The future of digital guitar amplifiers
The one thing that digital modellers still can't do as well as tube amps, IMO, is to produce great touch sensitivity in the critical clean-to-slightly dirty range. The Vox Tonelab does well in this area, but technically it's not using digital modelling to create the amp response. Line6 does well in the pristine clean (e.g. JC120) and dirty to stupid-amounts-of-gain sounds, but falls apart in the middle ground between clean and slightly dirty. I don't really care if it sounds like a cranked Bassman or whatever; the real issue is the response as a guitar amp - of any brand or style - and the modellers just can't hack what any simple tube circuit does naturally.
Don't get me wrong: modellers are great tools for getting distorted cranked-amp tones at low volumes. And it's tough to argue with all the built-in effects and patching options that come along for the ride, if you happen to like that kind of thing. I'd much rather have a modelling amp than a rack system having the same capabilities, not only for the size and weight savings but also for the considerable cost reduction.
There's one point that I will argue: I don't believe that digital models will ever become indistinguishable (in terms of quality, especially in that critical range that I've already mentioned) from tube amps. Digital modellers will continue to become less expensive and have more features. They may even kill the market for tube amps at some point (as solid-state almost did in the `70s). But the people programming these modellers have been making (and, for reasons of cost, will continue to make) simplifying assumptions about the general behavior of a tube amp, which is far more complex than they'll admit.
Frankly, I don't believe that programmers have the chops to accurately model a tube amp. That's not a dig at the programmers; they could be the best in the world at programming a computer and still not have the chops to create an accurate tube amp model in software. Programmers are doing amazing things with computers... things that I certainly couldn't have envisioned forty years ago. (That's about when I started getting interested in computers - my aunt worked for Cmdr. Grace Hopper programming the ENIAC to produce ballistics tables for the Navy. And all but a few years of my working life have been spent designing and programming computers.) However, skill in programming does not convey or imply knowledge or skill in any other domain. For the most part (there are no doubt exceptions) the programmers who are creating digital modelling amps are from the generation that grew up with computers, not tube amps. How many of them have a notion of a tube amp as something other than a few transfer functions (one for distortion, one for tone, one for FX)? Are these guys gonna grok the effect of the OT, shifting bias points, power supply sag, etc? It's a lot easier (and far more economical) to make simplifying assumptions, and that's exactly what they're doing.
There's nothing inherently wrong with a simplified model: it is what it is. It's not an accurate model, but a model. The guys who have a deep understanding of tubes (down to the physics level, not just as a copy/paste design exercise) are the ones who could really shed some insight on the design of good modellers, but they're long since retired (or even departed from this world).
I take issue more with the "modelling" label per se than with using digital techniques to generate or modify musical sounds. I have no doubt that someone will do something in the digital arena that will become a touchstone for future programmers, much as the early Fender amplifier designs have become a touchstone for modern-day tube amp builders.
But modelling...? Modelling something using an algorithm implies that you have an understanding of the system you're trying to model. The deeper the understanding, the better chance you have to create an accurate model. The point I've been trying to get across is not that programmers don't have the chops to do really interesting things, nor that they won't be able to nail a good tone using digital techniques. Rather, my point is that the intersection of the small set of people in the world who command a deep understanding of tube technology, and the set of people who are programming modelling amps is probably an empty set. That's not to say that the programmers will be incapable of doing outstanding, even groundbreaking, work - just that they will probably never have the intellectual tools to produce a truly accurate, nuanced model of a good tube amp.
All of the textbooks (and I mean professional engineering textbooks, not technical applications books like the RDH) that describe why tubes and tube circuits operate the way they do were published fifty to sixty years ago, are not well-known, and are somewhat difficult to find today even if you know what to look for. Just to read and understand these texts requires a grasp of physics, math, electronics, materials, etc. well beyond what the typical CS graduate is exposed to nowadays. (And when you get right down to it, how many working programmers are actually CS graduates as opposed to someone who's had vocational or on-the-job training combined with ongoing self-education?) So it's a pretty tall order for the would-be modelling programmer to try to construct an accurate model from first principles. It's much more productive to make a lot of simplifying assumptions and measure the transfer characteristics of some sample amps in order to program a model that captures perhaps 80% of a real amp's behaviors. And for a lot of players, that 80% is good enough. If it wasn't, the modelling amps wouldn't command the market share that they do today.
I look at modelling this way: In another 50 years, virtually no one is going to care about the "real" sound of a `59 Bassman or a `64 Deluxe. Guitar players (if there are any) will have a vast palette of synthesized tones to choose from. A few of those tones may be reminiscent of the sound of century-old amplifiers. The rest will be sounds that none of us could have forseen.