My name is Marvin Maechler, and I'm a cognitive neuroscientist and AI Fellow at the University of Pennsylvania, where I work in Alan Stocker's lab. I study brains and computational models of perception and decision-making.
Cognitive biases
Why would your brain evolve to underuse evidence that contradicts your beliefs? My current research focuses on choice-induced bias: the tendency to behave as if past choices were useful guides. Rather than treating this as a flaw, I'm investigating normative reasons brains might have developed this bias. I am generally interested in the trade-offs that make brains efficient, particularly when that goes wrong.
Visual illusions
During my PhD at Dartmouth College (with Peter Tse and Patrick Cavanagh), I studied how the visual system gets fooled. We worked on tracking the double-drift illusion, an effect where objects appear perceptually in locations where they physically aren't. We asked whether attentional tracking would target the physical or perceived locations of this stimulus. It turns out that certain cognitive operations can only operate over the outputs of perception and not the inputs directly. Here the illusion, an apparent mistake of our perceptual system, serves as one of the best windows we have into how cognition and perception interact.
Cognitive Resources
We can only pay attention to so many things at once, but where do those limits come from? More concretely, when keeping track of multiple moving things at once, we are limited to a certain maximum number of trackable targets. The exact number depends on factors like how fast these objects move, but also which visual hemifield and hemisphere they are in. In one of my projects we uncovered the visual brain maps that are the most likely places for this bottleneck to arise in.
AI and perception
More recently (and with support from the Penn AI Fellowship), I've started asking whether artificial neural networks are susceptible to the same illusions and biases humans are. When do artificial and biological vision agree, and when do they diverge? Similarly, can we use what we have learned about efficiencies in brains and apply it to artificial neural networks?
Alien minds
I also did some octopus research during my PhD and I hope some of that will be published soon. One of the things we tried and failed to do was show that octopuses too experience illusions. More importantly, if they can be surprised by their own experience (i.e., by finding out they were tricked by an illusion) that would have given us insight into whether they have metacognitive faculties like ours despite being evolutionarily removed from us by hundreds of millions of years.
Everything else
I'm originally from Munich, Germany. In my free time I enjoy trying the many great burgers in the US, reading weird blogs, and playing DnD.