ashtonabc

Karma: 1

ashtonabc 13 May 2018 2:46 UTC
1 point
on: AI Alignment is Alchemy.
Even if your characterization of AI was accurate, an unaligned AI could find it valuable to dedicate part of its resources to helping humans. In your analogy, isn’t that the role of myrmecologists today? If it were ants:us = us:AI, solving our problems wouldn’t require a significant expenditure of resources.

ashtonabc 5 May 2018 3:13 UTC
1 point
on: Finding a Girlfriend with Reinforcement Learning
I like the thinking that went into this post, but I also think it’s difficult to make any definitive statements here. None of the actions you’ve given are entirely independent of the others (for example, you can attend new events with friends). It’s also difficult to use an algorithmic approach without a good way of measuring expected returns, which are difficult to intuit and change significantly over time.
Even ignoring varying individual experiences with online dating (conventionally attractive/non-minority individuals tend to have better success), there may be actions that you can take to make it more efficient. It can be also done at times when you are unable to go out and can be done for even for small amounts of time.
I think algorithmic/rationalist approaches to dating are really interesting. I’m not certain that reinforcement learning is any different than non-algorithmic/rationalist approaches though. Aren’t humans always trying to maximize our expected reward?

ashtonabc 2 Oct 2017 0:39 UTC
2 points
on: Strategic Goal Pursuit and Daily Schedules
Habit formation is really difficult. The way I like to think about goal lists and daily schedules is as productivity supporting habits rather than directly productive habits.
It’s not unusual to have some productive habits. Going to the gym before work, doing X hours of uninterrupted work each weekday, cleaning your home once a week, etc, can be hard habits to build, but it can and is done. The problem is that you need to do the habit formation work every time you have a new goal.
Other habits can help support productivity even if they aren’t directly productive themselves. Being in the habit of creating schedules and sticking to them (or effectively using a to-do list) can help support other productive activities, but isn’t directly productive itself. These type of habits are nice because you only need to build them once. If you always follow your schedule, you can engage in a variety of productive activities without building each habit individually; that habit can serve as an action trigger.
Obviously it’s not always this simple. I think building up the habit of actually completing the work listed in your calendar can be very challenging, perhaps more so than building up individually productive tasks. I wasn’t able to do it, even after scheduling in leisure time. I settled on a to-do list workflow (styled off of GTD) that works nicely for me. I now use calendars only for hard times (events, deadlines, birthdays, etc), otherwise I start to feel like everything on the calendar is as flexible as the daily work that I schedule in.
Doing some of the meta work described in the Anna Salamon piece would likely improve my goal management. I’ll try to work some of those heuristics into my existing goal list, which is likely too vague. I wonder if there is a way to work some of those heuristics into a calendar or to-do list habit to improve their efficacy.