It'll likely remain only marginally better than polygraphs for decades or longer, but it seems probable that eventually it's going to become so effective that law enforcement organizations will feel like they need to use it to keep up with other law enforcement organizations. Once one or two big organizations see high success rates with it, many will feel like they're letting a lot of criminals get off scot free because they're behind on technology, and most departments/agencies will have very strong incentives to jump on board as they start to see more and more of their peers adopting it (just like with DNA collection and testing).

Even if consent is required, it'll just be like traffic stop no-win situations: if they ask you to consent to a search due to suspicion of drug possession and you say yes, they'll search, and if you say no, they'll wait until a drug dog comes, and then the dog will effectively conduct the search and/or will be used as a tool to permit a full search. Or if you refuse to take a breathalyzer test, they essentially presume guilt and treat you like you took the test and failed. Anyone who refuses to consent to the Truth-O-Matic test will probably suffer similar consequences. It's mostly an illusion of consent. I think this future is inevitable without some very serious legislation; perhaps a constitutional amendment.

A counter-point is that one could imagine a potential distant future exigent circumstance where some sophisticated criminal might have information that could truly prevent an imminent attack that'll very likely kill millions of people, or something, and which could be discovered through such a detector. In those fanciful scenarios I honestly do find myself wanting to agree with Comey: fuck privacy, siphon those fucking thoughts. It's certainly a much more ethical and effective alternative to torture, at least, which is how the US government would currently handle such a scenario (if they could cover it up).

Security and privacy/liberty are always a tradeoff, but even if you weigh privacy exponentially higher, there's always still some theoretical risk that would tip the scale back towards security, in my view. Nuclear armageddon, for example. The privacy hill to die on shouldn't be Mega-Corpse Mountain.

This scenario is absurdly far-fetched in 2019, of course, but in 2079? 2119? Who knows? If Aum Shinrikyo and ISIS can do what they did and plan what they planned in the 1990s and 2010s, what about new incarnations of zealous death cults in a world that may have extremely intelligent AIs and/or DIY-potential for almost anything?

Even a single instance of losing the final bastion of privacy is such an incredibly slippery slope that it's hard to imagine any safe way to allow its use in very extreme situations like these without setting up the conditions for horrific abuse and a Stasi/1984-esque kowtowing of all of society. But I think this aspect of the law enforcement perspective needs to be taken into consideration as well, even while simultaneously acknowledging that law enforcement will inevitably cry wolf and exaggerate the likelihood or impact of imminent threats. Of course this'll happen, but what happens if they do see an actual wolf one day?