Want to watch reductionism degenerate into reductio ad absurdum?
Want to see what happens when “my truth” gets installed in the hive minds of our nascent AI overlords?
Want to watch Karl Popper turn over in his grave as circular unfalsifiability gets built into AI arbiters of “settled science?”
Then stay tuned as AIs start training on “synthetic data” conjured up by … other AIs.
MIT’s Laboratory for Information & Decision Systems (LIDS) is at the forefront of the Synthetic Data movement. According to LIDS, synthetic data is generated by algorithmically extracting truthiness from a smidgen of reality in a way that “captures all the underlying rules and infinite patterns that exist in real data.”
Might this leave the barn door open for bias and mischief? Fear not! Credentialed Experts™ considered all the pros and cons and promise that the data they make up will be fully diverse, equitable, and inclusive.
There’s nothing new about synthesizing data. Climate scientists perfected the art when they learned how to virtually impute temperature data from weather stations that were shut down decades ago. And no one bats an eye when they characterize the complex climate of the entire planet for a whole year with a single number. Not only that but to support the conclusion that every new year is the hottest ever, they can calculate this synthetic super-datum to five hundredths of a degree of accuracy!
Scoff if you want, but these data manipulation techniques made Al Gore a billionaire and convinced a generation to stop having babies and start eating bugs.
There’s no telling what similar feats AI can accomplish when trained on the right synthetic data.
Story suggested by MIT News