Ensuring ethical outcomes from AI services

Artificial Intelligence – the creation of machines and software that can learn, reason, make decisions and act on their own – offers huge potential for governments to build more responsive services, policies and digital solutions.

However, the Policy Lab has identified a number of challenges when it comes to implementing and using AI. The first involves what AI can be used for with many agencies driven to use AI for cost savings, but we are encouraging agencies and service providers to also take advantage of non-economic outcomes when creating AI.

The second issue revolves around the quality of the AI algorithms, with poorly-trained AI creating sub-par outcomes. For example, a machine learning algorithm trained on existing, biased data can reinforce inequality. We need to make sure that government AI is designed well and doesn’t have unintended consequences. With the help of our AI technical experts, the Gradient Institute, we have built an ‘AI exemplar’ to demonstrate how such transparency and accountability can be achieved without compromising the underlying algorithm. This work has shown that the high standards and implementation is possible.

What’s happening in the world of AI?

We know agencies want to leverage AI for its many advantages, including speed and cost savings, but are concerned about getting it right. Many are waiting for guidance and a whole-of-government approach. The handful of agencies testing and using AI are facing increasingly complex ethical questions.

That’s where we come in. The Policy Lab is working with agencies and experts to develop a whole-of-government AI ethics policy framework to support agencies to use and test AI in full confidence that they are achieving ethical outcomes.

If government wants to leverage the advantages of AI to benefit people in NSW, we need to:

  • enable agencies to be able to consider and balance the ethical trade-offs
  • articulate what fair and ethical outcomes look like
  • identify minimum standards
  • provide guidance and tools to enable NSW government agencies to build, buy and partner with others to implement AI solutions safely, in a way that delivers ethical outcomes.

Imagining a positive AI-enabled future

Earlier this year we brought together AI and machine learning experts from industry, academia and the non-profit sector, as well as professional ethicists and current and potential users of AI from across the NSW Government. We asked participants to imagine an ‘optimistic future’ to help us understand what NSW agencies hope to achieve with AI in the long term. We also asked workshop participants to articulate their hopes and fears for an AI policy framework to help scope what kind of guidance the NSW Government needs to provide.

What agencies want What we will we be doing
Human-centred outcomes – directly involve citizens in the ongoing policymaking process Building tools to help agencies define their policy intent and desired outcomes on top of achieving internal efficiencies and savings.

These tools will inform agencies if AI is the right technological solution, identify useful metrics, perform data cleansing, and conduct trials and assess risks.

Enable data-based decisions and equitable representation Tools and guidance to address legal and human rights obligations, different citizen needs and to de-bias training data.
Build capability Advice to beginners on how to assess service providers during procurement, and for intermediate and advanced users on AI process design guidelines, including change management to avoid hidden costs and risks.
Enable safe experimentation Guidelines on key activities in the AI implementation process, including testing and modelling, engagement and public relations.
Advice on key roles and skillsets to build, including data architects, privacy specialists and ethicists.
Write digital policies flexibly to stay relevant Our framework will be tested with users of different levels of understanding and updated regularly to ensure relevance and practicality.

Providing practical guidance to enable ethical outcomes

Since the workshop, we’ve been working with ethical AI experts at the Gradient Institute and other technical and legal experts to develop a prototype policy framework. To support agencies to use AI in ways that will achieve ethical outcomes the framework will include:

  • guidance on the legal and international human rights obligations that will apply to the implementation of AI solutions in NSW
  • principles that shape outcomes in a variety of different AI technologies, but are also specific enough to effectively guide development of AI solutions
  • hypotheticals and use cases to model best practice
  • a suite of practical tools and guidance aimed at agencies and users with different levels of understanding and experience.

Next steps

The Policy Lab is looking to complete the prototype framework in the first quarter of the new financial year. We want to engage more broadly with experts and the community, test the prototype and co-design a further iteration with partner agencies in the NSW government. We’re also building an interactive public engagement plan to leverage invaluable insights.

Source: digital.nsw.gov.au