YES! I was trying to think of a way to express something like this this morning and you've nailed it. It really hit me after reading that bike messenger thread:https://twitter.com/mbgerring/status/993916520408363009 …
-
-
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
Fair, but Duplex takes it further by treating the human auditory comprehension system as just another API.
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
Imagine 7000B people as api endpoints. Imagine the power of Social Engineering leveraged by a neural network. Imagine you cannot trust voice over digital anymore. Imagine all the people living for today ... they said.
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
Agreed, was thinking of the nice people at my favorite local restaurant who definitely deserve better than this shoddy treatment. I'm in a ruralish area though, so appification of other humans hasn't really come here (yet)
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
I can’t believe they fucking named the thing after duplicity
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
Serious argument I've heard today: requiring AI's to identify themselves is the first step toward AI racism and anti-robot pogroms. (Serious meaning "it was not meant as a joke," not "deserves serious consideration.")
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
Oh good point, wow.
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
i think service industry workers & customers would both prefer we build machine to machine
#m2m systems as alternatives but we can't stomach nor innovate around@opentable's stranglehold as intermediator/data-aggregator.https://twitter.com/rektide/status/994280863826415629 … -
or, as
@cemerick said, google like us all is desperate to get outside of other's designed nexus of control.https://mobile.twitter.com/cemerick/status/915279337712230400 …
Fin de la conversation
Nouvelle conversation -
-
-
I think I'm most creeped out by the "um's" and such that they add. Makes it unequivocally clear they are trying to deceive the other party into thinking it's a real person
-
"they" in this case is the machine that learned to use those phrases. A big misconception is saying they are programmed to act this way. No, ML takes inputs and you tell it what do you want as output (a booking) and the machine tries all permutations until it succeeds.
-
Maybe I need to do more research, but from the demo it sounded like it relied pretty heavily on pre-recorded lines, not machine-generated speech. Besides, choice of training set (and/or decision to use overly-realistic model) is still down to devs
-
a huge/massive log of calls is fed to the ML engine. You give it a criteria for success (booking). The ML engine tries all sorts of permutations on its own, learning which one will work best. It's possible the machine realized that saying "hmmhm" meant less hung up calls.
-
Maybe it realized that most humans would hang up if they realize it's an AI so the ML Engine found out adding those words decreases the probability of it. NOBODY tells the machine which words to use (so nobody told the AI to say "hmmhm").
-
So there isn't a secret cabal of people thinking "lets make it pass as human" as an end goal. The AI on its own realized it needed to pass as human as much as possible to avoid being hung up on to increase the AI success rate in performing the bookings.
-
I think you're giving too much credit to the capabilities of ML. I think someone went "hey if we add filler words it sounds more realistic" and no one stopped to ask if that was the right call.
-
afaik Machine Learning is definitely not at the point where you can just point an ML black box at a bunch of call logs and expect it to learn how to make a phone call successfully. Has to be broken down into individual subsystems, each of which is heavily tuned by hand
- 3 réponses de plus
Nouvelle conversation -
-
-
Where this can get creepy [if it is not creepy enough!], imagine a country with malicious intent, and they develop an agent lets call it "Angela Merkel", that has sampled her characteristics, and it is then used to call Donald Trump to create a diplomatic incident..
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
-
-
maybe you should stop doing that then
Merci. Twitter en tiendra compte pour améliorer votre fil. SupprimerSupprimer
-
Le chargement semble prendre du temps.
Twitter est peut-être en surcapacité ou rencontre momentanément un incident. Réessayez ou rendez-vous sur la page Twitter Status pour plus d'informations.