NZ has led the world in issuing its Algorithmic Charter for Aotearoa New Zealand in July this year. The charter is designed to ensure NZ citizens have confidence in how government agencies use algorithms and demonstrating a commitment to transparency and accountability in using data.
Algorithmic accountability has been much on my mind for a while now. I first wrote about this as an emerging issue for recordkeeping in 2016. Since that time, algorithms have been identified as decision making tools in many contexts and are likely to increase in use by organisations as the trends towards datafication of all types of things, not least of which is people.
Many books have been written identifying the inherent bias built into assumptions controlling algorithms; one of the best known of these is Professor Sofiya Noble’s Algorithms of Oppression, but many other scholarly articles have emerged around the use of algorithms. Similarly there has been an expansion of attention to data ethics, following the Snowden revelations about widespread unauthorised data collection by multiple governments and the Cambridge Analytica Facebook scandals of 2015-2016. Some of the world’s biggest data players are even backing calls for ethical approaches to data use and regulation of AI.
Examples of misintended (we assume) results which severely impact people, are growing. They include the Google facial recognition service that labelled images based on skin colour and the earlier labelling of dark skinned individuals as ‘gorillas’; or more locally the use of AI and predictive technologies in our own home grown Robo-debt scandal linked to the suicide of a number of targeted individuals; and most recently the controversy surrounding the algorithmic grading of UK’s school leaver results in CoVID times. The emergence of the Indigenous Data Sovereignty initiatives reflect just how seriously Indigenous communities regard the appropriation and perpetuation of stereotyping in data use. Of course there are great benefits, new insights and advantages to also be found in greater activation of data – it’s just that the problem stories show the devastating results of misapplied use of technology.
Recordkeeping and Algorithms
So what is the role of recordkeeping in the process of documenting algorithms which learn and evolve? It’s a big topic, far too big for me to be able to answer and one which needs much more work. The UK Information Commissioner’s Office is doing interesting work in building an AI Auditing Framework which will likely have recordkeeping implications. But what seems likely as a starting point is far more systematic documentation about algorithms, AI and predictive learning. So here is perhaps some beginning thoughts:
- identify and maintain the data on which the algorithms are trained;
- document and records all processes in the design of systems and algorithmic development;
- clear documentation about the intent of the algorithm, and
- development of a regular, systematic and documented auditing approach to test whether the algorithm is still doing what it set out to do.
All this also brings with it questions of individual consent (and the use of umbrella consent models), social expectations and larger social conversations about consultation, about whether the outcomes are ethical etc.
It’s a big topic – I’m not going to be the one to ‘solve’ it. There is so much to know and find out. But I am actively thinking in this space, and would love to hear from like minded recordkeeping people.
About the Author
Ms Barbara Reed, Director of Recordkeeping Innovation Pty Ltd, has been a consultant in the fields of records, archives and information management since 1985. She is active in professional arenas, including the teaching and training environments. She has played a major role in the development of Australian and International standards for records management, digitisation, recordkeeping metadata and others.