7 ms·

This is a classic AI planning problem, solvable with STRIPS. Solution to the Sussman Anomaly https://github.com/primaryobjects/strips/blob/master/Readme.md#su

This is a classic AI planning problem, solvable with STRIPS. Solution to the Sussman Anomaly https://github.com/primaryobjects/strips/blob/master/Readme.md#sussman-anomaly-solution https://github.com/primaryobjects/strips/blob/master/Readme....

RyanShook 9y ago
Are there any unsolved/unsolvable anomalies similar to this one?
- graycat 9y ago
  I'd say that all the NP-complete problems are of this kind and much worse.
jerf 9y ago
I'd imagine it's solvable by just about everything. This is one of those cases where the core problem isn't that the problem is hard, it is that it is so easy that it is a challenge to get humans to even perceive it as a problem on such a small scale. Any planning system will hit something like this all but immediately. I'd say a lot of the earliest AI optimism was a failure to understand just how hard things that in the simplest case are so simple we don't even perceive them as problems in the first place. For a much more complicated example, I doubt anyone in 1970 would imagine that in 2018 we're still trying to get robots to walk from here to there. (Yes, lots of progress has been made, but it's still very much cutting edge, a research project, and a bespoke one-off effort each time, not a "Oh, I'll just swing on down to the robot store and pick up a walking chassis".) We don't see it as a challenge because for our conscious minds, by the time we're old enough to be doing robotics research, it's zero cognitive effort for us.
- FabHK 9y ago
  Moravec's paradox - stuff we thought is hard (chess, derivatives) is easy, while stuff we think is easy (face recognition, getting out of a car, opening a door) is really hard. https://en.wikipedia.org/wiki/Moravec%27s_paradox https://en.wikipedia.org/wiki/Moravec%27s_paradox One of my favourite videos is this compilation of robots falling (failing) during the simplest tasks at the 2015 DARPA challenge: https://www.youtube.com/watch?v=g0TaYhjpOfo https://www.youtube.com/watch?v=g0TaYhjpOfo
  - lloeki 9y ago
    To be fair, it takes us humans a solid couple of years of continuous training to make sense of our sensory inputs and get accustomed to our limbs enough to begin doing anything significant with them, so I am regularly flabbergasted at the expectations we place in our lowly contraptions to outdo us by two to three orders of magnitude.
    - bsaul 9y ago
      And yet humans need even more years to correctly play chess.
      - Frqy3 9y ago
        Humans have had millions of years of system level optimisation (evolution) which then only needs a couple of additional years to train the wetware to operate the system for daily activity. Chess has only been around a mere 1,500 years and has had an evolutionary impact on only a very minute portion of the population. Hence it takes a few additional years to specialise a human for optimal chess performance.
        Simon_says 9y ago
        Optimal chess performance is not remotely achievable by humans.
mannykannot 9y ago
This solves the problem by searching the graph of reachable states (and presumably doing cycle detection, at least in depth-first mode), which avoids the problem of picking an unsuitable sub-goal by considering all possible ones. While this is a perfectly good solution, it doesn't seem to throw much light on how human intelligence works. A person is likely to notice, for example, that extending any stack that does not have C on the bottom is not helpful, without mentally running through all possibilities, but just on the basis that it will have to be torn down at some point, to move the bottom block to where it belongs.
- CarolineW 9y ago
  It also just "blows up" on any problem of any real complexity, and as such isn't really a solution.
  - adrianratnapala 9y ago
    Of course human intelligence also stumbles on planing problems of true complexity. So it is still possible that human intelligence does something like a naive graph search, and this is good enough in the common case.
    - mannykannot 9y ago
      I do not think naive search would give rise to the sort of reasoning that I posit in my initial post, even as a post-facto rationalization rather than as part of actually solving the problem.

RyanShook 9y ago

Are there any unsolved/unsolvable anomalies similar to this one?

graycat 9y ago
I'd say that all the NP-complete problems are of this kind and much worse.

jerf 9y ago

I'd imagine it's solvable by just about everything. This is one of those cases where the core problem isn't that the problem is hard, it is that it is so easy that it is a challenge to get humans to even perceive it as a problem on such a small scale. Any planning system will hit something like this all but immediately. I'd say a lot of the earliest AI optimism was a failure to understand just how hard things that in the simplest case are so simple we don't even perceive them as problems in the first place. For a much more complicated example, I doubt anyone in 1970 would imagine that in 2018 we're still trying to get robots to walk from here to there. (Yes, lots of progress has been made, but it's still very much cutting edge, a research project, and a bespoke one-off effort each time, not a "Oh, I'll just swing on down to the robot store and pick up a walking chassis".) We don't see it as a challenge because for our conscious minds, by the time we're old enough to be doing robotics research, it's zero cognitive effort for us.

FabHK 9y ago
Moravec's paradox - stuff we thought is hard (chess, derivatives) is easy, while stuff we think is easy (face recognition, getting out of a car, opening a door) is really hard. https://en.wikipedia.org/wiki/Moravec%27s_paradox https://en.wikipedia.org/wiki/Moravec%27s_paradox One of my favourite videos is this compilation of robots falling (failing) during the simplest tasks at the 2015 DARPA challenge: https://www.youtube.com/watch?v=g0TaYhjpOfo https://www.youtube.com/watch?v=g0TaYhjpOfo
- lloeki 9y ago
  To be fair, it takes us humans a solid couple of years of continuous training to make sense of our sensory inputs and get accustomed to our limbs enough to begin doing anything significant with them, so I am regularly flabbergasted at the expectations we place in our lowly contraptions to outdo us by two to three orders of magnitude.
  - bsaul 9y ago
    And yet humans need even more years to correctly play chess.
    - Frqy3 9y ago
      Humans have had millions of years of system level optimisation (evolution) which then only needs a couple of additional years to train the wetware to operate the system for daily activity. Chess has only been around a mere 1,500 years and has had an evolutionary impact on only a very minute portion of the population. Hence it takes a few additional years to specialise a human for optimal chess performance.
      - Simon_says 9y ago
        Optimal chess performance is not remotely achievable by humans.

mannykannot 9y ago

This solves the problem by searching the graph of reachable states (and presumably doing cycle detection, at least in depth-first mode), which avoids the problem of picking an unsuitable sub-goal by considering all possible ones. While this is a perfectly good solution, it doesn't seem to throw much light on how human intelligence works. A person is likely to notice, for example, that extending any stack that does not have C on the bottom is not helpful, without mentally running through all possibilities, but just on the basis that it will have to be torn down at some point, to move the bottom block to where it belongs.

CarolineW 9y ago
It also just "blows up" on any problem of any real complexity, and as such isn't really a solution.
- adrianratnapala 9y ago
  Of course human intelligence also stumbles on planing problems of true complexity. So it is still possible that human intelligence does something like a naive graph search, and this is good enough in the common case.
  - mannykannot 9y ago
    I do not think naive search would give rise to the sort of reasoning that I posit in my initial post, even as a post-facto rationalization rather than as part of actually solving the problem.