USC · Thomas Lord Department of Computer Science

We study how humans and machines shape each other.

The HUMANS Lab (HUmans, MAchines, Networks & Social Systems): A USC group studying bots, AI safety & human-machine societies.

A research group of the Machine Intelligence & Data Science (MINDS) initiative · directed by Emilio Ferrara

! For the upcoming PhD cycle I can only recruit students who hold an individual fellowship. If you have one and want to join us, please reach out.

Emilio FerraraLuca LuceriFaith BacaLeonardo BlasEun Cheol ChoiPriyanka DeyPatrick GerardBumju JungDaniel RuizVictoria Bingxin XuJinyi YeSiyi ZhouGrace LiAmisha NambiarPreyashi PoddarPeiran QiuGiovanni VurroLei CaoKiran KhannaDing (Blair) ChenKatie HancockSanthosh NarayananRichard ZhangAmirkhan SerikbayMatteo BenvenutiMeilin Pan

The people of the HUMANS Lab.

Now

Latest from the lab

auto-synced — arXiv · Crossref · GitHub
  • PreprintOpen-Weight Masked Introspection, on whether LLMs can report what changes inside them, is out as a preprint on arXiv
  • PaperPALMs, on modeling population preferences in LLMs with multi-construct grounded rationales, is accepted to the EMNLP 2026 main conference
  • PaperTemporal Signatures of Coordination, on detecting and anticipating influence operations, is accepted at EPJ Data Science
  • PaperFive Myths About Influence Operations, from 25 million tweets across seven state campaigns, published in Systems
  • PaperMOSAIC, our benchmark for the moral, social, and individual dimensions of LLMs, is accepted to the CIKM 2026 Resource Track
  • PaperEssay in Tech Policy Press: the model you audit is not the model you ship
  • PaperOur work on backdoor attacks in vision-language models is accepted to Findings of ACL 2026
  • TalkClosing lecture at SICSS Singapore (NUS) — From Bots to Agents
  • PreprintManufactured Divisiveness, on invective across seven state-backed influence campaigns, posted as arXiv:2607.14491
  • PaperCommunity Detection as Adaptive Diffusion accepted at ACM AI Letters
  • DatasetWe release QuantiBias, a benchmark for quantization-induced bias in LLMs, on Hugging Face
  • PaperDefeat Devices in AI Systems published in Future Internet
  • PaperECHO: Encoding Communities via High-order Operators published in Machine Learning with Applications
  • PaperTwo papers at ICWSM 2026: coordinated inauthentic behavior on TikTok, and cross-platform narrative networks
  • PaperChange is Hard: Consistent Player Behavior Across Games with Conflicting Incentives accepted at CHI PLAY 2026

Programs

What we work on

What if societies and machines are one system?

01

Social Bots & Influence Ops

Detecting automated and coordinated accounts; measuring platform manipulation at scale.

02

AI Safety & Bias Auditing

Behavioral audits of LLMs — deceptive behavior, quantization-induced bias, cultural safety.

03

Misinformation & Integrity

How false and divisive content spreads, and which interventions actually work.

04

Network Science

Community detection and diffusion — from theory to scalable open-source tools.

05

Computational Social Science

Large-scale measurement of collective behavior: elections, crises, human-AI interaction.

06

Human-Centered AI

Designing AI that serves people and communities: interaction, trust, and alignment.

INITIATIVE

The Election Integrity Initiative

Our standing election-integrity effort: public datasets, analysis, and briefings on online manipulation around the U.S. elections.

election-integrity.online ↗

The team

People

The lab is its people, across computer science and the social sciences.

Leadership
Emilio Ferrara
Emilio FerraraPI, Professor
Luca Luceri
Luca LuceriLead Scientist, Research Asst. Prof
PhD Students
Students & Researchers

Alumni hall of fame. Former members now teach at UCLA, Dartmouth, and Minnesota, and research at Google DeepMind, Meta, and Amazon. Where they are →

Selected

Publications

A selection of the lab’s work, kept current from arXiv, Crossref, and ORCID.

→ all publications

Open Science

Data & Software

Benchmarks, corpora, and tools the community builds on.

OWMI ↗

Open-Weight Masked Introspection: a framework for measuring what LLMs can report about their own computation.

huggingface.co2026

QuantiBias ↗

Benchmark for quantization-induced bias and safety drift in compressed LLMs.

huggingface.co2026

MOSAIC ↗

Benchmark probing the moral, social, and individual value dimensions of LLMs.

huggingface.co2026

ECHO-GNN ↗

Graph-neural community detection: encoding communities via high-order operators.

github.com2026

Generalized Louvain ↗

A scalable, open-source revival of the generalized Louvain method for community detection.

github.com2026

EDTok ↗

TikTok video IDs for eating-disorder content, collected via TikTok's Research API.

github.com2025

Telegram 2024 Election Posts ↗

The largest public Telegram dataset on the 2024 U.S. presidential election.

github.com2024

X/Twitter 2024 Election Discourse ↗

Large-scale discourse on X tracking the 2024 U.S. presidential election.

github.com2024

Truth Social 2024 Election Posts ↗

1.5 million Truth Social posts on the 2024 election, Feb 2022 to Oct 2024.

github.com2024

TikTok 2024 Election Videos ↗

Video IDs for 2024 election content, collected via TikTok's Research API.

github.com2024

GET-Tok (Peru) ↗

Multimodal TikTok data on the 2022 attempted coup in Peru, enriched with Whisper and GPT-4.

github.com2024

Ukraine-Russia Conflict Tweets ↗

An ongoing collection of tweet IDs on the war between Ukraine and Russia, since Feb 2022.

github.com2022

COVID-19 Misinformation Labels ↗

Large-scale labeled misinformation cascades on COVID-19 vaccines, built by label refinement.

github.com2022

US 2020 Election Tweets ↗

A large-scale Twitter dataset tracking the 2020 U.S. presidential election.

github.com2020

COVID-19 Twitter Dataset ↗

One of the most widely used pandemic social-media corpora in the field.

github.com2020

League of Legends Dataset ↗

Behavioral traces from team-based online games for studying performance.

Harvard Dataverse2018

GEM ↗

Graph Embedding Methods: a Python library of node-embedding algorithms (Goyal & Ferrara).

github.com2018
→ all datasets→ code & resources