abuse. The Guardian Council meets every quarter to help us to navigate the complex world of identity and ensure we always do what’s right for our community. Their Role Description and Code of Conduct can be found online, as well as the minutes from their meetings. The Council has considered age estimation, and biometric technologies more broadly, several times.
Stakeholder Engagement
As well as our Guardian Council, we engage regularly with external stakeholders on a formal and informal basis, in order to open up our approach to age estimation, and technology more broadly. There are two primary channels through which that scrutiny takes place.
First, we host roundtables, bringing together diverse stakeholders to hear their views on our approach to developing and deploying age estimation. So far, we’ve held three roundtable discussions, focusing on different aspects of the age estimation lifecycle. We find that roundtables are particularly useful because they give stakeholders an opportunity to provide their personal or institutional advice, as well as creating an opportunity for the synthesisation of novel views as a result of stakeholder interaction.
Secondly, we commission formal reviews of our approach to technology. For example, we have had a review undertaken by Dr Allison Gardner. She works on the IEEE P7000 Global Initiative on the Ethics of Autonomous and Intelligent Systems and specifically P7003 on algorithmic bias, providing a framework for Algorithmic Impact Assessments. This review looked at our comprehensive approach to bias mitigation in the context of age estimation.
Transparency
Throughout the above, runs an adherence to the principle of transparency. The age estimation white paper, the Guardian Council’s publicly available minutes, pledges and roundtables have all been to further the transparency of our approach. Similarly, how age estimation is implemented aims to ensure the user is as fully aware of how the system works as possible, in order to make an informed decision as to whether to use it or not. We believe transparency is crucial to garnering the trust of individuals and organisations that would like to make use of our technology.
Conclusion
Hopefully, from the above, it’s clear that we take ethics and trustworthiness very seriously. I am often asked whether I think taking such a conscientious approach is an obstacle to growth for Yoti. I think that the opposite holds true: embedding responsible business practices at the heart of what we do means that we are less likely to have to revise our approach in the future, as well as lessening the regulatory risks we face in developing and implementing our technologies. In addition, in an age where social responsibility is becoming increasingly desirable, it sets us apart. Developing age estimation is an ongoing process of learning. However, we like to think that we are on the right trajectory.
Author
Samuel Rowe
Legal and Policy Associate, Yoti
Samuel Rowe is a Legal and Policy Associate at Yoti. He chairs the techUK Digital Identity Working Group and Yoti’s internal Ethics & Trust Committee. He is also one of the independent reviewers of the Independent Review of the Governance of Biometric Data, commissioned by the Ada Lovelace Institute.