Derek Parfit wrote up some thoughts along these lines in 1984:
I shall first distinguish threats from warnings. When I say that I shall do X unless you do Y, call this a warning if my doing X would be worse for you but not for me, and a threat if my doing X would be worse for both of us. Call me a threat‐fulfiller if I would always fulfil my threats.
Suppose that, apart from being a threat‐fulfiller, someone is never self‐denying. Such a person would fulfil his threats even though he knows that this would be worse for him. But he would not make threats if he believed that doing so would be worse for him. This is because, apart from being a threat‐fulfiller, this person is never self‐denying. He never does what he believes will be worse for him, except when he is fulfilling some threat. This exception does not cover making threats.
Suppose that we are all both transparent and never self‐denying. If this was true, it would be better for me if I made myself a threat‐fulfiller, and then announced to everyone else this change in my dispositions. Since I am transparent, everyone would believe my threats. And believed threats have many uses. Some of my threats could be defensive, intended to protect me from aggression by others. I might confine myself to defensive threats. But it would be tempting to use my known disposition in other ways. Suppose that the benefits of some co‐operation are shared between us. And suppose that, without my co‐operation, there would be no further benefits. I might say that, unless I get the largest share, I shall not co‐operate. If others know me to be a threat‐fulfiller, and they are never self‐denying, they will give me the largest share. Failure to do so would be worse for them.
Other threat‐fulfillers might act in worse ways. They could reduce us to slavery. They could threaten that, unless we become their slaves, they will bring about our mutual destruction. We would know that these people would fulfil their threats. We would therefore know that we can avoid destruction only by becoming their slaves.
The answer to threat‐fulfillers, if we are all transparent, is to become a threat‐ignorer. Such a person always ignores threats, even when he knows that doing so will be worse for him. A threat‐fulfiller would not threaten a transparent threat‐ignorer. He would know that, if he did, his threat would be ignored, and he would fulfil this threat, which would be worse for him.
If we were all both transparent and never self‐denying, what changes in our dispositions would be better for each of us? I answer this question in Appendix A, since parts of the answer are not relevant to the question I am now discussing. What is relevant is this. If we were all transparent, it would probably be better for each of us if he became a trustworthy threat‐ignorer. These two changes would involve certain risks; but these would be heavily outweighed by the probable benefits. What would be the benefits from becoming trustworthy? That we would not be excluded from those mutually advantageous agreements that require self‐denial. What would be the benefits from becoming threat‐ignorers? That we would avoid becoming the slaves of threat‐fulfillers.
We can next assume that we could not become trustworthy threat‐ignorers unless we changed our beliefs about rationality. Those who are trustworthy keep their promises even when they know that this will be worse for them. We can assume that we could not become disposed to act in this way unless we believed that it is rational to keep such promises. And we can assume that, unless we were known to have this belief, others would not trust us to keep such promises. On these assumptions, S tells us to make ourselves have this belief. Similar remarks apply to becoming threat‐ignorers. We can assume that we could not become threat‐ignorers unless we believed that it is always rational to ignore threats. And we can assume that, unless we have this belief, others would not be convinced that we are threat‐ignorers. On these assumptions, S tells us to make ourselves have this belief. These conclusions can be combined. S tells us to make ourselves believe that it is always irrational to do what we believe will be worse for us, except when we are keeping promises or ignoring threats.
Does this fact support these beliefs? According to S, it would be rational for each of us to make himself believe that it is rational to ignore threats, even when he knows that this will be worse for him. Does this show this belief to be correct? Does it show that it is rational ignore such threats?
It will help to have an example. Consider
My Slavery. You and I share a desert island. We are both transparent, and never self‐denying. You now bring about one change in your dispositions, becoming a threat‐fulfiller. And you have a bomb that could blow the island up. By regularly threatening to explode this bomb, you force me to toil on your behalf. The only limit on your power is that you must leave my life worth living. If my life became worse than that, it would cease to be better for me to give in to your threats.
How can I end my slavery? It would be no good killing you, since your bomb will automatically explode unless you regularly dial some secret number. But suppose that I could make myself transparently a threat‐ignorer. Foolishly, you have not threatened that you would ignore this change in my dispositions. So this change would end my slavery.
Would it be rational for me to make this change? There is the risk that you might make some new threat. But since doing so would be clearly worse for you, this risk would be small. And, by taking this small risk, I would almost certainly gain a very great benefit. I would almost certainly end my slavery. Given the wretchedness of my slavery, it would be rational for me, according to S, to cause myself to become a threat‐ignorer. And, given our other assumptions, it would be rational for me to cause myself to believe that it is always rational to ignore threats. Though I cannot be wholly certain that this will be better for me, the great and nearly certain benefit would outweigh the small risk. (In the same way, it would never be wholly certain that it would be better for someone if he became trustworthy. Here too, all that could be true is that the probable benefits outweigh the risks.)
Assume that I have now made these changes. I have become transparently a threat‐ignorer, and have made myself believe that it is always rational to ignore threats. According to S, it was rational for me to cause myself to have this belief. Does this show this belief to be correct?
Let us continue the story.
How I End My Slavery. We both have bad luck. For a moment, you forget that I have become a threat‐ignorer. To gain some trivial end—such as the coconut that I have just picked—you repeat your standard threat. You say, that, unless I give you the coconut, you will blow us both to pieces. I know that, if I refuse, this will certainly be worse for me. I know that you are reliably a threat‐fulfiller, who will carry out your threats even when you know that this will be worse for you. But, like you, I do not now believe in the pure Self‐interest Theory. I now believe that it is rational to ignore threats, even when I know that this will be worse for me. I act on my belief. As I foresaw, you blow us both up.
Is my act rational? It is not. As before, we might concede that, since I am acting on a belief that it was rational for me to acquire, I am not irrational. More precisely, I am rationally irrational. But what I am doing is not rational. It is irrational to ignore some threat when I know that, if I do, this will be disastrous for me and better for no one. S told me here that it was rational to make myself believe that it is rational to ignore threats, even when I know that this will be worse for me. But this does not show this belief to be correct. It does not show that, in such a case, it is rational to ignore threats.
We can draw a wider conclusion. This case shows that we should reject
(G2) If it is rational for someone to make himself believe that it is rational for him to act in some way, it is rational for him to act in this way.
Return now to B, the belief that it is rational to keep our promises even when we know that this will be worse for us. On the assumptions made above, S implies that it is rational for us to make ourselves believe B. Some people claim that this fact supports B, showing that it is rational to keep such promises. But this claim seems to assume (G2), which we have just rejected.
There is another objection to what these people claim. Even though S tells us to try to believe B, S implies that B is false. So, if B is true, S must be false. Since these people believe B, they should believe that S is false. Their claim would then assume
(G3) If some false theory about rationality tells us to make ourselves have a particular belief, this shows this belief to be true.
But we should obviously reject (G3). If some false theory told us to make ourselves believe that the Earth was flat, this would not show this to be so.
S told us to try to believe that it is rational to ignore threats, even when we know that this will be worse for us. As my example shows, this does not support this belief. We should therefore make the same claim about keeping promises. There may be other grounds for believing that it is rational to keep our promises, even when we know that doing so will be worse for us. But this would not be shown to be rational by the fact that the Self‐interest Theory itself told us to make ourselves believe that it was rational. It has been argued that, by appealing to such facts, we can solve an ancient problem_ we can show that, when it conflicts with self‐interest, morality provides the stronger reasons for acting. This argument fails. The most that it might show is something less. In a world where we are all transparent—unable to deceive each other—it might be rational to deceive ourselves about rationality.
Derek Parfit wrote up some thoughts along these lines in 1984: