sarah jeong

@mariuswatz @katecrawford And anything there is going to be horribly gameable.

@Dymaxion @mariuswatz bingo - that's what our paper found. Not even A++ AI cuts it. The context is too shifting and complex.

@katecrawford @Dymaxion @mariuswatz So how do we convince these companies that blind faith in the almighty algorithm isn't the way to go?

@bhpascal @katecrawford @mariuswatz e.g. things like publishable/subscribable blocklists, keyword filters. & they have new failure modes.

@Dymaxion @katecrawford @mariuswatz Hm, but underlying problem I'm seeing is belief by companies that computer science can fix everything.

@bhpascal @Dymaxion @mariuswatz I think we agree that's the underlying problem. :-)

@katecrawford @Dymaxion My point: Make abuse (more) expensive by making fake accts harder to run. Raise the threshold + effort needed.

@mariuswatz @katecrawford In most cases, it's high-follower accounts turning thousands of normal people into a hate machine.