supposedlyfun

Karma: 1,257

supposedlyfun 18 Sep 2020 2:04 UTC
50 points
on: Book Review: Working With Contracts
I’m a lawyer and have programmed very simple things in very simple programming languages, and I endorse both your Contracts 101 description, which is very easy to understand and also accurate, and your parallels with computer programming. Your focus on risk allocation is spot on. I promise to do something for you, you promise to pay me, and the other 50 pages are just risk allocation and rules on how to parse the risk allocation.
Re patches on top of patches: There is some truth here. For context, there are two reasons that contracts I work on end up looking like this. The first is that the patches are carve-outs for some very particular contingency the client is worried about, and because they’re not lawyers, they don’t understand the non-patchy things that address their concern, so they want something specific addressing their concern. The other side tends to agree because they, too, know that the patchy thing is superfluous. It’s sort of like if you were a programmer who was really bad at math, and you wanted to print “hello world” if x was between 5 and 10 but it was EXTREMELY PERSONALLY IMPORTANT to you that “hello world” be printed where x=7, you would code
{if(5<x<10), PRINT “hello world”
if(x=7), PRINT “hello world”}
The second reason is cost. I could write a bespoke contract for every client’s every need, but to save time==money, there are vast form books/online repositories for standard transactions, plus we ruthlessly plagiarize from each other—it really is the highest form of flattery. The patches are added to address the specific concerns or unique details of the particular deal. We then run a sanity check applying the patches to the rest of the agreement to make sure nothing weird happens.
Finally, I would characterize a “reasonableness” requirement as “we’ll just ask a jury” rather than a “do what I mean” button. Because taking a case to trial is so flipping expensive, and putting in a reasonableness standard or “course of business” all but guarantees (due to common law) that you will have to go to trial if there’s an unresolvable dispute, and juries are super unpredictable, you would almost never use “reasonableness” as a substitute for “do what I mean.” You would keep hammering at a verbalization of “what I mean”.
If the other side were insisting on a “reasonableness” standard for some action which we could define by reference to objective facts (like “200 widgets made out of ASTM A125-96 steel”), my mental model of them would update to “so likely to be planning to cheat my client that we should walk the deal”.
The other use of “reasonable” you mention is more common—to give the party some freedom but not absolute discretion.
Thanks again for this excellent post. I feel hopeful when I see non-lawyers understanding law stuff, because it makes me feel hopeless for the future of the Rule of Law (TM) when some part of my job is so complex that no person without specialized training could hope to do it. The law should be for everyone.
What links here?
- Rant on Problem Factorization for Alignment by johnswentworth (5 Aug 2022 19:23 UTC; 90 points)

supposedlyfun 19 Sep 2020 20:57 UTC
19 points
in reply to: johnswentworth’s comment on: Book Review: Working With Contracts
In most US jurisdictions, interpretation of a contract is a question of law (read: decided by the judge). However, some subparts of that inquiry are questions of fact (read: decided by the jury). By far the most common is to argue that some key term of the contract is ambiguous—that if we read the document as a whole, that key term has two mutually inconsistent meanings.

This opens the door to evidence about the parties’ intent, which is a jury question. We lawyers are nothing if not sensitive to language, so we’re very good at coming up with possible ambiguities. A judge confronted with a bona fide ambiguity, or even a borderline one, ought to let the case go to trial for the jury to resolve the intent question. (The “ought” comes from common law about when summary judgment is inappropriate.)

Another good one is in contracts for specialized goods and services with esoteric terms and requirements. You identify some inadequately defined word no jury is familiar with and hire an expert witness to opine that the word’s meaning in the particular commercial context is x. Your opponent hires an expert to say it’s y. Again, the appellate courts tell the trial courts that in this situation, a jury should resolve the “battle of the experts.” P.S. you can find an expert to say just about anything for $5,000, $10,000 on the high side.

As you can imagine, it’s much easier to litigate the interpretation of a contract than to draft one that can’t really be litigated. Transactional attorneys tend to give litigators a hard time about this, and litigators give transactional attorneys a hard time about being terrified of court rooms.

supposedlyfun 25 Sep 2020 18:43 UTC
3 points
on: Human Biases that Obscure AI Progress
This helped me to distill a lot of things that I already “knew” on a semi-verbal/intuitive level but had not put into words. I expect it will help me discuss these concepts with people unfamiliar with the ins and outs of AI. Thank you for doing the mental work of synthesizing and then translating it into plain English with only the necessary jargon.

supposedlyfun 26 Sep 2020 20:56 UTC
10 points
on: supposedlyfun’s Shortform
I’m grateful for MIRI etc and their work on what is probably as world-endy as nuclear war was (and look at all the intellectual work that went into THAT).
The thing that’s been eating me lately, almost certainly mainly triggered by the political situation in the U.S., is how to manage the transition from 2020 to what I suspect is the only way forward for the species—genetic editing to reduce or eliminate the genetically determined cognitive biases we inherited from the savannah. My objectives for the transition would be
1. Minimize death
2. Minimize physical suffering
3. Minimize mental/emotional suffering
4. Maximize critical thinking
5. Maximize sharing of economic resources
I’m extra concerned about tribalism/outgrouping and have been thinking a lot about the lunch-counter protestors in the U.S. practice/role-playing the inevitable taunts, slurs, and mild or worse physical violence they would receive at a sit-in, knowing that if they were anything less than absolute model minorities, their entire movement could be written off overnight.
I’m only just starting to look into what research there might already be on such a broad topic, so if you see this, and you have literally any starting points whatsoever (beyond what’s on this site’s wiki and SlateStarCodex), say something.

supposedlyfun 27 Sep 2020 21:16 UTC
1 point
in reply to: Matt Goldenberg’s comment on: supposedlyfun’s Shortform
I figure that at some point in the next ~300 years, computers will become powerful enough to do the necessary math/modeling to figure this out based on advances in understanding genetics.

supposedlyfun 3 Oct 2020 14:44 UTC
1 point
on: supposedlyfun’s Shortform
Does anyone have some good primary/canonical/especially insightful sources on the question of “Once we make a superintelligent AI, how do we get people to do what it says?”
I’m trying to hold the question to the question posed, rather than get into the weeds on “how would we know the AI’s solutions were good” and “how do we know it’s benign” and “evil AI in a box” as I know where to look for that information.
So assume (if you will) all other problems with AI are solved and that the AI’s solutions are perfect except that they are totally opaque. “To fix global warming, inject 5.024 mol of boron into the ionosphere at the following GPS coordinates via a clone of Buzz Aldrin in a dirigible...”. And then maybe global warming would be solved, but Exxon’s PR team spends $30 million on a campaign to convince people it was actually because we all used fewer plastic straws, because Exxon’s baby AI is telling them that the superintelligence is about tell us to dismantle Exxon and execute its board of directors by burning at the stake.
Or give me some key words to google.

supposedlyfun 3 Oct 2020 14:57 UTC
1 point
in reply to: MikkW’s comment on: MikkW’s Shortform
A cute senior in my high school Physics class asked me to tutor her after school because she was having a hard time. I can’t overstate the ways in which this improved me as a young-geek-person, and I think she got better at doing physics, too. Your proposal would tend to create more opportunities like that, I think, for cross-learning among students who are primarily book-intelligent and those who may be more social-intelligent.

supposedlyfun 4 Oct 2020 15:18 UTC
19 points
in reply to: jbash’s comment on: Postmortem to Petrov Day, 2020
The Web is full of jokey “red buttons”
This point is important. LW is the type of website I would expect to do some quirky applet that gives an object lesson in iterated game theory, or something, and as a person still trying to learn on LW’s key concepts, I would be inclined to try to play. Thus, the fact that chris pressed the button last year and entered fake codes is neutral evidence to me. It’s what I would have done if I were playing what I thought was an instructive game.
By contrast, Petrov believed (I assume correctly or at least justifiably) that all of the “buttons” available to him were real buttons that would cause the Soviets to launch at least one nuke and probably hundreds.
I don’t have a good solution for the problem of “Convince a relative newcomer to the site, solely via the button, that the button is real and will do what it says it does”. For me, as such a newcomer, this is because HPMOR and SSC (my entrees into the community) have such a screwball sense of humor.
Thus, the more times you try to say, “For real, clicking this button will actually cause the front page to go down for 24 hours,” or “Lots of people are going to be annoyed by this,” or “Clicking this button will tend to show the community that its xrisk calculations are not unreasonably high,” or even “Petrov would be ashamed of you (epistemic status: confident),” the more it looks like a game, because most of the Internet is so seamless that multiple caveats come off as satire.
Maybe you could add links to the postmortems for Petrov Day 2019 and 2020 in the button?

supposedlyfun 4 Oct 2020 15:50 UTC
1 point
on: Postmortem to Petrov Day, 2020
So in some ways this has been good for making the challenge level clear – this is a real coordination problem, that we can fail, and isn’t overdetermined in either direction.
I think every single part of this exercise, and all of the responses to it, have been extremely informative.
1. The priors of hundreds of thoughtful people have been updated substantially.
2. Others learned that they cared about something (either the site or the Petrov Day button) more than they thought they did, and, being thoughtful, learned something about themselves by thinking about why they cared so much.
3. At least one person (chris) received an object lesson in security that I doubt ze will forget.
4. The rest of us had the opportunity to learn that lesson as well, although lessons not learned in blood are soon forgotten, and for a given individual, “losing the front page for one day” is far less blood than what it cost chris.
5. I updated my priors about myself. If you’d asked me in 2019, after the successful no-press, whether I would have pressed it with codes, I probably would have said no. However, seeing the amount of thoughtfulness generated by what went down in 2020, if I get codes in 2021, I will give serious thought to pressing the button because doing so:
  1. will, no matter the reason, instill even further caution and security-paranoia in AI researchers, which is stressful for them but beneficial for humanity.
  2. causes the community to learn something about itself which can’t be predicted in advance but is almost certainly accurate.
  3. shows that if it’s something I would consider, it’s something certain types of AI would consider, too, if we’re not careful. AI risk researchers are aware of that problem, but pressing the button gives one the opportunity to make them feel it in their guts.
I wonder if there is instrumental value in pre-committing to being a Petrov Day ruiner unless certain types of conditions are met. “I am coming to the San Francisco Petrov Day 2022 celebration and will press the button five minutes before the end of the ceremony unless I first come to understand how we can solve the problem of whether an autonomous vehicle should swerve, and thereby kill the 18-year-old pregnant valedictorian, or not swerve, and thereby kill the 84-year-old semi-retired Medicin Sans Frontieres who hand-carves toys for kids at the orphanage.”

supposedlyfun 4 Oct 2020 19:09 UTC
12 points
in reply to: habryka’s comment on: Postmortem to Petrov Day, 2020
I understand and sympathize with the desire to know that people around you can hold that power without abusing it. I would also like to know that.
But it’s only ever a belief about the average behavior of the people in the community. It should update when new information becomes available. The button is a test of your belief. Each decision made by each person to press or not to press the button is information that should feed your model of how probable it is that people can hold the power without abusing it. A bunch of people staring at a red button with candles lit nearby can be an iterated Prisoner’s Dilemma depending on the participants’ utility functions.
If I decide not to Petrov-Ruin out of a desire to protect your belief that people can hold the power without abusing it, and I make that change because I care about you and your suffering as a fellow human and think your life will be much worse if my actions demolish that belief, then a successful Petrov Day is at risk of becoming another example of Goodhart’s Law.
I think, anyway. Sometimes my prose comes off as aggressive when I’m just trying to engage with thoughtful people. I swear, on SlateStarCodex’s review of Surfing Uncertainty, that I’m typing in good faith and could have my mind changed on these issues.

supposedlyfun 4 Oct 2020 19:19 UTC
1 point
in reply to: DanielFilan’s comment on: Postmortem to Petrov Day, 2020
It sounds like we draw the line differently on the continuum between persuasion and brain hacking. I’d like to hear more about why you think some or all parts of this are is hacking so I can calibrate my “I’m probably not a sociopath” prior.
Or maybe we are diverging on what things one can legitimately claim are the purposes of Petrov Day. If the purpose of an in-person celebration is “Provide a space for community members to get together, break bread, and contemplate, and the button’s raises the tension high enough that it feels like something is at stake but there’s no serious risk that the button will be pressed,” then I’m wrong, I’m being cruel, and some other forum is more appropriate for my Ruiner to raise zir concern. But I don’t get the sense that there’s a community consensus on this.
On the contrary, the fact that all attendees are supposed to leave in silence if the button is pressed suggests that the risk is meant to be real and to have consequences.

supposedlyfun 4 Oct 2020 20:18 UTC
10 points
in reply to: Ben Pace’s comment on: Postmortem to Petrov Day, 2020
I’m missing something fundamental here given that this is your view as organizer, so I’m going to go back to lurking for a while.

supposedlyfun 8 Oct 2020 1:01 UTC
3 points
in reply to: DanielFilan’s comment on: Postmortem to Petrov Day, 2020
I genuinely, sincerely appreciate that you took the time to make this all explicit, and I think you assumed a more-than-reasonable amount of good faith on my part given how lathered up I was and how hard it is to read tone on the Internet.
I think the space we are talking across is “without checking if this will make my beliefs more accurate.” Accuracy entails “what do I think is true” but also “how confident am I that it’s true”. Persuasion entails that plus “will this persuasion strategy actually make their beliefs more accurate”. In hindsight, I should have communicated why I thought what I proposed would make people’s beliefs about humanity more accurate.
However, the response to my comments made me less confident that the intervention would be effective at making those beliefs more accurate. Plus, given the context, you had little reason to assume that my truth+confidence calculation was well-calibrated.
There’s also the question of whether the expected value of button-pressing exceeds the expected life-worsening, and how confident a potential button-presser is in their answer and the magnitude of the exceeding. I do think that’s a fair challenge to your final thought.
Thanks again.

supposedlyfun 19 Oct 2020 2:11 UTC
1 point
on: AI safety from first principles: Alignment
“A robot arm with a purposely disabled gripper found a way to hit the box in a way that would force the gripper open”

The list of AI misbehavior in the blog you link to should be in the Sequences. I had no idea that we were programming things with this level of AI-ish sophistication in the 90s, much less that they were going so spectacularly wrong sometimes. Hello, Skynet.

This is a great series that is written in reasonably plain English with only necessary jargon, and I thank and salute you.

supposedlyfun 28 Oct 2020 12:12 UTC
3 points
in reply to: cousin_it’s comment on: Notes on Honesty
I’m not per se disagreeing with your sentiment about Odysseus, but it would be helpful if you could model it more explicitly than “lies like he tells are okay”. I can see why SPOILERS not telling Polyphemus your real name is potentially acceptable dishonesty, but Kant would disagree, I guess.

The problem with setting a fictional hero as your standard for any component of morality is that the omniscient narrator will often have your back in edge cases. The best example to me is Ender’s Game, SPOILERS in which he kills two people without intending to, but also intends to attack them in potentially deadly ways and does so; yet in the next chapter, the literal adults in the literal next room are there to tell the audience, “If Ender hadn’t killed Bonzo, Bonzo would have killed him, and we then wouldn’t win the war, so it’s fine.” To me, their belief is maybe true but definitely not justified on the evidence. Bullies don’t kill people all the time. Card is heavily motivated to protect Ender’s morality for various reasons and builds the story around him for that particular goal.

supposedlyfun 2 Nov 2020 1:37 UTC
3 points
in reply to: Adam Zerner’s comment on: What is the right phrase for “theoretical evidence”?
Imagine that you are working on a product. A/B tests are showing that option A is better, but your instincts, based on your understanding of how the gears turn, suggest that B is better.
Imagining it now. “are showing” makes it sound like your A/B tests are still underway, in which case wait for the study to end (presumably you designed a good study with enough power that the end results would give you a useful answer on A vs. B). But if the tests show A > B, why would you hold on to your B > A prior? Or if you think the tests are only 50% conclusive, why would you not at least update the certainty or strength of your B > A prior?
I think this is why Idan said, “Or, since they explicitly go against the empirical evidence, how about we just call it ‘stubbornness’?”

supposedlyfun 2 Nov 2020 2:09 UTC
1 point
in reply to: Adam Zerner’s comment on: What is the right phrase for “theoretical evidence”?
Okay, thank you for engaging. Those answers weren’t clear to me from the parent piece.
Maybe I reacted strongly because my current prior on my own intuitions is something like “Your intuition is just savannah-monkey-brain cognitive shortcuts and biases layered over your weird life experiences”. One thing I’ve been thinking about lately is how often that prior is actually justified versus how often it’s merely a useful heuristic (or a shortcut/bias? ha!) to remind me to shut up and Google/multiply.

supposedlyfun 3 Nov 2020 15:34 UTC
2 points
in reply to: cousin_it’s comment on: Notes on Honesty
Totally fair point. I agree that not all fictional heroes’ possibly justifiable lies are subject to my “author writes the story to protect reader perception of hero” concept. Maybe narrow my comment to “When using a fictional character’s lies as a model for when lying is acceptable, one should be alert for situations where the author has built the fictional world in such a way that Lie X is maybe justifiable in the fictional world but would not be in nearby counterfactual worlds.”

supposedlyfun 4 Nov 2020 23:47 UTC
1 point
in reply to: ChristianKl’s comment on: ChristianKl’s Shortform
I endorse gjm’s final comment 100%. (I wrote a much longer response but eventually decided that it was just repeating what ze said.)

supposedlyfun 5 Nov 2020 1:13 UTC
3 points
on: Multiple Worlds, One Universal Wave Function
This added to my layperson’s understanding of both MWI and quantum mechanics more generally.
Immediately under the subhead “The Apparent Collapse of The Wave Function,” what is a-sub-i in the initial state?