I understand very well how 2-of-2 lightning channels work, but 2-of-2 doesn't have risk of signatory collusion.

Lightning Factory details are interesting— thanks for the paper.

I'm going to blow your mind for a minute. You know lightning channels? Just the normal ones we make with CLN or LND. Those are n-of-n (2-of-2) multi-sigs with unilateral exit.

Its because you can exit by yourself if you follow certain rules, but the other party (or parties in the case of multi-party channels) have recourse if you try to exit using an old state.

Yes, 1 party can become uncooperative (this is where the all signatories must sign all transactions part comes in) and thereby prevent a payment from going through, but you can exit without them and them reopen with everyone except them later, and you can exit without their signature. 

"Using Lightning Factories, there is no risk for a counter party going unresponsive, since the factory can be closed uncooperatively with a small locktime ∆t to allow for disputes, instead of a period of time representative of
29
the lifetime of the factory"

This paper I linked for you is just one of many proposals for multi-party channels and it does have a concept of justice transactions

nerd2ninja

Hmm, my understanding is that the all signatories must sign all transactions in a n-of-n. So there is no unilateral exit and a single signatory has the ability to censor a given transaction. Isn't that what gives immunity to signatory collusion?

lightning

Scaling LN routing with Chaumian eCash

Well its n-of-n (so no signatory collusion) with unilateral exit. Even if you assume a trusted multi-party channel, if almost all members of the multi-party channel became uncoorperative (died, went offline, what have you) you can still exit (get your funds out) without them. Even if you assume a pretty bad multi-party channel where you're using current day mainnet and with no justice transactions (because you trust so much) the unilateral exit is still relevant. 

There is a multi-party channel proposals that do have justice transactions by the way, just want to put that out there: https://eprint.iacr.org/2018/918.pdf

AND when you build all of that out, you don't have to build an interoperability layer or convince the network to use a different routing mechanism like you guys were talking about in another comment