Best practices for transparency
in machine-generated personalization

Laura Schelenz, Avi Segal, Kobi Gal

The Problem:

Personalization is widely used in AI-based systems to improve users’ experience.
However, users are becoming increasingly vulnerable to online manipulation due to opaque personalization.
Such manipulation decreases users’ levels of trust, autonomy, and satisfaction.
Increasing transparency is an important goal for personalization based systems.
Unfortunately, system designers lack guidance on how to increase transparency in their systems.

Transparency Definition:

(Based on a review of ethics literature)

A practice of system design that centers on the disclosure of information to users, whereas this information should be understandable to the respective user and provide insights about the system. Specifically, the information disclosed should enable the user to understand why and how the system may produce or why and how it has produced a certain outcome (e.g. why a user received a certain personalized recommendation).

Best Practices and Checklist for Transparency:

System designers can consult the best practices below to evaluate transparency in their system.
To simplify their work, we have developed this checklist (23 questions) which they can fill.
They can then make adjustments to their system based on their checklist replies.
We help system designers understand where – in the system architecture – they are strong on transparency and where improvements are needed.

Architecture component	Best practices
Input	Disclose information about data collection Notify the user about the data collected for personalization Disclose information about pre-processing and possible biases in the dataset
Processing	Disclose information about the goals of the system Disclose information about the reasoning of a system (factors used and how they are weighted) Disclose information about how inferences influencing personalization were made If possible, provide expert testimony (by the designer) about how the system works If possible, disclose how the models affect users and interact with other models
Output	Disclose that a machine, not a real person, is communicating with the user Disclose information about those responsible for the models
User Control	Offer different options for personalization that users can choose from Allow users to provide feedback about their goals; Adjust personalization according to users’ self-reported preferences

User control: Enabling users to interact with the system to adjust elements based on their respective needs and preferences.

Case Study:

Applying the checklist to several online services as a reflective and assessment tool.
Taking a system designer’s point of view, using authors’ accounts in these systems.
Checked information and options available to registered users.
Y axis: the percentage of positive and partial replies in each checklist section.

Preliminary checklist, online sites

Conclusion and Future Research:

Take Aways:

Developed a Transparency Definition, Best Practices and Checklist for system designers.
Applied checklist to several online services.
Current systems lack transparency especially with regard to “Processing” and “User Control”.

Limits of Transparency:

No guarantee that designers and users really understand models: lack of resources, human capital, digital and technical literacy.
Information can obscure details, confusion of users.
Transparency vs. privacy of users/designers/companies.
Balancing transparency and economic interests.

Need for Personalized Transparency:

What is transparent to some, may not be transparent to others.
Configure information to user needs and capabilities.
How does other users’ behavior affect a particular user?

Poster 235

Best practices for transparency
in machine-generated personalization

The Problem:

Transparency Definition:

(Based on a review of ethics literature)

Best Practices and Checklist for Transparency:

Case Study:

Conclusion and Future Research:

Links:

Best practices for transparency in machine-generated personalization

The Problem:

Transparency Definition:

(Based on a review of ethics literature)

Best Practices and Checklist for Transparency:

Case Study:

Conclusion and Future Research:

Links:

Best practices for transparency
in machine-generated personalization