Alpha Health’s Delivery Plan for its Ethics Strategy

Posted / 22nd June 2020

Introduction

This paper provides a summary of how Alpha Health plans to deliver its Ethical Principles and Commitments. It also includes a brief summary of lessons learnt to date.

 

Delivering our ethics strategy

This paper is not intended to be a detailed delivery plan, but we nevertheless want to include a high-level description of how our ethics strategy is being delivered. We will describe progress against our plans in full in an annual external audit of our ethics strategy.

 

Our ethical principles and commitments are designed to apply to all products and teams within Alpha Health. Therefore we have organised their delivery into four work streams that best enable them to be embedded into the work of our teams.

 

Workstream 1: Ethics in Design

This workstream covers the following commitments: 2 – bias and inequality; 3 – putting the user in control; and 5 – explaining how our services work.

 

The aim of this workstream is to really embed ethics into product design, and so it is split into five parts:

  1. Creating simple tools that the team can use in their day-to-day work
  2. Writing a guidance manual to support how to use the tools described above
  3. Appointing ‘champions’ who are responsible for encouraging the use of the tools, in particularly identifying the right time to use them, and how
  4. Training on how to use the tools and manual
  5. Evaluation – a simple quarterly survey to understand how easy design and product teams find it to embed our ethics strategy into their work, how useful the tools are, and what could be improved

 

In terms of progress: design templates for consent have been created; tools for plain and non-biased language are in development; and drafts of plain language privacy policy and terms and conditions are being discussed with our in-house legal team.

 

In addition, it is worth noting that all of our design decisions are guided by clinical and scientific evidence for what will improve the efficacy and engagement of our products; content is created by a combination of our in-house experts and our collaborators such as Massachusetts General Hospital and the London School of Economics and Political Science. This grounding in the evidence base contributes to our first commitment to support health and happiness.

 

Workstream 2: Ethics in Technology

This workstream covers the following commitments: 2 – bias and inequality; 5 – explaining how our services work; and 8 – creating privacy preserving products.

 

In a similar vein to workstream 1, the aim of this workstream is to really embed ethics into the work of our R&D team, again split into five parts:

    1. Creating simple tools that the team can use in their day-to-day work
    2. Writing a detailed guidance manual to support how to use the tools described above
    3. Appointing ‘champions’ who are responsible for encouraging the use of the tools, in particularly identifying the right time to use them, and how
    4. Training on how to use the tools and manual
    5. Evaluation – a simple quarterly survey to understand how easy the R&D team finds it to embed our ethics strategy into its work, how useful the tools are, and what could be improved

 

In addition to embedding ethics into the working practices of the R&D team, the ethics strategy is also driving two important areas of technology development. The first of these is the creation of Explainable Artificial Intelligence (XAI), in which the team is aiming to provide both local and global explanations for any type of automated decision-making system, machine learning models. The second area of technology development is Privacy Preserving Artificial Intelligence (PPAI), in which the team is aiming to be able to provide automated decision-making services without Alpha Health ever being able to access our users’ personal data, so that no personal data need ever be sent to Alpha Health.

 

In terms of progress: an XAI proof of concept has been created for global explanations of some forms of neural network; work is underway to create a PPAI proof of concept; and a first version of a manual for avoiding algorithmic bias (including training) is almost ready.

 

Workstream 3: Ethics R&D

This workstream covers the following commitments: 1 – supporting optimal health and happiness; and 4 – avoiding addiction.

 

In supporting optimal health and happiness, the majority of our work is focused on being able to measure and improve both health, happiness, and control. This is at the core of Alpha Health’s work, both in terms of the creation of products that meet different health needs, and also within R&D as we seek to create new health interventions and improve the effectiveness of our existing ones. 

 

In terms of our ethics strategy, the additional work required is to understand how people make trade-offs between health, happiness, and control. Whilst these trade-offs will ultimately differ across individuals, we would like to see if there is enough similarity across groups to enable us to inform the recommendations that our products provide to users. To this end we are working with the London School of Economics and Political Science (LSE) to run a study of how representative populations of people in Spain, the UK, and the USA trade-off health, happiness, and control.

 

Turning to avoiding addiction, as smart phones have proliferated there is increasing concern that apps can be designed in such a way as to be addictive, and therefore detrimental to a user’s health. Clearly, for Alpha Health, pushing user engagement as far as addiction would be against our health mission. Therefore, this line of work will examine how we can identify addiction in the use of applications, and put in place methods to manage or mitigate it. This work is at an early stage, and we plan to begin by reviewing what is already known about apps and addiction, for instance drawing on lessons from online gambling.

 

Workstream 4: Ethics Communications and Accountability

This workstream covers the following commitments: 6 – publishing our work; and 10 – external audit.

 

The aim of this workstream is as much to ensure that the team has the skills to support our ethics strategy overall, as well as putting in place processes to ensure accountability. It is split into four parts:

    1. Establishing the process for audits. This includes running an internal audit to provide a baseline for an external audit, and appointing an external auditor
    2. Introducing a process of ethical approvals for research that we run internally, and to ensure that where we undertake research with external partners their ethics approvals processes is used
    3. Introducing a process to ensure that we are regularly and consistently publishing our audits, ethics approvals, and other documents related to our research
    4. Building team skills by creating a training course that each team member must attend

Alpha Health has already commissioned one external audit of our work, and will commission another audit during 2020. We have also begun to publish more of our work on our website, and this will grow over the year. Finally, we ran our first general ethics training session in March this year and are reviewing feedback with the aim of improving the course.

 

In terms of our research ethics, all of our existing collaborations are subject to IRB / ERB approval and adhere to the Helsinki Declaration. We are currently exploring with our collaborators how best to publish such approvals.

 

Ethics in wider Alpha Health work

The above four work streams don’t cover all of the ten commitments because some of them are fully addressed in existing Alpha Health work. This doesn’t make them any less a part of our ethics strategy, nor does it mean that they will not be audited, it just means that they don’t require separate workstreams to be established to ensure their delivery. See annex, below, for more details on this.

 

Lessons learnt

Alpha Health has been developing its ethics strategy since 2016 when it was established. During that time we have been developing and testing a range of ideas and we have already experienced a number of challenges. Three are highlighted in this section.

 

Firstly, principles can conflict. We saw this most acutely when undertaking an algorithmic audit of one of our prototypes during 2018/19. You can read the full story in the paper published here as part of the proceedings of AI Ethics and Society 2020. In summary, in creating a prototype app, we had put such a strong focus on privacy that we had no direct evidence to draw on later when we wanted to understand whether the recommendation algorithm within the app might be biased against any particular group. Fortunately, Eticas and University Pompeu Fabra, who ran the audit, were able to use indirect evidence to demonstrate that the app was not biased. However, it demonstrated to us that early consideration of the how principles may come into conflict is required for each of our products.

 

Secondly, to an extent, the delivery of an ethics strategy is all or nothing. What we mean by this is that it’s not really possible to say that we’re 60% or 80% ethical; in setting out five ethical principles we have to be addressing all five at the same time, a waterfall approach is not acceptable. This is most obviously true of the approach we have taken to understandability and transparency, where initially we put more emphasis on the former than the latter. Publishing this strategy is an example of us now redressing that balance.

 

Lastly, and related to the second lesson, it is challenging to find the right balance between ambition and realism. Whilst we want to be ambitious we also recognise that there is no point setting commitments that we cannot meet at all right now without risking appearing to be insincere. Equally, if we only focused on what can be done now with certainty, then we run the risk of appearing not to be serious in being ethical, and instead only doing the minimum required. To try to strike the right balance between these two poles we have proposed a series of commitments, sitting underneath our principles. These are a mix of ideas that we can already implement, alongside ones that are more challenging and which require real research and effort to deliver. 

 

Further reading

You can read our Ethical Principles and Commitments here

Plus you can also read about Alpha Health’s Approach to Ethics, and Alpha Health’s Governance Framework for its Ethics Strategy.

 

Appendix

Mapping of our ethical commitments to workstreams

 

Principles Commitments Delivery workstreams
Improving your health and happiness We aim to support you to have the greatest possible health and happiness, and we will never try to increase one if it will significantly reduce the other Ethics R&D – to understand how people trade off health and happiness
We will ensure that our recommendations are not based on gender, race, ethnicity, sexuality, age, belief, or other characteristic protected under anti-discrimination law, unless there is evidence that demonstrates that such a characteristic is an important driver of the desired outcome of the recommendation Embed in Design – to avoid bias in our visual and written language

Embed in Technology – to avoid biased algorithms

Putting you in control You will always know and control how we use your personal data Embed in Design – providing templates for data consent
We will deploy the best available techniques to prevent any user from becoming addicted to any of our services Ethics R&D – to create methods to identify and avoid user addiction in using our services
Being understandable and transparent We will explain how our services work to support you in having the greatest possible health and happiness. We will ensure that such explanations are comprehensible Embed in Design – design models for explainability, simplified language in privacy policy and terms and conditions

Embed in Technology – create and deploy XAI

We will publish the ethical approvals of our research and external audits of our work, although we may remove some commercially sensitive information Ethics Comms & Accountability – publishing our audits, and creating an internal ethical approval process for research
Securing your data We shall share only the minimum data about you on our internal networks, and when we do this we will use the highest industry standards of encryption to protect your data Delivered through Alpha Health’s Cyber Security team
We will create products and services that preserve as much privacy as possible, for you and your community Embed in Technology – create and deploy PPAI
Being accountable We will never generate revenue through advertising Delivered through Alpha Health’s overall commercial strategy
We will hold regular external audits of our algorithms, and always before commercial deployment, to test that it is doing its job and avoiding biases and other unintended consequences Ethics Comms & Accountability – set up systems for audit and ensure that the team have the skills to deliver