Posts
-
The Blogging Gauntlet: May 19 - Working Assuming Failure
CommentsThis is part of The Blogging Gauntlet of May 2016, where I try to write 500 words every day. See the May 1st post for full details.
Alternate title: This Was a Bad Idea.
I’ve been working all day on turning my final project for CS 281B into a NIPS paper. The submission deadline is tomorrow, 9 AM. I have no idea how I convinced myself this was a good idea.
First off, the odds that I actually get accepted to NIPS is almost nil. My final project has the bare minimum of algorithm design to make it even justifiable to write a paper in the first place, and the experiments I ran all gave negative results. Furthermore, I worked on this project entirely by myself, with little to no input from any grad students/professors. That means I’ll have no feedback from people who have experience with the submission process, and that I’ll be editing the paper entirely by myself, which is a huge recipe for disaster. Everyone I’ve worked with is busy attending ICRA, and the time it would take for me to explain what I’m doing and for them to read my paper isn’t worth it. Besides, even if this was the most well-written paper in the world, the combination of not very novel ideas and negative results feels like a death sentence.
What makes this an even worse idea is that I didn’t even learn about the NIPS deadline until 2 days ago. Sure, let’s submit to NIPS! Why not! It’s not like I’ve spent the past few days playing Phoenix Wright: Ace Attorney in my pajamas all day, meaning I have to go through a super jarring context shift into the work all day mindset. It’s not like this is my last week staying at Berkeley, and I should probably be trying to meet up with friends before we all head away from one another. OH WAIT, EXCEPT BOTH THOSE THINGS ARE TRUE? YEAH, THEY ARE.
UUUUUUGGGGGGHHHHHHHH.
(Yes, if you couldn’t tell I’m a bit annoyed with myself. It was probably a bad idea to even write a blog post when I’m in such a mood, but here we are. On the raggedy edge.)
Look, I even did an estimation in my head, and decided it wasn’t worth sending my paper around for feedback, because it would take away time from people working on their papers, and the marginally improved chance for my paper isn’t worth diminishing the chances of people who actually have significant progress. Maybe this is circular logic I’m giving myself to not talk to people, but at this point I’m too tired for self improvement or internal consistency.
By the way, if you were expecting a post about existential risk and why it’s justifiable to act in situations where you almost always receive failure, I’m sorry the title baited you. See, that’s what I thought this post was going to be about, but my stream of consciousness decide to hijack it into a rant post and I don’t have the time to do that topic full justice. I’m writing this on an insanely short deadline because I need to both finish this post and finish my paper before midnight to give myself sleeping time for the CS graduation ceremony tomorrow morning.
Anyways this post is a huge pile of bullshit and I’m sorry you sat and read through it. I’m cheating myself for writing this just to make sure I don’t have to pay $20, but I set the rules! Only myself to blame if they let me get away with this!
If I had the time to write a decent post, here’s the ideas I would have tried to convey.
- The odds of me getting accepted are small, but I was always planning to refine my final project for my research page currently under construction. The NIPS deadline just gave me a good deadline to actually work on it and spring me out of my hedonistic haze.
- When the paper is done, I’ll get to say I submitted a paper to a top tier machine learning conference, which is actually a decently big deal. Of course anybody can do this, so it’s secretly not a big deal.
- I’m not sure how I feel about existential risk, but in this case the journey of finding the best way to explain my ideas to a curious but unfamiliar audience and the experience I get doing so is the big thing I’m getting out of it.
I didn’t have the time to find a good way to naturally bring all this up, so instead you get a bullet point text dump. Sorry!
Okay this is actually closer to 750 words, not 500. I’m outta here, I’ve got a paper to write.
-
The Blogging Gauntlet: May 18 - More Time Travel Password Theorizing
CommentsThis is part of The Blogging Gauntlet of May 2016, where I try to write 500 words every day. See the May 1st post for full details.
This builds upon the ideas from yesterday’s post. Read that one first.
First off, here’s some trivia I learned from the TVTropes page on trust passwords. Harry Houdini and his wife Bess developed secret passphrases which he promised to say to her if he found a way to contact her from the afterlife. After his death, Bess used those passphrases to debunk several spirit mediums. Or, as put by MagicPedia, “Bess began the tradition of holding a seance to see whether Houdini could escape from death.” None of the spirit mediums ever convinced her.
Can I just say: this is totally badass. I never realized time travel passwords could also be used to authenticate yourself from the dead.
(Also, MagicPedia implies she did this as a coping mechanism for her loss, instead of as a way to debunk mystics. I like the latter interpretation more.)
Anyways, so I’ve been thinking about the security of time travel passwords. Suppose you’re kidnapped, and your kidnappers are trying to coerce you to reveal your time travel password. What do you do? Obviously, you could lie about your password. However, this may not help you. If your kidnappers can verify whether messages they sent into the past have changed the present, they’ll know you’re lying.
This doesn’t mean you shouldn’t create a time travel password. It still gives you some security, and if you are kidnapped, you always have the option of giving the adversary your password.
In some ways, this is the best it ever gets. Here’s the quick argument. Any time travel authentication scheme must have a way to convince your past self a message is from the future. This is true by definition. Otherwise, even when you’re sending a message of your own free will, your past self could reject your messages. In other words, every authentication scheme must have some way to let people through the door. If kidnappers have convinced you to cooperate with them, you’ll always be able to open the door for them. There’s no way around it.
Still, there’s room for improvement. Suppose I accidentally say my time travel password out loud. Anyone present can now pretend to be me from the future. And importantly, I have no way to tell my past self their time travel password is no longer secure. Time travel passwords are actually less secure than regular passwords! If we reveal our password in real life, we can request a password change, but this doesn’t work for time travel. You can’t retroactively change your memory! You’re screwed!
(Would this be a bad time to mention I’ve never seen Back to the Future? I know, I know, I’m sorry.)
At best, you can send a message ASAP telling your past self to change their time travel password. But, that assumes you get there first. If you get there 2nd or 3rd, you’re sunk. By that point, your past self has had years to act on malicious information. Who knows what could happen?
An ideal time travel auth-system should minimize the danger from revealing your password. I haven’t thought about it for long, but one option is two-factor authentication. Require every message to come with both a password and a physical token that only you own. That way, anyone who learns your password can’t fool your past self.
Unfortunately, this assumes you can send objects back in time, which might not be true. There’s another issue too: you’re not getting your physical token back after sending it to the past. If you want to send more than one message to the past, you’ll need a new token for each one.
At this point, we’re running into the usability vs security issue. Yes, if you try hard enough, you can make more secure time travel authentication schemes, but at some point it’s not worth the effort to implement them, and theorizing about them is just an interesting exercise. Personally, I’m going to stick to passwords, because they’re good enough for me.
-
The Blogging Gauntlet: May 17 - What's Your Time Travel Password?
CommentsThis is part of The Blogging Gauntlet of May 2016, where I try to write 500 words every day. See the May 1st post for full details.
As far as I know, time travel doesn’t exist. However, that doesn’t mean there’s nothing interesting to say about it. In fact, you can do a lot of theorizing about models of time travel.
Today, I made a time travel password. What I’m going to do is detail the argument for why everyone else should make one too.
***
Suppose we’re living in a world where time travel is possible, and that furthermore time travel will be invented within our lifetime. I know this is a huge assumption, but bear with me for a moment.
A world where people can transmit information to past versions of themselves is very hard to predict. The nature of that world depends on the model of time travel used, but the potential gains of time travel are huge.
However, that means the potential losses of forged future messages are also huge. If you’re in a world where time travel is possible, the strangest requests in the world can be explained away by saying the future depends on it. What makes decision making tricky is that it’s almost always done under uncertainty. Time travel removes that uncertainty, giving a perfect oracle towards achieving the best outcome. Or, if a message is forged, a perfect oracle towards achieving the worst outcome.
To prevent this, you need an authentication scheme that verifies future messages come from you. Hence, time travel passwords! In the present, you pre-commit to adding your time travel password to any message sent to a past version of you. That way, if you ever receive such a message, you can verify that message is legitimate.
Importantly, you never, under any circumstances, write down your time travel password or speak it out loud. After the password leaves your head, it’s no longer secure. As a corollary, every time travel password is one time use, and you must create a new one whenever you receive a future message.
***
There’s a very natural objection to this: isn’t time travel exceedingly unlikely? Why should I waste my time making a password for non-existent technologies?
Well, consider the following.
- Creating a time travel password takes less than a minute.
- If time travel is never invented, you lose one minute of your life.
- If time travel is invented, you prevent a large class of attacks that could be made against you.
The expected utility is vastly, vastly in favor of making a time travel password. I’ll fully admit this argument is a variant of Pascal’s mugging, which you might reject for one reason or another.
Here’s my counterargument. Yes, maybe I’m mugging you out of one minute of your time. However, the value of one minute of your time is very, very low. And even if time travel is never invented, you get to tell people you have a time travel password! How awesome is that? You can explain this very argument, or link to help increase my view counts (nod nod, wink wink).
Having a time travel password makes you a more interesting person, and that’s surely worth the one minute of time it costs you.