ALGOfest 2026 · AI & Machine Learning Track

Does AI make us
all the same?

MirrorMind is a 12-week longitudinal study measuring how AI recommendation algorithms reshape your expressed identity over time — using our novel Identity Drift Score.

60
Study participants
15.4×
× tighter IQR in treatment
0.033
Levene p-value (variance)
12 wks
Week study period
Identity Drift Score
IDS(t) =
0.5 × ΔSemantic(t)
+ 0.3 × ΔTopicEntropy(t)
+ 0.2 × ΔLexical(t)
Treatment group drift at W12
Semantic drift68%
Topic entropy52%
Lexical diversity44%
The study

Measuring algorithmic
identity capture

We introduce the Identity Drift Score (IDS) — a composite NLP metric tracking three dimensions of identity change: semantic embedding distance, topic entropy shift, and lexical diversity loss.

Both groups drifted from baseline. The difference was the shape of that drift — treatment group IQR was 15.4× tighter than control (Levene p = 0.033). The algorithm makes everyone drift the same way, regardless of where they started.

How it works

Built on rigorous methodology

Identity Drift Score

Novel composite metric combining semantic embedding distance, topic entropy, and lexical diversity loss into a single, interpretable number.

Longitudinal tracking

4 collection waves over 12 weeks. Watch your IDS evolve in real-time as you complete weekly survey responses.

NLP pipeline

Under the hood: NMF topic modeling, cosine similarity on sentence embeddings, KL-divergence, MTLD lexical diversity scoring.

IRB-approved & private

Fully anonymised at ingestion. SHA-256 participant IDs. GDPR-compliant data exports only. Right to withdraw at any time.

Key findings

What the data shows

Results from 60 participants across 12 weeks, with Bonferroni-corrected statistical validation.

Treatment group IQR was 15.4× tighter than control (0.082 vs 1.263) — Levene F(1,58)=4.80, p=0.033. The algorithm produced uniform, convergent drift across participants who started differently.

Both groups drifted significantly from baseline (Wilcoxon p<0.001 for both). The finding is not about magnitude — it's about consistency. Control drift was individual. Treatment drift was convergent.

IDS slope variance: treatment SD=0.135 vs control SD=0.399 (Levene p=0.039). The homogenization effect holds for rate of change, not just final IDS.

Know your drift score.

Join the study and receive a personalised Identity Drift Score report after 12 weeks. All data is anonymised and IRB-approved.