Why connecting ML to the Internet is a Bad Idea, James Mitchens @ #usesec18 
James Mitchens is using Tay as a cautionary tale, and it's hilarious.
Spoiler: connecting antisocial people to a chatbot to train it... is a very very bad idea.
"It's no longer appropriate to use a cartesian approach to computer security." We need to change the way we think.
"The tech industry has become detached from the ramifications of the things they've built"
Machine learning used for criminal justice risk assessment incorrectly labels black defendants as high risk at twice the rate of white defendants.
"We need to question the manifest destiny to which we're being pushed. Because as we learned from Oregon Trail, manifest destiny often leads to dysentery."
The first assumption of tech manifest destiny is that tech is value neutral, and will automatically lead to good outcomes for everyone
Thus, new tech should be deployed as fast as possible, even if we don't know what its outcome will be.
Thus, new tech should be deployed as fast as possible, even if we don't know what its outcome will be.
History is generally uninteresting because the past has nothing to teach us (another assumption of Tech Manifest Destiny)
We don't look at how bias creeps into ML. We as technologists don't know the History Channel exists. We assume statistical models have never led us astray.
We only know two things about nutrition: you're gonna die, try not to get diabetes on the way out.
This is instructive to computer science.
This is instructive to computer science.
TLS is the only good thing we have
Self explanatory
All previous tweets in this thread quotes or paraphrase from this talk #usesec18
Mickens. Not Mitchens.
More name typos are bad.
plz visually edit all typos.
