for fans of lectures, Janzing and Schoelkopf (who are among the author's of the paper Ben just linked and are among the most established people in the field afaik): https://www.youtube.com/watch?v=KsbftkwZTq4
On Tue, Jan 13, 2015 at 5:53 PM, Ben Goertzel via AGI <[email protected]> wrote: > See Figure 1 in > > http://arxiv.org/abs/1407.2256 > > Distinguishing between cases a) and b) is useful in narrowing down > causality > > But yeah, most interesting causal inference involves cases where the > agent can't actually carry out the exact manipulations in question. > So then IMO causal inference involves analogical inference, whether > the source of the analogy is some related case where the agent could > carry out a direct manipulation... > > ben g > > On Tue, Jan 13, 2015 at 11:48 AM, Aaron Hosford <[email protected]> > wrote: > >> Are you familiar with "causal networks" as developed by Judea Pearl, > >> etc.? > > > > > > I have heard the name thrown around but that's the extent of it. I'll > have > > to read up on this. > > > >> But once the causal network theory has done its > >> job, one is left with a set of events that are *potentially* > >> consistently labeled causal -- one still needs some other kind of > >> insight to tell what should really be considered causal or not > > > > > > That's a bit vague. Can you put your finger on precisely which types of > > event pairs would mistakenly be considered potentially causal by this > > approach? Based on what you have said so far, I would think it would > just be > > a matter of formulating an experiment. The agent could either test the > > ability to control one variable (the effect) with another (the cause) by > > controlling the other, or if the cause itself could not be controlled by > the > > agent then whether it is "truly" a cause is a moot point because the > > information is inaccessible and consequentially irrelevant to the agent's > > internal model. > > > > On Tue, Jan 13, 2015 at 11:34 AM, Ben Goertzel <[email protected]> wrote: > >> > >> Hi Aaron, > >> > >> Are you familiar with "causal networks" as developed by Judea Pearl, > >> etc.? The basic theory thereof involves precisely this type of > >> manipulation that you're talking about.... > >> > >> However, I've never been fully satisfied with causal network theory. > >> Ultimately my conclusion is that, given a network of probabilistically > >> related events, causal network theory helps rule out which > >> relationships are almost surely NOT causal but rather consequential > >> from other events. But once the causal network theory has done its > >> job, one is left with a set of events that are *potentially* > >> consistently labeled causal -- one still needs some other kind of > >> insight to tell what should really be considered causal or not > >> > >> > >> -- Ben > >> > >> On Tue, Jan 13, 2015 at 11:23 AM, Aaron Hosford <[email protected]> > >> wrote: > >> > Ben, > >> > > >> > I'm not sure this would be useful within the scope of what you're > >> > working > >> > on, but I don't think probabilities are the right basis for > >> > understanding > >> > causation. Causation, to me, requires interaction between objects or > >> > events > >> > that results in a transfer of information/control. Think of a > >> > simulation. If > >> > changing the simulation in some way (i.e. modifying the state of an > >> > object > >> > or introducing/removing an object or event) changes whether a > predicate > >> > of > >> > the simulation's (future) history is satisfied, then that change can > be > >> > said > >> > to be the cause of the predicate's (or its negation's) satisfaction. > In > >> > other words, you can control one locus of the model indirectly by > >> > modifying > >> > some other locus of it. (It makes sense that human beings would be > >> > preoccupied with flow of control, given our comparative expertise at > >> > controlling our environments relative to other species.) This notion > of > >> > causality is a purely deterministic one, which is in agreement with > most > >> > people's intuitive understanding of causality. Probabilities are > >> > introduced > >> > only when we take into account potential uncertainties, inaccuracies, > or > >> > imprecisions that may be present in the model used for simulation; > they > >> > are > >> > averages or densities of deterministic behaviors over the space of > >> > possibilities. If you start from this deterministic perspective and > then > >> > ask > >> > yourself how you can derive probabilistic models of such a > deterministic > >> > phenomenon under uncertainty, I think the relationships between > >> > causality, > >> > probability, correlation, and flow of time will become fairly evident, > >> > and > >> > it will be much easier to put together a system that reconstructs > >> > underlying > >> > causal relationships from sample data. > >> > > >> > On Tue, Nov 25, 2014 at 5:53 AM, Ben Goertzel via AGI < > [email protected]> > >> > wrote: > >> >> > >> >> Hmmm... > >> >> > >> >> Having thought about this more, while I was indeed traveling > backwards > >> >> in time when I wrote the previous email, it's not too relevant anyhow > >> >> because the Second Law only holds globally, and in complex systems > >> >> there are many subsystems that are behaving anti-entropically. So > I'm > >> >> no sure one can use the law of entropy increase to draw conclusions > >> >> about local causality. > >> >> > >> >> However, I was thinking about section 6.3.2 of > >> >> > >> >> http://cqi.inf.usi.ch/qic/94_Lloyd.pdf > >> >> > >> >> where Seth Lloyd observes that > >> >> > >> >> "Having a common effect does not induce correlation between events, > >> >> while having a common cause does." > >> >> > >> >> I.e. > >> >> > >> >> -- In the case of two causes with a common effect ... there is an > >> >> increase of information from past to future (the probability spread > >> >> across two causes is now concentrated on a single effect). There no > >> >> correlation in the past (between the causes). This is the opposite > >> >> direction of the Second Law of Thermodynamics. > >> >> > >> >> -- In the case of two effects with a common cause ... there is a > >> >> decrease of information from past to future (the probability > >> >> concentrated in one cause is now spread across two effects). There > >> >> is correlation in the future (between the effects). This is in the > >> >> direction of the Second Law of Thermodynamics. > >> >> > >> >> ... > >> >> > >> >> I.e. in many cases the direction of causal influence may be > >> >> identifiable as the direction of increasing correlation.... I'm not > >> >> sure exactly what are the limits of this conclusion though. > >> >> > >> >> ... > >> >> > >> >> Soo -- What if one has two sets of variables, S and T, and there is > >> >> significant mutual information between the values of S and the values > >> >> of T, as evaluated across different cases...? So, suppose we have > >> >> both > >> >> > >> >> S --> T > >> >> > >> >> and > >> >> > >> >> T --> S > >> >> > >> >> in a sense.... But, if there is significantly more correlation > >> >> among the variables within T, than among the variables within S, then > >> >> we can say that it's more likely that T is the effect and S is the > >> >> cause... > >> >> > >> >> The asymmetry used to identify causation is then one of correlation > >> >> rather than of temporality directly... > >> >> > >> >> This may be a way of heuristically inferring causality from > >> >> non-temporal data, if one has a sufficient ensemble of data > samples... > >> >> > >> >> -- Ben > >> >> > >> >> > >> >> On Tue, Nov 25, 2014 at 1:46 PM, Ben Goertzel <[email protected]> > wrote: > >> >> > > >> >> > Hmm, maybe you're right , maybe I was traveling backwards in time > >> >> > when I > >> >> > wrote that ... > >> >> > > >> >> > (More later) > >> >> > > >> >> > On Tuesday, November 25, 2014, martin biehl <[email protected]> > wrote: > >> >> >> > >> >> >> hm, sounds interesting, but I don't get it either. If entropy > >> >> >> increases, > >> >> >> the uncertainty of the state increases and information (about the > >> >> >> state) > >> >> >> decreases as you say, but why would the past then contain more > >> >> >> information > >> >> >> about the future than vice versa? Let X be the past, Y be the > >> >> >> future, > >> >> >> then > >> >> >> as mutual information is symmetric: > >> >> >> H(X) - H(X|Y) = H(Y) - H(Y|X) > >> >> >> now H(Y) > H(X) because of entropy increase. > >> >> >> then > >> >> >> H(Y|X) > H(X|Y) > >> >> >> and the future should be more uncertain given the past than vice > >> >> >> versa. > >> >> >> Where did this go wrong? > >> >> >> > >> >> >> > >> >> >> On Tue, Nov 25, 2014 at 2:13 AM, Ben Goertzel via AGI > >> >> >> <[email protected]> > >> >> >> wrote: > >> >> >>> > >> >> >>> Information is negentropy, so increase of entropy implies > decrease > >> >> >>> of > >> >> >>> information... > >> >> >>> > >> >> >>> Acquiring information about a system is associated with entropy > >> >> >>> production... > >> >> >>> > >> >> >>> On Tue, Nov 25, 2014 at 9:59 AM, Aaron Nitzkin < > [email protected]> > >> >> >>> wrote: > >> >> >>> > Sorry, I must be a little confused -- probably thinking from > the > >> >> >>> > wrong > >> >> >>> > perspective . . . I would think that there is more information > >> >> >>> > in the future about the past than vice versa, because we know > >> >> >>> > more > >> >> >>> > about the > >> >> >>> > past than we do about the future, and also, doesn't > >> >> >>> > increase in entropy imply increase in information (because it > >> >> >>> > requries > >> >> >>> > more > >> >> >>> > information to specify the configuration of a system > >> >> >>> > with higher entropy than the same system with lower entropy?) > >> >> >>> > > >> >> >>> > On Tue, Nov 25, 2014 at 8:27 AM, Ben Goertzel < > [email protected]> > >> >> >>> > wrote: > >> >> >>> >> > >> >> >>> >> In the early part of the paper, the author clarifies that > while > >> >> >>> >> he > >> >> >>> >> assumes "temporal precedence as an aspect of causality" for > >> >> >>> >> simplicity, actually his approach would work with any other > >> >> >>> >> systematic > >> >> >>> >> way of assigning asymmetric directions to relationships > between > >> >> >>> >> events > >> >> >>> >> > >> >> >>> >> I have been thinking a lot about how to infer causality from > >> >> >>> >> non-time-series data (e.g. categorial gene expression data), > and > >> >> >>> >> this > >> >> >>> >> is a case where looking at some other sort of asymmetry than > >> >> >>> >> temporal > >> >> >>> >> precedence (but that may generally correlated with temporal > >> >> >>> >> precedence) seems to make sense. E.g. I've been thinking > about > >> >> >>> >> looking at informational asymmetry: If one has P(A = a | B=b), > >> >> >>> >> one > >> >> >>> >> can > >> >> >>> >> look at whether the distribution for A gives more information > >> >> >>> >> about > >> >> >>> >> the distribution for B, or vice versa. This informational > >> >> >>> >> asymmetry > >> >> >>> >> can be used similarly to temporal asymmetry in defining > >> >> >>> >> causality. > >> >> >>> >> Furthermore, it on the average is going to correlate with > >> >> >>> >> temporal > >> >> >>> >> asymmetry, because the past tends to contain more information > >> >> >>> >> about > >> >> >>> >> the future than vice versa (due to entropy increase, roughly > >> >> >>> >> speaking... but there's more story here...) > >> >> >>> >> > >> >> >>> >> -- Ben > >> >> >>> >> > >> >> >>> >> > >> >> >>> >> On Tue, Nov 25, 2014 at 5:34 AM, Michael van der Gulik > >> >> >>> >> <[email protected]> wrote: > >> >> >>> >> > "Chapter 1. Quantum mechanics... " > >> >> >>> >> > > >> >> >>> >> > It's a nice article; I'll add it to my reading list. > >> >> >>> >> > Prediction > >> >> >>> >> > involves > >> >> >>> >> > working out what causes what, so it's pretty fundamental. > >> >> >>> >> > > >> >> >>> >> > I have a question. Causation in my mind seems to always > >> >> >>> >> > involve > >> >> >>> >> > time, > >> >> >>> >> > and I > >> >> >>> >> > suspect it's impossible to have causation without including > >> >> >>> >> > timing. > >> >> >>> >> > So... > >> >> >>> >> > > >> >> >>> >> > Is it possible for a cause to happen at exactly the same > >> >> >>> >> > moment > >> >> >>> >> > as > >> >> >>> >> > its > >> >> >>> >> > effect? > >> >> >>> >> > > >> >> >>> >> > Is it possible for a cause to happen after its effect? > >> >> >>> >> > > >> >> >>> >> > One instance I'm trying to get my head around is when an > >> >> >>> >> > intelligence > >> >> >>> >> > anticipates a cause (which is an event in the future), which > >> >> >>> >> > results > >> >> >>> >> > in > >> >> >>> >> > the > >> >> >>> >> > intelligence acting such that the effect occurs before the > >> >> >>> >> > cause. > >> >> >>> >> > Perhaps > >> >> >>> >> > the anticipation itself is the causal event. > >> >> >>> >> > > >> >> >>> >> > Regards, > >> >> >>> >> > Michael. > >> >> >>> >> > > >> >> >>> >> > > >> >> >>> >> > On Sun, Nov 23, 2014 at 7:17 AM, Ben Goertzel > >> >> >>> >> > <[email protected]> > >> >> >>> >> > wrote: > >> >> >>> >> >> > >> >> >>> >> >> I just happened across this 2011 paper on the probabilistic > >> >> >>> >> >> foundation > >> >> >>> >> >> of causality, > >> >> >>> >> >> > >> >> >>> >> >> > http://philsci-archive.pitt.edu/9729/1/Website_Version_2.pdf > >> >> >>> >> >> > >> >> >>> >> >> which seems to carefully clarify a bunch of issues that > >> >> >>> >> >> remain > >> >> >>> >> >> dangling in prior discussions of the topic > >> >> >>> >> >> > >> >> >>> >> >> It seems to give a good characterization of what it means > for > >> >> >>> >> >> "P > >> >> >>> >> >> to > >> >> >>> >> >> appear to cause Q, based on the knowledge-base of observer > O" > >> >> >>> >> >> > >> >> >>> >> >> -- > >> >> >>> >> >> Ben Goertzel, PhD > >> >> >>> >> >> http://goertzel.org > >> >> >>> >> >> > >> >> >>> >> >> "The reasonable man adapts himself to the world: the > >> >> >>> >> >> unreasonable > >> >> >>> >> >> one > >> >> >>> >> >> persists in trying to adapt the world to himself. Therefore > >> >> >>> >> >> all > >> >> >>> >> >> progress depends on the unreasonable man." -- George > Bernard > >> >> >>> >> >> Shaw > >> >> >>> >> >> > >> >> >>> >> >> -- > >> >> >>> >> >> You received this message because you are subscribed to the > >> >> >>> >> >> Google > >> >> >>> >> >> Groups > >> >> >>> >> >> "Artificial General Intelligence" group. > >> >> >>> >> >> To unsubscribe from this group and stop receiving emails > from > >> >> >>> >> >> it, > >> >> >>> >> >> send > >> >> >>> >> >> an > >> >> >>> >> >> email to > >> >> >>> >> >> > [email protected]. > >> >> >>> >> >> For more options, visit https://groups.google.com/d/optout > . > >> >> >>> >> > > >> >> >>> >> > > >> >> >>> >> > > >> >> >>> >> > > >> >> >>> >> > -- > >> >> >>> >> > http://gulik.pbwiki.com/ > >> >> >>> >> > > >> >> >>> >> > -- > >> >> >>> >> > You received this message because you are subscribed to the > >> >> >>> >> > Google > >> >> >>> >> > Groups > >> >> >>> >> > "Artificial General Intelligence" group. > >> >> >>> >> > To unsubscribe from this group and stop receiving emails > from > >> >> >>> >> > it, > >> >> >>> >> > send > >> >> >>> >> > an > >> >> >>> >> > email to > >> >> >>> >> > > [email protected]. > >> >> >>> >> > For more options, visit https://groups.google.com/d/optout. > >> >> >>> >> > >> >> >>> >> > >> >> >>> >> > >> >> >>> >> -- > >> >> >>> >> Ben Goertzel, PhD > >> >> >>> >> http://goertzel.org > >> >> >>> >> > >> >> >>> >> "The reasonable man adapts himself to the world: the > >> >> >>> >> unreasonable > >> >> >>> >> one > >> >> >>> >> persists in trying to adapt the world to himself. Therefore > all > >> >> >>> >> progress depends on the unreasonable man." -- George Bernard > >> >> >>> >> Shaw > >> >> >>> >> > >> >> >>> >> -- > >> >> >>> >> You received this message because you are subscribed to the > >> >> >>> >> Google > >> >> >>> >> Groups > >> >> >>> >> "opencog" group. > >> >> >>> >> To unsubscribe from this group and stop receiving emails from > >> >> >>> >> it, > >> >> >>> >> send > >> >> >>> >> an > >> >> >>> >> email to [email protected]. > >> >> >>> >> To post to this group, send email to [email protected] > . > >> >> >>> >> Visit this group at http://groups.google.com/group/opencog. > >> >> >>> >> For more options, visit https://groups.google.com/d/optout. > >> >> >>> > > >> >> >>> > > >> >> >>> > >> >> >>> > >> >> >>> > >> >> >>> -- > >> >> >>> Ben Goertzel, PhD > >> >> >>> http://goertzel.org > >> >> >>> > >> >> >>> "The reasonable man adapts himself to the world: the unreasonable > >> >> >>> one > >> >> >>> persists in trying to adapt the world to himself. Therefore all > >> >> >>> progress depends on the unreasonable man." -- George Bernard Shaw > >> >> >>> > >> >> >>> > >> >> >>> ------------------------------------------- > >> >> >>> AGI > >> >> >>> Archives: https://www.listbox.com/member/archive/303/=now > >> >> >>> RSS Feed: > >> >> >>> https://www.listbox.com/member/archive/rss/303/10872673-8f99760d > >> >> >>> Modify Your Subscription: > >> >> >>> https://www.listbox.com/member/?&; > >> >> >>> Powered by Listbox: http://www.listbox.com > >> >> >> > >> >> >> > >> >> > > >> >> > > >> >> > -- > >> >> > Ben Goertzel, PhD > >> >> > http://goertzel.org > >> >> > > >> >> > "The reasonable man adapts himself to the world: the unreasonable > one > >> >> > persists in trying to adapt the world to himself. Therefore all > >> >> > progress > >> >> > depends on the unreasonable man." -- George Bernard Shaw > >> >> > > >> >> > >> >> > >> >> > >> >> -- > >> >> Ben Goertzel, PhD > >> >> http://goertzel.org > >> >> > >> >> "The reasonable man adapts himself to the world: the unreasonable one > >> >> persists in trying to adapt the world to himself. Therefore all > >> >> progress depends on the unreasonable man." -- George Bernard Shaw > >> >> > >> >> > >> >> ------------------------------------------- > >> >> AGI > >> >> Archives: https://www.listbox.com/member/archive/303/=now > >> >> RSS Feed: > >> >> https://www.listbox.com/member/archive/rss/303/23050605-2da819ff > >> >> Modify Your Subscription: > >> >> > >> >> https://www.listbox.com/member/?&; > >> >> Powered by Listbox: http://www.listbox.com > >> > > >> > > >> > >> > >> > >> -- > >> Ben Goertzel, PhD > >> http://goertzel.org > >> > >> "The reasonable man adapts himself to the world: the unreasonable one > >> persists in trying to adapt the world to himself. Therefore all > >> progress depends on the unreasonable man." -- George Bernard Shaw > > > > > > > > -- > Ben Goertzel, PhD > http://goertzel.org > > "The reasonable man adapts himself to the world: the unreasonable one > persists in trying to adapt the world to himself. Therefore all > progress depends on the unreasonable man." -- George Bernard Shaw > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/10872673-8f99760d > Modify Your Subscription: > https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
