Research Assistant / PhD Candidate (all genders) - Research Lab Challenging AI of the Cluster of Excellence Reasonable Artificial Intelligence

TU Darmstadt stands for excellent and relevant science. We are playing a decisive role in shaping far-reaching processes of global change—from energy transition to artificial intelligence—through outstanding scientific knowledge and innovative academic programmes. We group our cutting-edge research into three fields: Energy and Environment, Information and Intelligence, Matter and Materials. We are a university with strong ties to the Frankfurt Rhine-Main metropolitan region and a very strong international focus. We are committed to European values and European integration.

The Cluster of Excellence 3057 Reasonable Artificial Intelligence, RAI, was selected as one of 70 Clusters of Excellence to receive funding in a highly competitive process. The RAI Cluster of Excellence conducts research focusing on AI systems that acquire human-like communication and thinking skills, as well as the ability to recognize, classify, and adapt independently to new situations. The Cluster of Excellence conducts cutting-edge research at the highest international level with regular publications in flagship venues such as NeurIPS, AAAI, ACL, RSS, CVPR, SIGMOD, ICSE, Nature Communications, Nature Human Behaviour, Nature Machine Intelligence, and offers scientists in early career phases the best possible working and research conditions. RAI contributes to Germany's continued expansion of its international competitiveness in artificial intelligence.

The Cluster of Excellence is closely integrated with the Hessian Center for Artificial Intelligence (hessian.AI). The cluster has exclusive access to state-of-the-art GPU power (A100 and H100) through the AI supercomputers fortytwo and fortythree from hessian.AI, providing a reliable basis for cutting-edge research in AI training and inference. The Cluster of Excellence involves researchers from other universities, including the University of Bremen, Goethe University Frankfurt am Main, Saarland University Saarbrücken, the University of Tübingen, and Julius Maximilian University Würzburg.

RAI is structured into four Research Labs. Your tasks will focus on the Challenging AI Lab: CAI will help to understand the limitations of AI systems by confronting them with tasks that humans can handle but that current AI systems struggle with. The goal of Challenging AI with Cognitive Science (CAI) is to develop and co-evolve tasks that challenge current and future AI systems. This applies in particular to those developed in the research areas of Observational AI (OAI) and Active AI (AAI). This will enable testing and quantifying progress made towards the goals of RAI concerning human-like reasoning and learning capabilities. Specifically, we will develop tasks and benchmarks that evaluate (A) how well systems can bridge from low-level perception to high-level-reasoning by abstracting from training data and transferring knowledge, (B) how adaptable and flexible they acquire new skills in new situations, (C) how they actively discover new representations and knowledge, and (D) how understandable the symbolic explanations they find are to humans, and (E) how they perform theory-of-mind inference about themselves and other agents to enable cooperative interactions.

Possible supervisors/mentors with a focus on Challenging AI

• Prof. Dr. Kristian Kersting: Artificial Intelligence and Machine Learning
• Prof. Dr. Martin Mundt: Lifelong Machine Learning
• Prof. Dr. Gemma Roig: Computational Vision and AI
• Prof. Dr. Constantin Rothkopf: Psychology of Information Processing
• Prof. Dr. Angela Yu: Computational Modelling of Intelligent Systems

For further information, as well as a list of the participating professors, please refer to the RAI website: https://hessian.ai/projects/reasonable-ai-rai/. See the hessian.AI website under 'Graduate School' for information on what it is like to be a PhD candidate in Germany.

Your tasks will be related to the research conducted in the Research Lab Challenging AI with Cognitive Science.

Possible topics for dissertations are:

• Comparing AI models and humans on textual abstraction and reasoning tasks
• Comparing vision-language models and humans on abstraction and reasoning tasks
• Vision-language puzzles
• Causal puzzles: from SMART-101 to CAUSE-101
• CogLab: RAI models walk into a cognitive science lab
• Neuro-symbolic puzzles
• (Differentiable) Bongard solver
• Emergent world representations in games
• Assessing the scientific mind of RAI models
• Interactive compositional learning of functions in humans and machines
• Using games to understand RAI models
• Interacting with the world: causal physical sensorimotor interactions under risk
• Does a tight integration of learning and reasoning make AI systems smarter?
• Do AI system get smarter by interacting with the environment and other intelligent entities?
• Is there a compositional language of thought in AI models?
• What human biases are there in AI judgements?
• Do additional modalities make AI systems smarter?
• How do we design novel tests to accurately measure intelligence in both AI and humans?
• Can AI systems reason about their own behavior and human behavior?
• Can AI identify and predict humans’ states of mind in novel benchmarks?
• Can Theory of Mind improve AI communication and cooperation?

Candidates apply for the Challenging AI Research Lab, not for a specific PhD project and not for several labs in parallel. While the doctoral candidates should first discuss their respective research interests and identify possible supervising professors during the first year (lab rotation with permanently assigned mentors), the specific dissertation topics are determined by the beginning of the second year at the latest. The doctoral candidates are then supervised by two RAI professors, preferably across RAI Research Labs.

• University science degree (master's or equivalent) with very good performance in computer science, machine learning, artificial intelligence and cognitive science or related fields
• Very good programming skills, e.g. with deep (probabilistic) learning, compilers and/or robots
• Specific technical knowledge, that enables collaboration on the central research questions of the Challenging AI research lab and the cluster
• Very good English skills (written and spoken)
• Ability to cooperate in a dynamic and interdisciplinary research environment 
• Ability to work both independently and in a team, as well as a high degree of motivation and proactive behaviour

Technical University of Darmstadt offers varied and challenging assignments, freedom to work independently, the latest technologies, good collaboration between colleagues in partnership, needs-based training opportunities and customised personnel development.

Opportunity for further qualification (doctoral dissertation) is given. The fulfillment of the duties likewise enables the scientific qualifications of the candidate.

• Development and organisation – comprehensive in-house training offers, including the opportunity for continuing education and development;
• Annual leave/educational leave – 30 days annual leave (full-time employment) and 5 days educational leave;
• Sustainable and mobile – eligibility to free public transport in the state of Hesse with the LandesTicket Hessen (Hesse StateTicket) in accordance with the currently valid collective agreement, in addition to opportunities to working mobile at times;
• Fit and healthy – free of charge preventive medical check-ups and a wide-ranging subsidised sports programme;
• Work-life balance – flexible working time models, plus BGM (Betriebliches Gesundheitsmanagement – University Health Management);
• Pension scheme – supplementary public service pension scheme (VBL) in accordance with the currently applicable regulations;
• University bicycle
• Family-friendliness/compatibility of family/care/career – (university-run) childcare services, child allowance (based on the collective agreement), childcare programmes during school holidays.

TU Darmstadt intends to increase the number of female employees and encourages female candidates to apply. In case of equal qualifications, applicants with a degree of disability of at least 50 or equal will be given preference. Remuneration is in accordance with the collective agreement for the Technical University of Darmstadt (TV - TU Darmstadt). Part-time employment is generally possible.

By submitting your application, you agree that your data may be stored and processed for the purpose of filling the vacancy. You can find our privacy policy on our webpage.

If you have any questions about this position, please contact Dr. Susann Weißheit, Managing Director RAI, +49 6151 1628548 or susann.weissheit@hessian.ai

Please note the information provided in the application form regarding the application documents to be submitted.