Hawking thought - and could "rigorously" deduce from semiclassical gravity - that the information has to be lost because it can't get out of the black hole interior once it gets there.
To see why, look at this "Penrose causal diagram" that may be derived for a black hole solution. Diagonal lines at 45 degrees are trajectories of light, more vertical lines are time-like (trajectories of massive objects), more horizontal lines are space-like.
If you look e.g. at the yellow surface of the star (its world line), you see that it ultimately penetrates through the green event horizon into the purple black hole interior. Once the object - and the information it carries - is inside, it can no longer escape outside (don't forget: time is going up), into the light green region, because it would have to move along spacelike trajectories.
So the information carried by the star that collapsed to the black hole inevitably ends at the violet horizontal singularity and it may never be seen in a completely different region outside the black hole.
The information loss of course depends on the detailed geometry of the black hole that is not shared by a helium nucleus, so it shouldn't be surprised that helium nuclei and black holes have different properties.
As we know today, the information is allowed to "tunnel" along spacelike trajectories a little bit in quantum gravity (i.e. string/M-theory). This process is weak but this weak "non-local process" allows the information to be preserved.
Here are my two cents.
There is a community in which Hawking's solution was ignored, and the only accepted one was the black hole complementarity of Susskind, Thorlacius, and Uglum. The firewall discussion takes place within that world.
Susskind claims in his book The Black Hole War: My Battle with Stephen Hawking to Make the World Safe for Quantum Mechanics, and others follow him, that he defeated Hawking, who, in 2004, conceded the bet against Preskill. In fact, Hawking was probably not convinced by Susskind's proposal, but by Maldacena's AdS/CFT correspondence. But AdS/CFT doesn't give the explanation how the information is recovered, and Hawking proposes his own solution, not being based on stretched horizon and black hole complementarity. In fact he still believes that, if we consider only a history, for General Relativity + Quantum Physics the problem persists, and it is resolved only when summing over all topologies.
It seems like the AMPS paper considers only the black hole complementarity. They don't consider other proposals, such as Hawking's. So, with respect to that framework, AMPS find a problem with the black hole complementarity, namely that it is not enough, and a firewall should be added. Susskind considered this idea earlier in his book An Introduction To Black Holes, Information And The String Theory Revolution, page 84, when he named it "brick wall". The "paradox" is that the firewall seems to be required by unitarity, but the existence of such a firewall contradicts the principle of equivalence.
So, in my opinion, yes, Hawking's solution doesn't need a firewall, and black hole complementarity needs it. And if we accept the firewall, the black hole complementarity is no longer needed. So Hawking should write a book about his (non-action) war with Susskind.
Update.
In my answer I argued that the firewall discussion takes place in a circle in which Hawking's solution is not acknowledged. Following a comment, let me get closer to the question about why Hawking's solution was not accepted. I don't know of any decisive argument against Hawking's solution.
My main reason why I find his argument insufficient is that it doesn't really solve the problem, unless you sum over different topologies, and the used measure allows the solutions that violate unitarity to cancel each other. It is again a personal opinion.
I think that the reason why his solution was not accepted like Susskind's is because it relies on a less popular approach to quantum gravity. The Susskind, Thorlacius, and Uglum (STU) argument was presented in a form which make it look as it only relies on three principles accepted by everyone:
- Information conservation
- No cloning theorem
- Equivalence principle,
so it doesn't seem confined to a particular quantum gravity approach.
Also, it seemed to solve the problem for each spacetime, and not only in a sum over topologies.
Another reason may be that, at the time when Hawking proposed his solution, the black hole complementarity was considered for over a decade to be the good solution by a dominating community. It stimulated research in superstring theory of black holes, and other approaches to quantum gravity tried to explain the information from the stretched horizon.
It is also possible that this is a historical accident, and if Hawking had proposed his solution before Susskind, it would have been accepted his, and not Sussikind's. I actually think that Susskind's would have been rejected long time before the AMPS argument, if there was an alternative to save unitarity. Probably the arguments would have been
- Susskind, Thorlacius, and Uglum (STU) claim to rely on the no-cloning theorem, but actually it admits cloning, only that it claims that there is no observer who will see both copies.
- STU claim to rely on the equivalence principle, but let's consider instead of the event horizon, a Rindler horizon. Say Bob is moving with acceleration, and sees Alice going through the Rindler horizon. Bob sees her destroyed, and she sees nothing. But now, unlike the case of the Schwarzschild event horizon, Bob can go back and check Alice, and find her well. So, this won't work for Rindler horizon, so the equivalence principle is in fact violated by BH complementarity.
- BH complementarity claims it is OK to admit contradiction, so long as the contradiction is not observed directly. I don't really think that, if the contradiction is seen only in theory, and never in experiment, it is OK.
- It has argued that if Alice sends a signal right after passing through the horizon, Bob may dive too, and receive it after he enters in the horizon, so the two viewpoints can be compared. Susskind claims that he can't, because he will reach the singularity before receiving the message, and indeed there is a proof for this. But, this works only for Schwarzschild black holes. If the black hole is rotating or charged, then the singularity is timelike, and can be avoided for indefinite long time. So Alice and Bob really can meet and compare the two copies, violating the very principle STU claims to save.
Maybe this made Robert Wald in his recent talk at the Fuzzorfire workshop, to state that the proposed cures (including complementarity) are worse than the disease:
I find it ironic that some of the same people who consider “pure ->
mixed” to be a violation of quantum theory then endorse truly drastic
alternatives that really are violations of quantum (field) theory in a
regime where it should be valid.
So, I agree with the question, that the AMPS argument and firewall discussions are just the realization that the Hawking's paradox was not solved by BH complementarity.
Let me mention another possibility, more recent and less known. Most solutions concentrate on the event horizon, and what happens there. While this important, let's not forget that the information appears to be lost not on the horizon, but at the singularity.
There is an analytic extension of the Schwarzschild solution through the singularity. This replaces the usual Penrose diagram (fig. A) with another one (fig. B), which is globally hyperbolic, and might allow information to be recovered.
I will stop here, because it becomes self-advertising. There are more questions, but I will not detail here. Some of them answered in the papers here (where there is also my email address). A less technical paper is here. Also I plan to write more about this soon on my blog.
Best Answer
Short answer: Force is generally a bad way to think about things in relativity. It can occasionally be useful, but this is most commonly the case when you are closely approximating Newtonian cases or you're already well within the relativistic framework.
In general (and special) relativity, all curves are of one of three types: timelike, spacelike or null. For timelike curves, the magnitude of the $\dot t$ component of the 4-velocity exceeds the spatial velocities. For spacelike curves, the opposite is true. Null curves are the boundary between the two. In special relativity, material particles follow timelike curves, and light follows null curves. Spacelike curves represent spatial seperations as measured in a particular coordinate frame.
It turns out that, in the case of a black hole, for all points inside the horizon, there are no timelike or null curves that point into the future and that leave the horizon. All future-pointing timelike and null curves are therefore trapped. Leaving the inside of the horizon would require that they cease being timelike/null, which would be contrary to their character in the theory.