How much is that Algorithm Doggie in the Measurement Window?
Turns out, programming an algorithm use a lot energy & is expensive.
So much so that researchers are developing tools to measure how much cost vs return.
Need for us to understand the impact of this carbon foot print on the Environment.
Training an algorithm is not easy or cheap.
Involves days of processing by data center servers & chips.
One algorithm can produce emission footprint equal to lifetime of 5 cars.
Researchers & Scientists encouraged to mitigate their carbon footprint.
By using computers that run on solar power or hydroelectricity.
Resulting in less training time and data for an algorithm coding.
Tools are being developed to automatically determine amount of dioxide emitted.
Now have an online carbon emissions calculator for machine learning researchers.
Measures hardware type, hours, and cloud provider for each algorithm program.
“ If people could see the true cost of an algorithm systems, there would be harder questions asked about the convenience vs. the worth of the planetary cost.” – Kate Crawford, AI Now Institute, NYU.
“Our main goal is to just raise awareness” – Peter Henderson, PHD Computer Science, Stanford.
Companies – Element AI [Developed the Calculator]; Comet.ml [Track & compare machine learning experiments.
_”Researchers Focus on AI’s Energy Use” – WSJ July 20 – https://www.wsj.com/pro/artificial-intelligence