No. The problem is that pretty much everyone trusts them, at this point in time. That was Peter's point.
Sure, there are researchers and engineers who rightfully don't trust CAs. But we don't really matter. The users, the consumers, the parents, the grandparents, the activists do.
Sure, there's a lot of research being done to find the Next Great Thing (tm), but how about a short/mid-term emphasis on shoring up the glaring problems in the existing technologies first? Tighten the number of default CAs, shore up bad SSL and TLS code, tighten default settings in client software.
Things like Chrome popping up warnings about self-signed, expired, or invalid certs may have been a great start, but nobody's really tidying up much on the server end, so the end effect is that the users blindly click through the Chrome warnings.
TL;DR: The Next Big Thing (tm) is going to be great, I'm sure, but how about fixing/tightening existing configurations in the mean time?
You and I might also agree: browsers make it too easy to click through the bad-cert warnings. It used to be a trendy thing to argue on HN that these warnings were entirely pointless and should be done away with, which, of course, would have done grievous harm to security above the harm already done by the click- click-click- you're- done UX browsers have already established here.
In fact I think that so many people trust CAs that if someone provides a more secure alternative it should look like an evolution of CAs so it doesn't piss off people who have been trusting CAs all this time.
Generally almost everyone has experience with malware, bugs and broken software. No one who has used a computer for more than 10 minutes absolutely trusts computers in any capacity including security.
Ask some random people on the street if they trust a mainland Chinese entity with ensuring the security of their communication with their banks, however, and I think you'll get a different answer.
Hundreds of millions of consumers use devices that implicitly trust CAs today.
You're usually spot on, but in this case you're dead wrong. Most humans that use the internet use devices that trust CAs absolutely - which is exactly why they're being subverted for government interception.
The distinction between these two vantage points isn't particularly relevant to my point; at least, I don't think it is.
If we want real security we have to build it separate from the CA model, but as you point out CAs do provide some security against government inception, just as a locked door does provide some security against police wondering into your house, and we should not abandon it wholesale.
With TACK you depend on CAs to establish the initial connection, set up a "pin", and then no longer rely on the CA for future connections. That initial connection is fungible by a MITM, so it's not secure. We need a term for connections that are "probably" secure, but for which there is no complete assurance of security, which is what any connection based around a CA really is.
This is why the postcard method is superior. Unless someone spends the time to intercept every anonymous postcard with a secret code on it, rewrite the message with perfect imitation handwriting, and send it on to the unknown P.O. box, without more than a couple days delay, it's next to impossible to circumvent this initial key exchange. For those that need real privacy I would recommend this method; for those that just want to order Jolt Cola on an open wifi connection, CAs are good enough.
Meanwhile, the world in which most of the mainstream browsers support TACK is imperfect, but immediately better than what we have now. TACK also sets us up to continue decoupling ourselves from the CA system.
Pre-loaded public key pinning shipped by modern browsers is a better way forward IMHO. If you're trusting your browser enough to run their software, you might as well get your public keys from them, too. IMO, TACK works about as well as self-signed certs with the 'remember this certificate' option - just don't use a Starbucks connection the first time you browse the site.
This really gets to the crux of it. We might deem the CA's untrustworthy but we currently trust them anyway because there isn't really that much choice.
[1] http://www.cl.cam.ac.uk/~rja14/tcpa-faq.html - Ross Anderson (Trusted Computing FAQ - Q24)
http://tack.io/draft.html
The problem with simply abandoning CAs is that it creates a situation in which it's even easier for government sponsored agencies to mass-intercept traffic, at least for a window of time (probably several years), and all that window buys us is the ability for sites that don't really care about security to save $10-$100 dollars on cert costs.