The Lightning Network solves the problem of the decentralized commit

Before reading this, see Ripple and the problem of the decentralized commit.

The Bitcoin Lightning Network can be thought as a system similar to Ripple: there are conditional IOUs (HTLCs) that are sent in "prepare"-like messages across a route, and a secret p that must travel from the final receiver backwards through the route until it reaches the initial sender and possession of that secret serves to prove the payment as well as to make the IOU hold true.

The difference is that if one of the parties don't send the "acknowledge" in time, the other has a trusted third-party with its own clock (that is the clock that is valid for everybody involved) to complain immediately at the timeout: the Bitcoin blockchain. If C has p and B isn't acknowleding it, C tells the Bitcoin blockchain and it will force the transfer of the amount from B to C.

Differences (or 1 upside and 3 downside)

1. The Lightning Network differs from a "pure" Ripple network in that when we send a "prepare" message on the Lightning Network, unlike on a pure Ripple network we're not just promising we will owe something -- instead we are putting the money on the table already for the other to get if we are not responsive.

2. The feature above removes the trust element from the equation. We can now have relationships with people we don't trust, as the Bitcoin blockchain will serve as an automated escrow for our conditional payments and no one will be harmed. Therefore it is much easier to build networks and route payments if you don't always require trust relationships.

3. However it introduces the cost of the capital. A ton of capital must be made available in channels and locked in HTLCs so payments can be routed. This leads to potential issues like the ones described in https://twitter.com/joostjgr/status/1308414364911841281.

4. Another issue that comes with the necessity of using the Bitcoin blockchain as an arbiter is that it may cost a lot in fees -- much more than the value of the payment that is being disputed -- to enforce it on the blockchain.¹

Solutions

Because the downsides listed above are so real and problematic -- and much more so when attacks from malicious peers are taken into account --, some have argued that the Lightning Network must rely on at least some trust between peers, which partly negate the benefit.

The introduction of purely trust-backend channels is the next step in the reasoning: if we are trusting already, why not make channels that don't touch the blockchain and don't require peers to commit large amounts of capital?

The reason is, again, the ambiguity that comes from the problem of the decentralized commit. Therefore hosted channels can be good when trust is required only from one side, like in the final hops of payments, but they cannot work in the middle of routes without eroding trust relationships between peers (however they can be useful if employed as channels between two nodes ran by the same person).

The next solution is a revamped pure Ripple network, one that solves the problem of the decentralized commit in a different way.