Accountability and regulation of algorithms – Flash 02 2017
The increasing prevalence of personalization of online services to optimize search results, news and product recommendations and even political messaging has led to growing concerns regarding responsible implementation and management of these algorithmic systems. Join us to explore how best to govern the information access mediating systems in the context of an internationally connected world.
algorithm, algorithmic bias, algorithm transparency, algorithmic accountability, personalization, recommender systems, standards, regulation, trust
Algorithmic decision processes that rely on high-dimensional user profiles, derived from personal data and possibly involving machine learning methods, pose difficult challenges for providing meaningful transparency and guarantees of reliably fair performance.
In recognition of these issues, various professional bodies, academics and regulatory organizations have started investigating this space, looking at causes/remedies against unjustified bias/disparate impact, means of establishing accountability, safeguarding privacy and exploring means for providing meaningful transparency/explanation of automated decision making.
Proposed methods for addressing these issues include industry self-regulation, establishing of Standards, voluntary certification and government oversight (e.g. “an FDA for algorithms”). A recurring theme in many of these proposals is the need for trusted third-party experts to be involved in the evaluation process as means of establishing trust and protecting intellectual property.
Given the cross-borded nature of services provided via the Internet this raises questions about jurisdiction. To maintain a free, open and trusted Internet where services are not restricted by nation or region, certification of algorithmic internet services will need international multi-stakeholder coordination.
Some of the questions we will explore include:
- Can this be solved by self-certification based on international industry standards (e.g. IEEE Standards P7001 "Transparency of Autonomous Systems", P7002 "Personal Data Privacy", P7003 "Algorithm Bias Considerations")
- Are efforts by industry groups going to be enough to satisfy and retain the trust of peoples and governments (e.g. Partnership on AI)?
- Is there a place/need to establish an international expert oversight body to provide trusted-third party certification?
We will use a 'round-table' format.
I will start off the session with a 5 minutes talk outlining our current work on developing a Standard for Algorithm Bias Consideration and the editorial responsibility arising from the use algorithmic systems to mediate content accessibility by online services.
The discussion will then be opened to all participants with the following structure.
- Round 1 - comments regarding existing regulation, what works and what doesn't?
- Round 2 - what additional (self-)regulation is required and how can this be achieved, with special focus on the international cross-border dimension of algorithmic systems used in internet services.
- UnBias: Emancipating Users Against Algorithmic Biases for a Trusted Digital Economy
- Report of UnBias mulit-stakeholder discussion on algorithm fairness
- “Facebook’s algorithms give it more editorial responsibility – not less“, theConversation, 14 September 2016
- IEEE P7003 Standard for Algorithm Bias Considerations working group
Ansgar Koene, Horizon Digital Economy Research institute, University of Nottingham, UK