12

A

RTIFICIAL intelligence may be impressive,

but it sure is hungry. In September, the

computing giant Microsoft announced a

20-year deal to buy all the power from a restarted

nuclear reactor in Pennsylvania, US, just to feed

its ever-growing AI services. By comparison, the

human brain is estimated to be a million times more

efficient. Whatever daily calculations Microsoft will

be able to perform with its extra 835 megawatts could,

theoretically, be accomplished by a dozen people on a

good breakfast.

Admittedly, this is an unfair comparison: AI and

humans have different strengths. Nevertheless, the

brain is a fascinating organ, and one that remains

poorly understood. The reason is its sheer complexity,

with upwards of 80 billion neurons and 100 trillion

connections. Conventional imaging techniques have

a hard time untangling this network – what scientists

refer to as the “connectome”. Magnetic resonance

imaging MRI is able to scan entire human brains

but only down to millimetre resolution which is

far too coarse to pick out individual neurons on the

E S R F A P A C U R E A N U

Four years ago, the ESRF’s Alexandra

Pacureanu began a research project to

construct “connectomes” of the neurons

inside brains with EBS X-rays. T he f irst

results are already helping us understand

how nervous systems work.

CONNECTOMICS

Connecting

the dots

Conventional imaging

techniques have a hard time

untangling the connectome