Linkpost: SBF sentenced to 25 years jail

Please note that my previous post took the following positions:

1. That SBF did terrible acts that harmed people.

2. That it was necessary that he be punished. To the extent that it wasn't implied by the previous comment, I clarify that what he did was illegal (EDIT: which would involve a finding of culpable mental states that would imply that his wrongdoing was no innocent or negligent mistake).

3. The post doesn't even take a position as to whether the 25 years is an appropriate sentence.

All of the preceding is consistent with the proposition that he also acted with the intention of doing what he could to better the world. Like others have commented, his punishment is necessary for general deterrence purposes. However, his genuine altruistic motivations make the fact that he must be punished tragic.

I think a conclusion that he acted from mixed motives is better supported by the evidence.

I disagree, but it obviously depends what exactly we're discussing. 

Was his judgment for not coming clean when things were only starting to get bad compromised by not wanting to lose his influence, money, and reputation? Probably!

However, do I think he made some of his most consequential decisions to a significant degree because he thought he could get nice things for himself that way? I actually don't think so!

Making big decisions for reasons other than impact would ruin his score in the game. It looks to me like his primary drive was optimizing his life for impact like a video game, trying to score the most points. Making a big life decision for self-oriented reasons would be forfeiting points he could otherwise have gotten, which would feel deeply unsatisfying for someone who's always thinking about how to get the highest score. 

Also, $16 million is peanuts for someone who's personally worth $2-10 billion, which SBF probably ""somewhat reasonably"" thought he was worth when things were going well (or when he optimistically thought he could make them well again soon enough). It's like what a couple hundred bucks is to someone with a net worth of $80,000 – not that significant! Imagine you're making decisions for tens of millions every day. I can't emphasize enough how unimportant it then becomes whether you pay 1 million or 16 million for your parents' house (which you may or may not also plan to use for yourself or some of your coworkers/flatmates). Just like with offices or the building where he planned to sleep in, the price isn't the relevant variable (what's more relevant is, "Does it take up your time or people's time with high opportunity costs? Is it close to the office where you work? Is it convenient?" Etc.) 

I feel like, 

• if someone displays as much single-minded obsessiveness as SBF, 
• and his conversations with the people around him tell you he's hyperfocused on impact (he wasn't just obsessing about how to grow FTX for money, he was also constantly discussing ideas around grantmaking strategies or paying Trump to not run again for election),
• and his close social circle is reinforcing the impact-orientedness,
• and there's a credible ideology behind it which generated many other examples of similarly impact-obsessed people (hopefully the vast majority of them without the recklessness and "no fear of bad consequences" part),

then I'd say the evidence points to being primarily impact-driven pretty much as strongly as it gets. Of course, no one has the motivations of a perfect utilitarian robot, but I think SBF was trying, so it doesn't seem fair to call it "acted from mixed motives."

Edit to add: We could discuss whether he was "unusually biased" for someone who's trying to have a ton of impact – hopefully yes (otherwise EAs don't stand a chance of getting impact right), but that's still meaningfully different from "acting from mixed motives." (Also, I'd say the most severe bias was probably a blindspot around risk taking and risks of things going badly, rather than e.g., being too much into a rich lifestyle or being too much into fame or whatever. FWIW, my best guess is that he also was too much into wanting power/influence; I'm just saying that it doesn't seem like the primary reason things went so poorly.) 

Edit2: Somewhat relevant comment I made a year ago about why I think SBF having dark triad traits is compatible with the above picture. 

SBF likely had mixed motives, in that there was likely at least some degree to which he acted in order to further his own well-being or with partiality toward the well-being of certain entities (such as his parents). The reasoning that you mentioned above (privileging your own interests instrumentally rather than terminally such that you as an agent can perform better) is a fraught manner of thinking with extremely high risk for motivated reasoning. However, I think that it is one that serious altruists need to engage with in good faith. To not do so would imply giving until one's welfare was at the global poverty line, which would probably impair one too much as an agent. Of course, I'm not saying he was engaged in good faith regarding this instrumental privileging argument, but I cannot preclude the possibility.

Regardless, I have been persuaded by everything that I have seen that a significant part of SBF's motivations were to help advance a world of higher well-being. Of course, from a deontological perspective he did wrong by his dishonest and fraudulent actions. From a consequentialist perspective, the downside risks had such incalculable costs that it was terrible as well.

But the sincere desire of his to make the world a better place makes me sympathetic of him in a way that I probably would not be with similarly sentenced other convicts. Given a deterministic or random world, I understand that all convicts are victims too. But I cannot help but feel more for one who was led to their crime by a sincere desire to better the world, than say, to kill their spouse in a fit of rage, or advance themselves financially without any such altruistic motivation.

To clarify, you would sacrifice consistency to achieve a more just result in an individual case, right?

But if there could be consistently applied, just, results, this would be the ideal result...

I don't understand the disagree votes if I am understanding correctly.

Totally fair! I think part of my reasoning here relies on the difference between "sentence I think is longer than necessary for the purposes of sentencing" (which I would not necessarily classify as an "evil" in the common English usage of that term) and an "unhinged" result. I would not support a consistent sentence if it were unhinged (or even a close call), and I would generally split the difference in various proportions if a sentence fell in places between those two points.

It's a little hard to define the bounds of "unhinged," but I think it might be vaguely like "no reasonable person could consider this sentence to have not been unjustly harsh." Here, even apart from the frame of reference of US sentencing norms, I cannot say that any reasonable person would find throwing the book at SBF here to have been unjustly harsh in light of the extreme harm and culpability.

As far as what they predict, 40-50 years as explained in the government's sentencing memo.

As far as what the impact should be -- I would have to write a book on that. To start with, I see multiple, related harm-related measures here:

• The amount of expected harm the offender knew or should have known about (this is the culpability-flavored measure);
• The actual expected value of harm (this is more of a general deterrence-flavored measure); and
• The actual harm (more of a retribution-flavored measure).

I also don't see a unified measure of harm in economic-loss cases, as the harm associated with stealing $1,000 from a working class person is substantially higher than the harm of stealing it from me. Targeting vulnerable victims also gets you an enhancement for other reasons (e.g., it suggests a more extensive lack of moral compass that makes me value incapacitation more as a sentencing goal).

But most fundamentally, both harm and culpability go into the mix, filtered through the standard purposes of sentencing, to produce a sentence that I think is sufficient, but not greater than necessary, to accomplish those goals.

So I can tell you that the relationship between harm and sentence in fraud cases shouldn't be -0-, both because there is little or no general deterrence against making your frauds bigger, and because there is some relationship between culpability and fraud size. It also shouldn't be linear, both because this is impractical given the wide variance in harms, and because the degree of culpability does not ordinarily vary in a linear manner. 

Most people in the sentencing realm think the federal sentencing guidelines increase the sentence too much based on loss amount (~ 25% uplift for each doubling of amount, plus some other uplifts tend to scale with size) and give too much weight to loss size in general. I agree with both of those views. Roughly and after considering a fuller measure of harm than aggregate financial loss, I might consider such an algorithm appropriate for weighing the harm factor in arriving at a sentence rather than for uplifting the sentence itself. And maybe I would increase the average sentence about 10% for each doubling, if all other factors were equal (low confidence)? But usually non-harm culpability factors increase with harm size, so the average increase in practice would probably be higher (low confidence).

As a bonus exercise: The differential impact between harm and non-harm culpability factors seems to vary quite a bit based on offense type. In drunk driving cases, actual harm seems to explain the bulk of the variance? No harm, first offense -- 2 days. Kill someone, no harm -- perhaps 10 years. The non-harm culpability factors and even the expected-harm measures aren't really that different in a lot of cases. Even here, the extra punishment for multiplied harm isn't linear; you might get a 50% uplift for the second fatality, and a total of 85% more for three fatalities (low confidence)? 

On the other hand, if you rob five convenience stores on five days before getting caught, you might get twice the sentence for robbing one (low confidence) -- even though the culpability scales here much more obviously with the number of offenses / amount of harm. One theory here is that getting caught and punished gives you the opportunity to learn your lesson; if you rob again after going through that then the argument for long-term incapacitation is stronger, etc.

For some purposes of sentencing, social context affects what is reasonably "necessary" and what I would count as "harm." Society is the other participant in the sentencing exercise.

Suppose A thinks from first principles that five years was enough for SBF. In the context of the American system, that sentence would convey something about offense severity -- and it would drastically understate it where a street dealer in illegal drugs might get the same sentence. So because of the background severity of sanctioning, the five-year sentence is inadequate to convey the appropriate degree of societal condemnation. If you think the social educative function of the criminal process is important, that's a problem. A sentence that tells everyone in the US that what SBF did isn't really that big a deal on the grand scheme of criminal activity would be quite harmful.

The social meaning also matters to the victims. Seeing the person who harmed you receive a penalty in line with their crimes mitigates the suffering of many victims; seeing them get off easy imposes additional suffering on others. And those victims are experiencing the sentence through the lens of what other offenders in their society receive.

But the main harm I had in mind is the harm to the norm that all are equal before the law and will be treated equally. Now, if I hand out below-norm sentences to all offenders, maybe that particular harm isn't present. But those kinds of breaks are not equally distributed; we know that the system is harder on people of color, poor people, and other disadvantaged groups. Giving SBF too much of a break from what others with comparable culpability and harm receive gives the perception that the privileged are more favored before the law than the commoners. Or maybe the reality -- some people have argued that the court should consider SBF's loss of reputation and loss of massive earning capacity, which sounds a whole lot to me like taking his privilege into account. Avoiding that harm is more important to me than whether SBF spends a few extra years in prison.

(I should clarify that I think a sentence of more than 25 years was necessary and appropriate here on/near first principles, mainly for reasons described in the quick take linked to my comment. I should note that this case hits a lot of the aggravating factors I tend to weigh heavily, and I see no significant mitigating factors here at all.)