If we are to understand what makes human intelligence unique, we need to demystify high-level intelligence, or ‘common sense’. Much of my own research program focuses on the case study of ethical intelligence, aka ‘moral reasoning’. In particular, I study:


(1) How ethical considerations affect our everyday thoughts about various concepts that are relevant to how we should live our lives, long studied by philosophers— including the true self, happiness, and meaning in life.

(2) What perceptual and cognitive capacities underlie our ability to behave ethically in social life, including capacities like animate attention, theory of mind, recursion, common knowledge, and the ability to convert sensory representations to a mental format that guides social inference and behavior.

(3) How ethical intelligence relates to real-world policy issues, especially in the domain of machine ethics. 


I study these phenomena by explaining both their:

(1) Cognitive basis (through experimental psychology methods); and

(2) Origins (by engineering AI systems that give rise to them).


In practice, this means that I employ a diverse set of empirical and engineering methods, including measuring judgments and decision-making, psychophysics, eye-tracking (fixed and mobile), and VR in adults; and by employing various AI models (e.g., various supervised and unsupervised models in computer vision, reinforcement learning, and natural language modeling).


Together with colleagues, I also work some with developmental and neural data, and, theoretically, I draw heavy inspiration from philosophy, game theory, and evolutionary biology.


The True Self

Representations of and beliefs about the concept of “a self” vary across cultures, perspectives (first vs. third), and individuals. Yet my collaborators and I have found evidence suggesting that people exhibit a robust, invariant tendency to believe that deep inside every individual there is a “good true self” calling them to behave in a morally virtuous manner. We propose that this belief arises from a general cognitive tendency known as psychological essentialism.

Common Knowledge and Recursive Mentalizing

Most work in psychology has studied the representation of other's beliefs about the world, aka theory of mind. My collaborators and I have investigated how representations of knowledge -- including knowledge that others have about our own beliefs (e.g., you know X, I know that you know X), and common knowledge (you know X, I know that you know X, you know that I know that you know X, ad infinitum) -- affect diverse social phenomena such as the bystander effect and perceptions of charitability. We propose that -- rather than being represented as an explicit, multiply nested proposition -- common knowledge may be a distinctive cognitive state, corresponding to the sense that something is public or "out there". 


Origins and Variability of Animate Attention

This work explores the idea that complex behaviors, like animate attention, can be explained by the interaction of a world model (which predicts future states of the world) and an intrinsically-motivated self model (which motivates the agent to spend time predicting parts of the world with certain features). In particular, we find that paying attention to aspects of the environment where one is continuing to learn new information (which we term progress curiosity) is a particularly powerful way to give rise to human-like behaviors like animate attention, without the need for built-in modules or hand-written rules. This work utilizes a 3D, photorealistic environment that we created for measuring artificial and human agents-- either while they wear mobile eye trackers or virtual reality goggles. We also run setups in which the displays are conveyed using real robots. 

  • Kim, K. H., Sano, M., De Freitas, J., Yamins, D. L. K., Haber, N. Towards modeling the variability of human attention. International Conference on Learning Representations Workshop

  • Gan, C., Schwartz, J., Alter, S., Schrimpf, M., Traer, J., De Freitas, J., Bhandwaldar, A., Sano, M., Kim, K. H., Wang, E., Mrowca, D., Lingelbach, M., Curtis, A., Feigelis, K., Haber, N., Gutfreund, D., Cox, D., DiCarlo, J., McDermott, J., Tenenbaum, J., Yamins, D. L. K. ThreeDWorld: A Platform for interactive, multi-modal physical simulation. arxiv. []

Ethics of Autonomous Machines

How do we create complex autonomous systems that have 'ethically acceptable' behavior? The current work considers the case study of autonomous vehicles, since for the first time in history they are a truly autonomous system that is beginning to operate in populated environments. Ethics is relevant to how they should be programmed, regulated, be perceived.

  • De Freitas, J., Censi, A., Anthony, S. E., Di Lillo, L., Smith, B. W., & Frazzoli, E. From driverless dilemmas to more practical ethics tests for autonomous vehicles. PsyArXiv

  • De Freitas, J., Anthony, S. A., Censi, A., & Alvarez, G. A. (2020). Doubting driverless dilemmas. Perspectives on Psychological Science. 

  • De Freitas & Cikara. Deliberately prejudiced self-driving vehicles elicit the most outrage. PsyArXiv.

Bridging Sensory Representations to Social Inference

How does sensory input get transformed into representations that serve full-blown social inference?

  • De Freitas, J., & Alvarez, G. A. (2018). Your visual system provides all the information you need to make moral judgments about generic visual events. Cognition, 178, 133–146. 

  • De Freitas, J., Hafri, A., Alvarez, G. A., Yamins, D. L. K. From pixels to moral judgment: Extracting morally-relevant information in minds and machines. Journal of Vision

  • Tarhan, L., De Freitas, J., Alvarez, G. A., & Konkle, T. Semantic embeddings of verbal descriptions predict action similarity judgments. Journal of Vision

The Limit of Self-Representation

This work tries to characterize the limits of self-representation, much as vision scientists have studied the limits of processes like working memory and attention. We do this by focusing both on intuitive psychology, and previous work on the so-called self-reference effect. This previous work finds that visual perceptual learning and memory are enhanced by "self-relevant" material, implying a "self" representation that is associated with visual information, thereby to these performance improvements. Applying an approach from perception science, we ask: is there a limit to this process? That is, can you have multiple "self representations" at once (e.g., "young me, old me"), or can there only be one self? We find that self-representation appears to be severely limited: the mind can only represent one item as the self at a time.  

  • De Freitas, J., Rips, L. J., & Alvarez, G. A. The capacity of self-relevant long-term memories. PsyArXiv.

  • De Freitas, J., Rips, L. J., & Alvarez, G. A. The limit of active self-representation. PsyArXiv.

  • De Freitas, J., Rips, L. J., & Alvarez, G. A. Personal identity, self-reference, and intransitivity. Under review. 



Other Papers