When mechanistic explanation fails: a neuroscientist rethinks what "understanding" means
burny_tech · x · 2026-10-21
Harvard neuroscientist Bence Ölveczky argues in a Transmitter perspective essay that the classic mechanistic approach—decompose a system into parts and explain each—worked beautifully for simple circuits like retinal motion processing and zebra finch song, but has stubbornly failed for the cortex and complex behaviors.
Such behaviors may depend on distributed, holistic system properties that get destroyed by decomposition, so researchers may need to embrace complexity and even adopt a new concept of "understanding" in neuroscience.
The poster notes the parallel with mechanistic interpretability in AI: both fields face the same question of what new methods—and what new definition of understanding—are needed when traditional mechanistic approaches stop working.
More from AGI Musings
- Rename 'alignment' to 'loyalty'? X users debate reframing AI alignment — lumpenspace · 2026-10-21
- AI is the wave, quantum computing the sea: the real bet is assurance, not speed — AryHHAry · 2026-10-21
- Report warns speculative data center debt sold to pension funds echoes subprime bubble — SumitGup · 2026-10-21
- Stripe new grad's AI-era guide: master a niche or be the model's 'meat proxy' — emilygsands · 2026-10-21
- Neuroscience hits the same wall as mech interp: complex behaviors resist mechanistic explanation — burny_tech · 2026-10-21
- Nearly 30% of US STEM PhD dissertation abstracts are now AI-written, NBER paper finds — lmoroney · 2026-10-21