The Center for Humans and Machines (CHM) at the Max-Planck-Institute for Human  Development in Berlin seeks Applicants for

Voluntary Internships 
39,00 hours/week

Internships will last between 6-12 weeks and start from July 21, 2025.

The Center for Humans and Machines (CHM) at the Max-Planck-Institute for Human Development in Berlin conducts interdisciplinary science to understand, anticipate, and shape major disruptions from digital media and Artificial Intelligence (AI) to the way we think, learn, work, play, cooperate, and govern. Our goal is to understand how machines are shaping human society today and how they may continue to shape it in the future. The Center is comprised of an interdisciplinary, international, and diverse group of scholars, and a science support team.

This summer, we are looking for motivated student interns who are excited about working at the intersection of computer science and social sciences. The following projects are on offer:

Project 1: Reinforcement learning on human teaching
By tuning large language models (LLMs) to enhance users’ skills rather than merely satisfy their preferences, we aim to create AI assistants that empower rather than entice. In our study, participants engage in challenging tasks—such as writing exercises, game-based puzzles, or coding problems—while being assisted by an AI system. We compare two systems: one optimized for immediate task success and another optimized for long-term learning outcomes. Our investigation focuses on how the underlying optimization objective shapes the AI's communicative strategies and the generalizability of the information it conveys. Although ambitious, the project is building upon extensive expertise in the lab with related setups.
Project Type: hypothesis-driven; Prime Audience: Science; Lead Researcher: Levin Brinkmann

• Internship on Human-Machine Interaction
  Design and develop a web-based application to support a human–AI interaction experiment. Responsibilities include implementing the experimental interface and piloting the study with online participants.
  Required Experience: Practical experience in web development
  Beneficial Experience: React, FastAPI (Python)
  Duration: 8 to 12 weeks

• Internship on LLM integration and fine-tuning
  Tasks include simulating participant responses, training a value model to rank AI-generated messages based on task performance and learning outcomes, and exploring reinforcement learning (RL)-based fine-tuning techniques.
  Required Knowledge: Python; Machine Learning Fundamentals
  Beneficial Experience: Hands-on experience with fine-tuning of LLMs and High-Performance Computing
  Duration: 8 to 12 weeks
  

Project 2: Internal representations of Emotions in LLMs
This project explores the viability of mechanistic interpretability methods and toolkits for identifying and monitoring the internal circuits of large language models (LLMs) that may represent emotional states. Building on recent advances in the interpretability of Transformer architectures, we study state-of-the-art theories and tools (e.g., TransformerLens, NNsight) to investigate how emotions are internally encoded, detected, or simulated within these models.
Project Type: explorative; Prime Audience: Science; Lead Researcher: Luis Mienhardt

• Internship on Neural Interpretation
  The internship will involve hands-on experimentation with interpretability toolkits, comparative analysis of internal representations, and attempts to isolate emotion-related subnetworks.
  Required Knowledge: Python, Machine Learning Fundamentals
  Beneficial Experience: Hands-on experience with fine-tuning of LLMs and methods of interpretability
  Duration: 6 to 12 weeks

Project 3: Elderbot: AI to combat loneliness among the elderly
Elderbot investigates the potential of a generative, AI-based social agent to reduce loneliness in older adults (60+). We examine whether such technology, used anonymously in private households, can foster social interaction, strengthen social skills and self-confidence, and thus contribute to improved mental health. The study explores the feasibility and effectiveness of this novel intervention within the context of limited access to traditional support services and increasing loneliness in aging populations. Through this work, we aim to better understand how responsible and ethical use of conversational AI may offer scalable, low-threshold support in mental health care.
Project Type: hypothesis-driven; Prime Audience: Science; Lead Researcher: Rodrigo Bermudez Schettino and Chaewon Yun

• Internship on LLM integration into Mobile App
  Migrate deprecated MongoDB integration in Android app written in Kotlin from Atlas Device Sync to Parse. The role involves future-proofing the existing software stack of an experiment. It requires evaluating database integrations to save conversations conducted via an Android app into a cloud database.
  Required Experience: Software Engineering, RestAPI basics
  Beneficial Experience: Android app development (Kotlin), MongoDB, Parse
  Duration: 12 weeks

• Internship on Online Survey Design
  Prepare a Qualtrics survey to collect data from study participants before and after the experiment. This internship gives an insight into the design of online surveys to collect data during research experiments.
  Required Experience: Basic understanding of online survey platforms
  Beneficial Experience: Qualtrics
  Duration: 6 to 12 weeks

Project 4: The influence of LLM vs Human Generated Options on Moral Judgment
Much of human moral judgment depends on the ability to determine which possibilities are relevant in a given situation. Assessing an agent’s responsibility requires considering what other actions were available to them. The extent to which we judge an agent responsible for an action or outcome reflects the alternative actions perceived as available. As humans increasingly rely on Large Language Models (LLMs) for idea generation and sampling of actions, the set of possibilities they consider may expand beyond what they would typically generate on their own. This shift in the perceived range of available actions may, in turn, affect how responsibility and blame are assigned. In this project, we explore whether the consideration of self vs. LLM generated possibilities influence people’s blame judgments.
Project Type: follow up project, project consolidation, exploratory; Prime Audience: Science; Lead Researcher: Lara Kirfel

• Internship on Online Survey Design & Data Analysis
  Develop an online study that investigates how human vs. LLM-generated options influence moral judgment, decision-making and norm perception. Responsibilities include implementing the experimental interface and running the study with online participants. The intern might also assist with data analysis and academic write-up.
  Required Experience: python, web-based development, otree, qualtrics
  Beneficial Experience: R, NLP, Psychology, Cognitive science
  Duration: 6-9 weeks

Project 5: Observatory of Collective Experience
The Observatory of Collective Experience (OCE) is an online platform that invites visitors to engage with a generative AI system by sharing personal stories and emotions, thereby co-creating machine-mediated collective experiences. The project explores whether fine-tuning on emotional narratives can enhance the empathic capacities of AI systems—such as emotional mirroring, context-sensitive response generation, and the recognition and modulation of affective states. Like earlier initiatives at the Center for Humans and Machines—such as Spook the Machine—the OCE operates at the intersection of art, science, and technology.
Project Type: explorative; Prime Audience: Public and Science; Lead Researcher: Levin Brinkmann

• Internship on Human-Machine Interaction
  Design and develop the OCE web application. This role involves building an interactive platform that facilitates human–AI interaction through the sharing of personal stories and emotional experiences.
  Required Experience: Practical experience in web development
  Beneficial Experience: React, FastAPI (Python), Google Firebase, Figma
  Duration: 8 to 12 weeks

• Internship on LLM integration and fine-tuning
  Develop the AI backend for the OCE and fine-tune a LLM using simulated visitor responses. Tasks include implementing backend services, designing APIs, and training models for emotionally responsive interaction.
  Required Knowledge: Python, Machine Learning Fundamentals, RestAPI Basics
  Beneficial Experience: Hands-on experience with fine-tuning of LLMs, API design and High-Performance Computing
  Duration: 8 to 12 weeks

Formal requirements:

• Availability at 39 hours/week for the duration of the internship (depending on the project)
• Currently enrolled in a program at a German university related to the project(s) 
• Attendance at the institute in Berlin (no remote working)

What we offer

• Get hands-on experience with research on human-AI interaction and scientific processes

• Close collaboration with CHM researchers

• Possibility of transitioning to position of Student Research Assistant

• Salary: 450€ / month (39 h/week)

The Max Planck Society strives for gender and diversity equality. We welcome applications from all backgrounds. The Max Planck Society is committed to increasing the number of individuals with disabilities in its workforce and therefore encourages applications from such qualified individuals.

Please apply, including a CV without a photo and your latest transcript, via our career site on this page. Please apply until June 1, 2025.

Applications will be reviewed in the first week of June. Interviews will be scheduled 10th-13th of June.

The data protection declaration for the processing of personal data within the scope of your application can be found here:
https://www.mpib-berlin.mpg.de/1589569/en_infos_bewerbung.pdf