The term "artificial intelligence" should not be used when what's meant is "optimization of a not terribly complicated function in a not terribly large number of dimensions".
[EDIT: I see someone disagrees with this, or with the fact that I said it out loud, or something, strongly enough to downvote it. I would be interested to know why, though I don't guarantee to be persuaded.]
Thanks for your comment. I don't see how the difficulty of a problem alters the class of the algorithm that is applied towards finding a solution; the algorithms that we used here are mentioned in almost every classic textbook on AI (Russel, Norvig etc.). If you're alluding towards the difference b/w weak and strong AI, then that's another discussion (btw, I didn't down-vote your comment).
Yes, these optimization algorithms are used in AI. But that doesn't make them AI. Loops are used in AI; strings are used in AI; but a loop isn't an "AI algorithm" and a string isn't an "AI data structure", and writing a program that uses strings and loops wouldn't be "using AI to [do whatever]".
Simulated annealing is a less trivial thing than loops and strings. But it still isn't specifically an AI technique. Simple hill-climbing optimization, even less so. I'm pretty sure that a hefty majority of the uses of these things is unambiguously not AI unless AI is defined super-broadly (e.g., as "doing something tricky with a computer").
(No, I wasn't alluding to the distinction between "weak" and "strong" AI. I'm perfectly happy to call things AI that don't come close to actually implementing human-like intelligence!)
I completely agree.
This really bothered my during my AI course.
More then half of the course were just search algorithms, calling that intelligence is stretching the term.
What i actually have been thinking about is that ai researchers should instead look at having several databases of knowledge in one agents with maybe one "neutral" base database.
This is because of my recent interest in MBTI and Socionics systems based on carl jungs work. Here every human has four main functions and each function has two positions (attitudes) giving a total of 8 real functions and a total of 16 human types. Each real function vies and models the world differently and each human has a priority of their function chain, every human has all the functions just in varying strengthen.
So i have concluded that for an ai to actually be more like humans it needes multiple databses of knowledge, or at least one neutral database with views on it. Example could be a physical database but were the things could be label and valued according to function in the real world or emotional attachment based on the memory of the agent.
The important thing is the coexistent of multiple worlds instead of the singularity focus that our ai focus has. Most of the ai debate is between choosing of multiple instead of seeing the solution to have both (all) models.
Here i am finished i will just give an overview of the functions for anyone interested:
Thinking
Feeling
Sensing
Intution
Thinking and feeling are judging functions and sesning and intution are perception functions. Each function has a subjective (intravert) and an objective (extraverted) attitude (position).
For example intraverted feeling is paying attention to your own feelings and value system and being true to them. Extraverted feeling is knowing the groups values and taking care of your fellow humans.
It seems to me like a spring-mass system would be another good thing to try here. That would give you similar behavior to hill climbing, but would also be better at avoiding local minima since I can push neighboring labels away.
I thought the article was well written and interesting but I had the same reaction you did: the first thing I would have tried would have been a force-graph.
Thanks for your comments. Plugging in a Physics engine to do just that did cross my mind but I ended up using more traditional optimization algorithms b/c they're incredibly easy to implement. Plus, I had worked with them before.
7 comments
[ 2.8 ms ] story [ 26.0 ms ] thread[EDIT: I see someone disagrees with this, or with the fact that I said it out loud, or something, strongly enough to downvote it. I would be interested to know why, though I don't guarantee to be persuaded.]
Yes, these optimization algorithms are used in AI. But that doesn't make them AI. Loops are used in AI; strings are used in AI; but a loop isn't an "AI algorithm" and a string isn't an "AI data structure", and writing a program that uses strings and loops wouldn't be "using AI to [do whatever]".
Simulated annealing is a less trivial thing than loops and strings. But it still isn't specifically an AI technique. Simple hill-climbing optimization, even less so. I'm pretty sure that a hefty majority of the uses of these things is unambiguously not AI unless AI is defined super-broadly (e.g., as "doing something tricky with a computer").
(No, I wasn't alluding to the distinction between "weak" and "strong" AI. I'm perfectly happy to call things AI that don't come close to actually implementing human-like intelligence!)
What i actually have been thinking about is that ai researchers should instead look at having several databases of knowledge in one agents with maybe one "neutral" base database.
This is because of my recent interest in MBTI and Socionics systems based on carl jungs work. Here every human has four main functions and each function has two positions (attitudes) giving a total of 8 real functions and a total of 16 human types. Each real function vies and models the world differently and each human has a priority of their function chain, every human has all the functions just in varying strengthen.
So i have concluded that for an ai to actually be more like humans it needes multiple databses of knowledge, or at least one neutral database with views on it. Example could be a physical database but were the things could be label and valued according to function in the real world or emotional attachment based on the memory of the agent.
The important thing is the coexistent of multiple worlds instead of the singularity focus that our ai focus has. Most of the ai debate is between choosing of multiple instead of seeing the solution to have both (all) models.
Here i am finished i will just give an overview of the functions for anyone interested: Thinking Feeling Sensing Intution
Thinking and feeling are judging functions and sesning and intution are perception functions. Each function has a subjective (intravert) and an objective (extraverted) attitude (position). For example intraverted feeling is paying attention to your own feelings and value system and being true to them. Extraverted feeling is knowing the groups values and taking care of your fellow humans.
Well i will end with coffe makes wonders.