Days ago at Frankfurt airport I played this game where you dance to a music video in front of an Xbox and a camera captures your moment.
I set the highest score. The trick: to accentuate movements the ML recognizes, and forget about the rest.
Will it be a behavioral trend?
I set the highest score. The trick: to accentuate movements the ML recognizes, and forget about the rest.
Will it be a behavioral trend?
One of my theses re ML is that we’re overlooking how it will change human behavior, on the dimensions of:
- MLs training humans to change behaviors so they are more legible
- Humans changing their (verbal/behavioral/cognitive) vocabulary to match it to the MLs’ (eg autocorrect)
- MLs training humans to change behaviors so they are more legible
- Humans changing their (verbal/behavioral/cognitive) vocabulary to match it to the MLs’ (eg autocorrect)
3/ Another issue. In my gum.co/twtamg I described “prioritization by specificity”, a principle for which, when people receive information through multiple channels (eg verbal + body language) they prioritize the own which conveys info which tends to less interpretations
4/ This is marked in people with Autism SD. Since for them body language conveys very unspecific information, whereas verbal language is very specific, they disregard the former.
This is relevant to ML because…
This is relevant to ML because…
5/ Making MLs with a balanced prioritization by specificity will be hard;
and if we don’t, we encourage communication (and behavior) to happen on spefic channels or to become lower-entropy,
Changing the way we communicate with other humans too, if NNT’s minority rule kicks in.
and if we don’t, we encourage communication (and behavior) to happen on spefic channels or to become lower-entropy,
Changing the way we communicate with other humans too, if NNT’s minority rule kicks in.
6/ Oh, wait, it’s already happening
Opening a Pandora’s box of signaling adaptation
Q: Why is it worse?
A: Because it’s legible, explicit & low-variance.
Q: But won’t AI adapt to get cheaters?
A: Yes. But it will only work if it doesn’t adapt centrally.
Opening a Pandora’s box of signaling adaptation
Q: Why is it worse?
A: Because it’s legible, explicit & low-variance.
Q: But won’t AI adapt to get cheaters?
A: Yes. But it will only work if it doesn’t adapt centrally.
I’m not saying it will happen. I’m saying it will happen
- at least in a transition period
- and if ML will be centralized
Right now people-to-people signaling is limited by variance: different people infer different conclusions from the same signal.
- at least in a transition period
- and if ML will be centralized
Right now people-to-people signaling is limited by variance: different people infer different conclusions from the same signal.
