Technology and more recently AI (Artificial Intelligence) have underpinned much innovation all around us. To ensure user acceptance and success, data is collected to determine aspects like feasibility and fitness of a technology. However data, while powerful, can also be toxic if it is poorly handled. In particular, to address privacy and protecting users, a good cultural practice is to ensure that both the data and collection method adheres to the principles of ethical standards.
Interestingly when ethics is a pro-active cultural consideration, technology development teams face a chicken and egg problem. So what is the problem, why is it challenging and how can we address it? This blog covers these elements in two parts. The first part explains how the desire to be good presents a set of challenging problems while the second part presents how to overcome these concerns.
The primary objectives of any technology are:
- efficacy – to assist users, and
- effectiveness – acceptable within an economic sense (cost effective, ease of use, scalable etc.)
To confidently determine efficacy and effectiveness, we need to collect data to support and demonstrate the outcome. However to ensure that we do this ethically, we need to decide on the type of data to collect (during the design phase) and get consent from users to provide this data. This leads us to the chicken-egg problem. To be good, we want to collect just enough data that is collected with consent. However to know what type of data to collect, we need to build an initial version of the product (Figure 1). This problem also arises in pharmaceutical development as well. How can we develop medicines without trialling first?
In technology development, the extreme approach (ie, not ethical):
- make assumptions about what the users need (or) start with weak assumptions
- hide the data collection from the users (while we experiment on them)
- subject the user to some features (we do not know what they need)
- use the hidden data collection (to determine what they need)
- iterate till the solution is appropriate.
At the other end (ie, following a good ethical process):
- define data collection protocol & seek an independent review from an ethics committee
- clearly tell users about the data being collected and the purpose
- build multiple iterations of the product (till we get it right – starting with a best guess).
In both cases, multiple iterations are unavoidable. Hence it is better to consider data ethically. But, this can be done well and with minimal impact to the overall project – which I will discuss in Part 2.
* Refer to attachments below – Checklist for ethics application processes & HREA Project Description Template for greater than low risk ethics applications.
Associate Research Fellow, ARC Industrial Transformation Research Hub for Digital Enhanced Living
Applied Artificial Intelligence Institute, Deakin University
NB: The author reserves the right to showcase/publish this blog piece elsewhere and/or in a different medium.
Editorial review by:
Professor Rajesh Vasa, Chief Investigator
Kevin Hoon, Hub Manager