I am a researcher working at the crossroads of Artificial Intelligence, Robotics, and Cognitive Science. The goal of my research is to build human-aware AI systems, i.e. systems that take into account humans in their decision-making process.

Building human-aware AI systems raises several challenges. In order to efficiently take into account humans, AI systems should not treat humans as black boxes. Unlike ant physical object, humans are capable of sophisticated reasoning and are subject to inter-individual variability due to the specificity of each person's preferences, personality and biases. All these factors contribute to the complexity of designing human-aware AI systems.

During my PhD at the Institute for Intelligent Systems and Robotics, I designed a Machine Learning framework that enables humans to teach robots new tasks by using feedback and instructions. The main contribution of this work was to remove the need to hand-code the meaning of instructions, in order to provide more flexibility to the human in the way he/she wants to communicate with the robot.

After my PhD, I joined the Human Reinforcement Learning Team, to investigate the computational processes underlying human learning, from both autonomous and social learning perspectives. Beyond the importance of answering fundamental questions about human cognition, the computational modeling of cognitive processes is an important building block for building human-aware AI systems. The main advantage of this approach is to look beyond behavioral patterns by investigating the causal relationships between cognitive processes underlying human behavior.

Even though neuroscience and AI have always been in synergy throughout history, both domains have evolved in different directions, which lead to the development of distinct research communities with different research questions and methodologies. Nevertheless, there is still some common threads that lie at the crossroads of these two distinct disciplines. One example of such aspects is the Reinforcement Learning framework (RL), which has been invented by behavioral psychologists, extended by AI researchers, and then readopted by the neuroscience community.

As such, RL is a powerful computational framework that has a wide range of potential applications from both engineering and neuroscience perspectives. For example, it can be used to build artificial agents capable of beating humans in playing games, or to explain the mechanisms underlying psychiatric and neurological diseases. Notably, RL was the main link between my PhD thesis and my postdoc research. In the former case, it was used to make robots capable of learning new tasks by themselves and from humans; in the latter case, it was used to explain how humans learn to solve problems by themselves and from other humans.

My current objective is to push further towards merging AI and Cognitive Science, while tackling real-world problems in order to build better machines for better human well-being.