Day 4 at the 90th Cold Spring Harbor Laboratory Symposium on Quantitative Biology. A day of provocations -- from neuroscience, agents, and a public lecture that brought the week into focus.
CSHL’s
@TonyZador opened the NeuroAI session with a challenge: AI can reason, but biology can sense, move, and act. The missing ingredient may not be better learning algorithms, it may be the right priors, shaped by evolution’s genomic “compression strategy.”
@sejnowski (
@salkinstitute) followed on the same gap from another angle: how cortical traveling waves handle temporal context in ways our architectures still can't match. Two giants, one morning, asking the same question from different directions.
Then
@JonahCool (
@AnthropicAI) gave the most honest talk of the week. He shared how critic agents can challenge AI outputs in real time and revealed new insights into how models like Claude represent concepts internally.
@marinkazitnik (
@Harvard ) showcased AI agents grounded in biology, including PROTON and newly launched AutoScientists teams of agents that generate hypotheses, critique one another, and learn from failure.
To close out the evening
@ewanbirney took the stage for the Dorcas Cummings Lecture. A CSHL alumnus,
@emblebi Director, co-author of the human genome in 2001. He offered a simple test for AI: Goal. Metric. Data. Do you trust them? He demonstrated it with Delphi-2M, a model that predicts diagnoses from health records. Trained on UK Biobank data and validated on 1.9 million Danish patients, its performance held.
Birney's closing message: AI has already revolutionized biology and we're still in the foothills.
After the lecture, some neighbors of the lab hosted selected meeting participants for dinner parties at their homes. The rest headed to the CSHL beach for a picnic.
Some conversations need open sky.
#CSHL AI in Biology Symposium is like a lit fire and by the last night, everyone was adding wood.
#CSHLSYMP26 #AIinBiology #NeuroAI #InspiringDiscovery #ScienceMakesLifeBetter