Algorithmic Identity
ROLE :
Creative technologist, Design researcher
TIMELINE :
8 months
TOOLS :
claude code, vercel, github, ChatGPT.
Presented at
The Problem
Algorithms now control what we see and do every day. Good design should work for everyone, not just big companies. How can we build tools that show people what algorithms know about them so they can take back control from platforms?
social media algorithms curate billions of feeds daily, shaping what we see and how we think. Yet users remain passive recipients. Recent platform transparency efforts show what algorithms display but don't show the personalised inferences the algorithm makes based on our attention patterns. This research closes that gap by bringing the algorithmic identity (John Cheney-Lippold, 2019) to surface by making the explainability of algorthms, narrative based rather than technical.
Research Method
Method:
A Research through Design (RtD) study, building a design probe workshop with 25 participants.
Process:
Upload 3 screenshots from instagram explore GPT-4o generates personalised narrative portrait participants reflect on their alorithmic identity. It enable participants to examine the digital selves constantly being constructed about them as highlighted by Shoshana Zuboff on surveillance capitalism.
"It helped me name the difference between who I think I am and how my online behaviour quietly shaped another version of me." - P13
Key Findings
Conceptual Gifting: Users gained agency not from seeing their data, but from receiving language for previously inarticulate patterns.
"I couldn't describe my 'aesthetic' or 'visual interest' properly, but the algorithm used words such as 'soft luxury', 'symmetry', 'curator-in-training', and 'domestic intelligence.'" (P24)
Algorithmic Authorship: Rtaher than asking "Does the algorithm see me accurately?', Participants began asking "who do I want the algorithm to help me become?" reframing the tool from mirror of past behaviour into scaffold for future intention.
"Sometimes I have to 'clean up' the explore page."
— P10
Identity First Design Framework: Ehsan & Riedl (2020) distinguish technical explainability (why this post?) from experiential explainability (what does this say about who I am becoming?). This study demonstrates that meaningful transparency requires behaviour- inferred identity over data-led information: showing users who their data has made them, not what their data shows.
"I give a goal to the system in the beginning of the year and it matches my content to it— for example, if I want to eat better, I want it to align my content to that." (P25)
Shifts agency from platform to users through reflection on Algorithmic Identity
Makes invisible systems emotionally available
Democratizes algorithmic literacy through accesible language
Why this? Why now?
Research shows that messages personalised through behavioural data can be significantly more persuasive than non-personalised ones, reinforcing feedback loops, echo chambers, and increasingly narrow worldviews., where users rarely get to see the profile being built about them.
As emerging tech builds on increasingly personalised products, regulations like the EU AI Act push for greater transparency, the question of how these systems should become legible is no longer just technical, but deeply personal and cultural. Yet most explainable AI systems focus on explaining how algorithms work, rather than helping people understand their relationship with them.
Instagram as emotional regulation
Attention shapes Identity
Many participants realised they used Instagram less for entertainment and more for emotional regulation, distraction, or escape.
Participants became aware that small behaviours, lingering, liking, scrolling quietly construct a second version of themselves online.
"I don't even know anymore… maybe inspiration? Being on IG makes me feel horrible and I feel so much better off it." — P21
"It reveals a side of myself I don't see because of my ego." — P17
Reflection comes from language, not data
Productive discomfort creates awareness
The strongest reactions came when AI gave participants vocabulary for things they already felt but could not previously articulate a phenomenon termed “conceptual gifting.”
Even inaccurate interpretations triggered reflection, because users questioned why the algorithm perceived them in a particular way.
"It felt like someone put what I have been struggling with lately, for quite a while in fact, into words. It was a call out, an invitation to consider what I knew I needed to consider but did not have proof for. Or permission for." — P15
"It helped me name the difference between who I think I am and how my online behavior quietly shapes another version of me." — P13