In yesterday's #MoneyStuff newsletter by @matt_levine @business, I really liked this quote:

"Another view of quant trading is that it reinforces market inefficiencies: Your algorithm notices some correlation, everyone else’s algorithm notices the same correlation, you all ...

... pile into the correlation, the correlation increases, and the original reason for the correlation goes away, or was spurious to begin with. The butterflies and the wheat prices were just a coincidence, but now all the hedge funds are buying wheat whenever they see ...

... butterflies, so wheat prices correlate with butterfly populations for no good reason. Wheat prices get further away from fundamentals, but eventually the fundamentals reassert themselves and the trade collapses.

I see a lot of the same happening in the field of #AI too unfortunately. Mitigating this is going to require deeper instrumentation and understanding of production models.

Some of the work from @mtlaiethics by @ErickGalinkin and I looks at this in Green Lighting #ML: #Confidentiality, #Integrity, and #Availability of #MachineLearning Systems in Deployment that we presented at @ICMLconf this past week

Would be great to hear from the #responsibleAI and #ethicalAI community on this - especially @KayFButterfield @mireillemoret @ShannonVallor @Abebab @timnitGebru @jovialjoy @rcalo who have a wealth of experience with real-world ML systems.

Tagging the team to engage in discussions @AllyCohen2 @alexandrineroy2 @CamLante @Csi_wright @MariannaBergama @ChiefMoAkif @muriamfancy @renjie @RyanKhurana @taniadegasperis @victoria_heath7