Much of this is academic once you realize that in communication, especially effective communication (and, really, if we're rationalists concerned with "winning communication" what other sort are we concerned with?), the truth value of a statement from our internal interpretation is irrelevant, and, indeed our intention is also irrelevant unless it is indicated over some other channel during our communication, because none of these are accessible to the people we are communicating with.
Your statement may be accurate and come from honest intent when it started out, but that gives you no guarantee that it will be interpreted as such when it is received by others. The whole discourse of lies, NTLs, honesty, and radical honesty is only giving excuses that you have tried hard enough when communication fails. After that you can just claim that it was the other person(s) fault.
But does the rationalist (who is playing to win), let herself lose the game just because trying to anticipate which cooperative strategy another player could choose is difficult?
I suggest the solution isn't choosing which single person model to permit or reject, rather it's to recognize that communication involves more than one person, and that reducing it to such a simple model is causing too much error to be truly effective.
only giving excuses that you have tried hard enough when communication fails
bears repeating.
I'm sorry I took so long to post this. My computer broke a little while ago. I promise this will be relevant later.
A surgeon has to perform emergency surgery on a patient. No painkillers of any kind are available. The surgeon takes an inert saline IV and hooks it up to the patient, hoping that the illusion of extra treatment will make the patient more comfortable. The patient asks, "What's in that?" The doctor has a few options:
-The first explanation is not only true, but maximizes the patient's understanding of the world.
-The second is obviously a lie, though, in this case, it is a lie with a clear intended positive effect: if the patient thinks he's getting morphine, then, due to the placebo effect, there is a very real chance he will experience less subjective pain.
-The third is, in a sense, both true and a lie. It is technically true. However, it's somewhat arbitrary; the doctor could have easily have said "It's the weakest painkiller I have," or "It's the strongest sedative I have," or any other number of technically true but misleading statements. This statement is clearly intended to mislead the hearer into thinking it is a potent painkiller; it promotes false beliefs while not quite being a false statement. It's Not Technically Lying. It seems that it deserves most, if not almost all, the disapproval that actually lying does; the truth does not save it. Because language does not specify single, clear meanings we can often use language where the obvious meaning is false and the non-obvious true, intentionally promoting false beliefs without false statements.
Another, perhaps more practical example: the opening two sentences of this post. I have been meaning to write this for a couple weeks, and have failed mostly due to akrasia. My computer broke a few months ago. Both statements are technically true,1 but the implied "because" is not just false, but completely opposite the truth - it's complex, but if my computer had not broken, I would probably never have written this post. I've created the impression of a quasi-legitimate excuse without actually saying anything false, because our conventional use of language filled in the gaps that would have been lies.
The distinction between telling someone a falsehood with the intention of promoting false beliefs and telling them a truth with the intention of promoting false beliefs seems razor-thin. In general, you're probably not justified in deceiving someone, but if you are justified, I hardly see how one form of deception is totally OK and the other is totally wrong. If, and I stress if, your purpose is justified, it seems you should choose whichever will fulfill it more effectively. I'd imagine the balance generally favors NTL, because there are often negative consequences associated with lies, but I doubt that the balance strictly favors NTL; the above doctor hypothetical is an example where the lie seems better than the truth (absent malpractice concerns).
For what common sentiment is worth, people often see little distinction between lies and NTLs. If I used my computer excuse with a boss or professor, and she later found out my computer actually broke before the paper was even assigned, my saying, "Well, I didn't claim there was a causal connection; you made that leap yourself! I was telling the truth (technically)!" is unlikely to fix the damage to her opinion of me. From the perspective of the listener, the two are about equally wrong. Indeed, at least in my experience, some listeners view NTL as worse because you don't even think you're lying to them.
Lying does admittedly have its own special problems, though I think the big one, deception of others, is clearly shared. There is the risk of lies begetting further lies, as the truth is entangled. This may be true, but it is unclear how Not Technically Lying resolves this; if you are entirely honest, the moment your claim is questioned seriously, you either admit you were misleading someone, or you have to continue misleading them in a very clever manner. If you were actually justified in misleading them, failing to do so does not appear to be an efficient outcome. If you're able to mislead them further, then you've further separated their mind from reality, even if, had they really understood what you said, you wouldn't have. And, of course, there's the risk that you will come to believe your own lies, which is serious.
Not Technically Lying poses a few problems that lying does not. For one, if I fill in the bottom line and then fill in my premises with NTL's, omitting or rephrasing difficult facts, I can potentially create an excellent argument, an investigation of which will show all my premises are true. If I lied, this could be spotted by fact-checking and my argument largely dismissed as a result. Depending on the context (for example, if I know there are fact-checkers) either one may be more efficient at confounding the truth.
While it may be a risk that one believes their own lies, if you are generally honest, you will at least be aware when you are lying, and it will likely be highly infrequent. NTL, by contrast, may be too cheap. If I lie about something, I realize that I'm lying and I feel bad that I have to. I may change my behaviour in the future to avoid that. I may realize that it reflects poorly on me as a person. But if I don't technically lie, well, hey! I'm still an honest, upright person and I can thus justify visciously misleading people because at least I'm not technically dishonest. I can easily overvalue the technical truth if I don't worry about promoting true beliefs. Of course, this will vary by individual; if you think lying is generally pretty much OK, you're probably doomed. You'd have to have a pretty serious attachment to the truth. But if you have that attachment, NTL seems that much more dangerous.
I'm not trying to spell out a moral argument for why we should all lie; if anything, I'm spelling out an argument for why we shouldn't all Not Technically Lie. Where one is immoral, in most if not all cases, so is the other, though where one is justified, the other is likely justified as well, though perhaps not more justified. If lying is never justified because of its effect on the listener, then neither is NTL. If lying is never justified because of its effect on the speaker, well, NTL may or may not be justified; its effects on the speaker don't seem so good, either.
To tie this into AI (definitely not my field, so I'll be quite brief), it seems a true superintelligence would be unbelievably good at promoting false beliefs with true statements if it really understood the beings it was speaking to. Imagine how well a person could mislead you if they knew beforehand exactly how you would interpret any statement they made. If our concern is the effect on the listener, rather than the effect on the speaker, this is a problem to be concerned with. A Technically Honest AI could probably get away with more deception than we can imagine.
1-Admittedly this depends on your value of a "little while," but this is sufficiently subjective that I find it reasonable to call both statements true.
As a footnote, I realize that this topic has been done a lot, but I do not recall seeing this angle (or, actually, this distinction) discussed; it's always been truth vs. falsity, so hopefully this is an interesting take on a thoroughly worn subject.