Three weeks ago I got tested at the Planned Parenthood on Telegraph Avenue in Oakland. The results came back on July 22. I haven’t opened them and don’t intend to. I also won’t share them with the four women I’ve slept with since. Under any serious reading of Nick Bostrom’s work on information hazards, the contents of that MyChart message are dangerous true information, and the responsible thing to do with dangerous true information is to contain it.
I know how this sounds outside the community. But we have a long tradition here of recognizing that some knowledge does more harm than good. In 2010 Eliezer Yudkowsky deleted a post from this site and banned discussion of it for five years because he judged that merely reading an argument could hurt people. Nobody called that cowardice. They called it stewardship. My results are a far more ordinary hazard than Roko’s basilisk, but the principle is the same. Unlike the basilisk, my hazard has burned like a struck match every time I piss since the Fourth of July.
The taxonomy
Bostrom’s 2011 paper sorts information hazards into types. My results fit most of them.
Data hazard: specific information that is dangerous to spread. A positive result would move through the East Bay dating graph faster than anything I might be carrying.
Evocation hazard: information that triggers harmful mental states in the people who receive it. My girlfriend, Lauren, has an anxiety disorder. I’m not going to hand her a document that could set it off over something that may turn out to be nothing.
Signaling hazard: the act of sharing sends information of its own. A man who discloses STD results signals that he’s the kind of man who has STD results worth disclosing.
Knowing-too-much hazard: knowledge that creates obligations. In most states, the question that matters is what you knew. I have arranged not to know.
Attention hazard: even drawing attention to the existence of the information does damage. This is why I deleted the MyChart app and blocked the clinic’s number.
Why I haven’t opened them
The value of information is the improvement it makes to your decisions. Here is mine:
V(open) = P(+)·[U(stop, disclose) − U(continue)] + P(−)·0
If the results are negative, nothing changes. If they’re positive, I’d have to stop sleeping with people and start telling the ones I already have. By my own preferences, that’s worse than continuing. Every term is negative or zero. I’m not avoiding the truth. I’m declining a bet with no upside.
Containment
Bev from the clinic has left four voicemails, each more urgent than the last, and the urgency is itself information I’ve had to firewall. Before I could look away, I saw the first line of the iPhone transcription of the second one: “Hi, this is Bev from Planned Parenthood, I’m calling about your results, it’s important that you—”. That fragment alone moved my estimate more than I’d like. I’ve since turned off voicemail transcription.
The four women are Lauren; Marisa from Hinge; Dana, who sits two rows from me at work; and Kelsey, from a bachelorette party in Nashville. I never learned Kelsey’s last name, which is its own kind of containment.
Coordinated disclosure
In computer security, a researcher who finds a vulnerability tells the vendor privately and gives them ninety days to patch before anything goes public. I’ve adopted the same norm. I’ve notified the vendor, my immune system, and I’m giving it ninety days. The window closes October 20. If the symptoms resolve by then, there will be nothing to disclose.
Meanwhile I’m hedging. I’ve been taking my roommate Derek’s leftover amoxicillin from his wisdom teeth. I’m treating a hypothesis I haven’t tested, which is what any good Bayesian does with a high-prior, high-cost possibility.
Objections
People say the women have a right to know. Rights-talk is how people avoid doing the math. A right to information that makes you miserable, ends a relationship and changes nothing about what has already happened is a right to a worse life. This community has spent fifteen years establishing that true information is not automatically good to share. I am simply one of the few people willing to apply that to my own dick.
Lauren asked me last night, point blank, whether I’d been tested. I told her the truth: yes. She didn’t ask about the results, and I have always believed it’s wrong to volunteer hazardous information to someone who hasn’t asked for it.
Please call the clinic.
Coordinated disclosure gives the vendor time to patch before users are exposed. Here the users are already exposed, every day of the window. The closest analogue is a vulnerability that's being actively exploited, and for those the norm is a deadline measured in days, sometimes seven. Ninety days is the timeline for bugs nobody is using yet.
The exploitation rate is lower than you're implying. I've been seeing fewer people since the symptoms started.
Bostrom's paper is about information whose spread harms the world. Your results are information whose suppression harms the world. The sign is flipped.
The sign depends on whose welfare you're measuring. I've been explicit that I'm measuring mine.
V(open) leaves out your own treatment. Opening the results changes whether you get the right treatment, which matters for your welfare even if you count nobody else's. Amoxicillin isn't the standard drug for most of the usual suspects, and a partial course can take the edge off symptoms without clearing anything. By your own preferences the information is worth something.
Treatment is already in progress.
That's the point. You don't know whether it's treating what you have, and the only way to find out is in the MyChart message.
I got tested last week and I'm emailing you my results. I understand you consider unsolicited test results hazardous, so I've put them in the subject line.
On the knowing-too-much hazard. Plenty of courts treat deliberately avoiding knowledge as having it. Willful blindness is the term. Blocking the clinic's number after four voicemails is close to a textbook fact pattern.
That was my last amoxicillin. I still have the other wisdom tooth.
For what it's worth, the 2010 deletion you cite is the most famous Streisand effect in the history of this site. Containment made a forgotten post into the one thing everyone has heard of. I wouldn't use it as the model for keeping anything quiet.
October 20 is ninety days from the 22nd. What's the plan if the symptoms haven't resolved by then?
Then the vendor has failed to patch and I'll disclose to myself first.
bro IS the hazard