• Posts
  • RSS
  • ◂◂RSS
  • Contact

  • Conversation with Bryce Wiedenbeck

    July 14th, 2017
    airisk, giving  [html]
    A few days ago I spoke with Bryce Wiedenbeck, a CS professor at Swarthmore teaching AI, as part of my project of assessing superintelligence risk. Bryce had relatively similar views to Michael: AGI is possible, it could be a serious problem, but we can't productively work on it now.

    Before our conversation he looked some at Concrete Problems in AI Safety (pdf) and Deep Reinforcement Learning from Human Preferences (pdf). His view on both was that they were good work from the perspective of advancing ML but very unlikely to be relevant to making AGI safer: the systems that get us to AGI will look very different from the ones we have now.

    One reason is that he saw a lot of learning from humans as being mediated by learning utility functions, but he sees utility functions as a very limited model. Economists and others use utility functions when talking about people because that's mathematically tractable, but it's a bad description of how humans actually behave. Trying to come up with utility functions that best explain human preferences or behavior probably solves some problems nicely and is helpful, but while Bryce wouldn't completely rule it out he thought it was very unlikely to get us to AGI.

    We tried to get more into why he thinks implementations for AGI will look vastly different from what we will have today, and couldn't make progress there. Bryce thinks there are deep questions about what intelligence really is that we don't understand yet, and that as we make progress on those questions we'll develop very different sorts of ML systems. If something like today's deep learning is still a part of what we eventually end up with, it's more likely to be something that solves specific problems than as a critical component.

    (This has been a common theme in my discussions with people recently: very different intuitions on the distance to AGI in terms of technical work required, and also on whether work we're doing today is likely to transfer.)

    Comment via: google plus, facebook

    Recent posts on blogs I like:

    What should we do about network-effect monopolies?

    Many large companies today are software monopolies that give their product away for free to get monopoly status, then do horrible things. Can we do anything about this?

    via benkuhn.net July 5, 2020

    More on the Deutschlandtakt

    The Deutschlandtakt plans are out now. They cover investment through 2040, but even beforehand, there’s a plan for something like a national integrated timetable by 2030, with trains connecting the major cities every 30 minutes rather than hourly. But the…

    via Pedestrian Observations July 1, 2020

    How do cars fare in crash tests they're not specifically optimized for?

    Any time you have a benchmark that gets taken seriously, some people will start gaming the benchmark. Some famous examples in computing are the CPU benchmark specfp and video game benchmarks. With specfp, Sun managed to increase its score on 179.art (a su…

    via Posts on Dan Luu June 30, 2020

    more     (via openring)


  • Posts
  • RSS
  • ◂◂RSS
  • Contact