Sometimes very active discussions about peripheral issues overwhelm a thread, so this is a permanent home for those conversations.
I’ve opened a new “Sandbox” thread as a post as the new “ignore commenter” plug-in only works on threads started as posts.
https://arxiv.org/pdf/1803.03453.pdf
https://aiweirdness.com/post/172894792687/when-algorithms-surprise-us
relevant…
Normally, computation is defined in terms of Turing machines. No, I do not tie this to GOFAI.
Start with data.
Apply some sort of rules based process
Get output data.
The problem for all of this, is that there is no data to begin with.
So a photon strikes a retinal cell. Maybe that counts as data. But it is, at most, data about the retinal state. It is not data about the external world.
The problem for an AI system, is getting data about the world.
I would guess that those trying to build AI systems are using gyroscopes, gps tranducers and similar to get data about the world. But the human body has neither gyroscopes nor gps transducers.
The principle problem that needs to be solved, is the problem of getting data or similar, and without gyroscopes or gps transducers. What’s needed is a completely self-contained way of getting data about the external world. This is, of course, the same thing as the problem of perception. Solve that problem, and you will likely understand consciousness. At least that is what works for me. And you may find that computation is not particularly helpful in solving that problem.
I see this as a philosopher’s problem, not a reverse-engineering problem. But professional philosophers don’t seem interested in considering it.
I don’t see how you can believe anything like that last sentence about philosophy. But, I’ll leave it to others to probe.
I’m not sure what you mean by a self-contained; for me, evolutionary heritage, interaction with the world, and interaction with other people are all part of understanding human perception and cognition.
“Reverse engineering” was just my way of saying we need to do science to understand how babies develop into adults successfully interacting with the world. Philosophy can be part of the discussion of what concepts and models are successful for doing that science and why they are so, as long as that philosophy is consistent with science.
So eg Predictive Processing and Bayesian modeling in general is one research program for perception and learning which philosophers examine in that way. Innate hyper priors are the math in those research programs for innate concepts/learning processes.
I would agree philosophy is the domain for discussions of what successful scientific theories and models tells us about reality. That applies equally to physics (eg quantum entities, spacetime) as well as to the cognitive sciences (eg perception, mental representation)
It is all implicitly theistic and dualistic. And yes, I did use the “Look inside” at Amazon before making that assessment.
To explain further: Gopnik is implicitly imposing her own way of seeing the world on the baby. So it is implicitly theistic, with the mother as the implied god. And for that way of understanding the baby to be applicable, there would need to be some way for those ideas (from the mother) to be communicated to the baby — call that a spiritual soul.
(1) it is obvious.
(2) you do not see it.
So I guess it is well hidden. It is hidden in plain view. You do not see it, and philosophers do not see it, because you take too much for granted.
I see it, because I have carefully avoided taking anything for granted.
There’s an example of the problem. You cannot do predictive processing and Bayesian modeling unless you already have data. But until the data problem is solved, there is no data.
Philosopher’s won’t look at this. You criticize me for saying that about philosophers. But you, too, won’t look at this.
I don’t understand the problem of getting data about the real world.
The generic problem of living is doing whatever is necessary to survive.
One possibility is to assume that whatever is needed will come to you.
Another possibility is you will need to move about scholastically.
Or move toward a resource, or move away from something noxious.
Or take action to evade or dissuade an enemy or predator.
Such actions are within the capabilities of single celled organisms.
One can ramp up the complexity, but these are the basic modes of behavior.
I don’t understand the concept of self-contained, nor it’s relevance. I do think living things are fully integrated, and that there are no “sensors” that are distinct from awareness. Even when we can replace sensory components with implants.
Most philosophy is posed in terms of propositions about the world.
Yes, and I’m fine with that. But that’s not the language most people use to talk about questions of knowledge and cognition.
That’s automatic if you are looking at things in terms of doing whatever is needed to survive. So this isn’t an issue for you.
It’s no secret that I am uncomfortable with most philosophy. I admit I haven’t studied deeply, but I have reasons.
There are things I haven’t worked at, such as musical performance, but I can easily tell who are good and who aren’t. There are matters of taste, but in general, one can discern competence.
In philosophy I have difficulty understanding the standards for competence. I can recognize verbal fluency, but cannot judge quality of thought. But more importantly, I have difficulty seeing the value. I don’t wish to disparage other people’s vocations, but I don’t get it.
What I mostly don’t get is the assumption that questions of theology, free will, and such are even answerable, or that some stances are better than others.
In music there are dozens of musical scales and styles, and most people think another culture’s melodies are noise, but it is just entertainment. If it isn’t enjoyable, it isn’t important.
Philosophy aspires to be important. People talk about making life decisions or political decisions based on philosophizing. I don’t find this a useful approach.
Except for the rather narrow field of logic that deals with actions and consequences, which I see as science or scientific.
That seems reasonable.
As best I can tell, the standards of philosophy come from tradition, though the traditions themselves also evolve over time.
https://getpocket.com/explore/item/the-great-ai-paradox?utm_source=pocket-newtab
Thanks, walto.
That actually seems like a pretty good assessment. And I agree with much of what O’Reilly says. I’m skeptical of Tegmark’s viewpoint.
Interesting how hypothetical AIs tend to solve problems the way we would if we were dictators.
I play this game in my head. My AI is a common carrier, like the phone company, that offers everyone a common core of good advice and information. The intellect equalizer.
I just don’t have enough data. Specifically, I have no reliable idea what you are talking about when you say “measurements” and “conventions” and “data”. You seem to me to use those words to mean things only you can understand, and then you complain when no one else gets your viewpoints.
If you have ideas that will revolutionize the field, why don’t you publish them? I have asked them in the past, and I recall you replying something like you tried by no one could understand your position.
“I … cannot judge the quality of thought”
“I have difficulty seeing the value”
” I don’t get [why] … some stances are better than others”
Do you think there might be some kind of correlation there?
I’m not claiming that philosophy has the same potential material value to society that applied science does or even that it has the same value that open-ended scientific research does. Nor that philosophical progress can be measured in the same way scientific progress can.
A better comparison might be pure mathematics. I sure am unable to judge the quality of thought or see why some stances (eg proof approaches) are better than others. But I have no qualms in accepting the value of pure mathematics.
Perhaps you will rightly claim that some pure mathematics is later recognized to be of value to science and that justifies all of it. The first problem with that is that the bulk of it does not have later scientific value
But a much more important issue is this: why should we use value to science as a the only metric for measuring value to society? Is it not helpful to have citizens with a well-rounded education, including understanding of how great minds have thought about free will, or morality, what is the best type of government, and how to live a meaningful life?
Bruce, to Neil:
I think Neil wants to be misunderstood. It’s easier on the ego than being clear and having his ideas rejected.
I publish some of it on my blog.
I’ve tried to publish in journals, but I never get very far. The editors appear to think that I am obviously wrong, although they never say that. And they are never able to tell me where I went wrong.
Quite the contrary. I would love to have somebody tell me where I went wrong. I would love to have people understand me well enough to be able to point out mistakes. I really want that feedback. But I never get it. And note that criticisms by keiths miss by miles (or even by light years).
Pretty much the entire world seems committed to theistic and dualistic thinking, while often asserting that they are not theist and not dualist. That is to say, their way of thinking about the world makes assumptions that only makes sense if theism/dualism is implicitly assumed.
I view theology and philosophy as advancing, but at really slow paces, measured in centuries.
As for contributions to politics, philosophy has given us democracy, capitalism, fascism and communism. The problem remains that I see no way to judge the quality of philosophical thought, nor any way to form an effective feedback loop to correct political philosophy. The lack of iterative testing and revising seems to be related to the slowness of progress.
From my point of view, politicians adopt the wrong set of oughts. Political incentives are perverse. Politicians are rewarded for managing problems rather than eliminating them.
Philosophy, like music and art, contributes to life, and in the same way. Ideas can be beautiful. I just don’t see the Big Questions leading to Big Answers.
Theology and philosophy are parasitic on science. They advance only because science advances and they try to avoid being too inconsistent with science.
If you insist on using “theism” and “dualism” in a highly idiosyncratic way, different from how everyone else uses those terms, and you don’t put the time and energy into explaining how you are using those terms and why that use makes sense, you don’t have much grounds for complaint that you’re misunderstood or for claiming that you’re the only one who really understands those concepts.
KN, to Neil:
Right. And those two terms are just the tip of Neil’s redefinitional iceberg.
I find Neil’s posts mostly interesting, and generally less confused than the norm.
I do try to get under the skin of posts that i disagree with. I have no trouble understanding where creationists and IDists are coming from, but I don’t think it is worth the effort. Ideas can be internally consistent without being interesting to me.
I find it ironic that people who deride god concepts for being vacuous (not falsifiable) are willing to spend hours on the question of free will. Or other philosophical quandaries.
To be clear: I don’t object to Neil that he conceptualizes a cognitive system as informationally closed (there is no information outside the system that get inside of it). That was Humberto Maturana’s view, and he’s one of the best neo-cyberneticists of the 20th century. A rather similar position has been developed by Daniel Hutto and Erik Myin in their books.
What I object to is (1) Neil’s insistence that any rejection of the informational closure of a cognitive system is tantamount to theism and/or dualism and (2) his insistence that this is just obvious and that everyone is confused except him, even though he cannot convince anyone else that he’s right.
Granted, (2) is pretty standard for the Internet, so he’s not an outlier as far as that goes.
I think the word information is pretty fogged up.
Seems to obscure meaning rather than clarify it.
Neil,
Why not take one of your rejected papers and post it here at TSZ?
Nobody listens to me, either 🙁
Philosophy, maybe but theology? Seems to me it’s been on a gentle retreat since the Enlightenment. Even in the USA if the recent Pew survey is to be believed.
By the political standard, laws against criticizing religion are spreading into western countries.
By the standard of rationality, religions are tending to conform to the findings of science. Even fundamentalists are tending to find accommodations. Both sides can refuse to see this trend.
The US constitution and the subsequent development of the US government are an example of that kind of testing of political philosophy. Of course both academics and everyone else are learning from it.
I understand you do not see the value of philosophy. But many greater minds than ours have.
I wish I understood you, but I do not think I bear all the reponsibility for that failure.
For me, the issue is that you do not express your ideas by describing how they differ from others working in the fields you disparage. Similarly, when you use terms, you do not describe how your usage differs from standard scientific or philosophical usage.
Instead of taking the time to engage seriously with contemporary philosophy or science, you summarily dismiss it.
I think knowledge is the result of an inquiring community following proven practices. Solitary thinkers have ideas that may turn out to be knowledge when those ideas are subject to that process. But the ideas may also turn out to be variations of existing knowledge or simply nonsense. There is no way to tell which category applies to your ideas.
How does that differ from Hohwy’s “inferentially secluded and neurocentrically skull‐bound” version of Friston’s ideas?
https://onlinelibrary.wiley.com/doi/abs/10.1111/nous.12062
When I try to understand Neil’s ideas, I guess that (semantic) information for him is what organisms create by “measuring” the environment external to their boundaries by interacting with that boudnary
Somehow Gibson needs to be worked in too.
Taking semantic information roughly as a difference that makes a difference to the organism, that idea seems reasonable to me, with PP being one approach to formalizing it.
But when it comes to Neil, truth about his ideas seems to be an ever-receding target for me.
Are you mimicking Neil here?
Just to be clear, I am not blaming you. I am not blaming anyone. I fully recognize that it is my responsibility to attempt to communicate what I know. And thus far, I have not been successful.
My ideas are not built on those of others working in the same area. The differences are so huge, that there is nothing to compare.
I recognize this. In order to communicate, we need shared concepts. So I need to persuade people to develop suitable concepts. And a lot of what I post is an attempt to do that. But I cannot just plant concepts in people’s brains. They have to develop them, which requires work on their part. Unfortunately, I have been unable to persuade anyone to put in the effort.
begin-serious-engagement:
it is mostly bullshit
end-serious-engagement:
Yes, I’m sure you don’t see that as serious. I have actually attempted serious engagement. It never gets anywhere. People are strongly attached to their view of science, and resist questioning it.
One book that I looked at (a long time ago) was Patterns of Discovery. As best I recall, Hanson seriously looked at the possibility that scientific laws are analytic (as opposed to synthetic).
I don’t see that much mentioned these days. The idea that scientific laws might be analytic is usually dismissed out of hand. Philosophers seem more comfortable with Quine’s dismissal of the analyitic/synthetic distinction. Why?
If philosophers of science really understood how science works, then they should see that the best scientific laws should be analytic. But they do not see this.
I’ll put this in perspective.
I am not blaming anyone. I periodically try to get my ideas out, because I sense that I have a responsibility to do that. But maybe the world just isn’t ready to understand cognition. Maybe they never will be. My ideas will probably die when I die. C’est la vie.
That’s about right.
Not really. I mention Gibson to give credit where credit is due. But I am not actually building on Gibson’s work in any important sense.
The problem with PP, is that it depends on a lot of prior knowledge (but not in the form of beliefs) before it can even get started. My concern is primarily with acquiring that prior knowledge.
As far as I know, petrushka has a background in psychology, or something related. And that seems make it easier for him to see some of what I am getting at.
I’ve not noticed that. Which countries are you referring to?
Neil,
And:
Why not post one of your rejected papers at TSZ?
I stopped trying when I retired. And that was 13 years ago.
TSZ won’t reject your paper, and you’ll get the feedback you say you want.
Thank for this long and thoughtful reply.
I don’t know if there is any constructive approach to working through some of the issues you touch on; I think it would have to involve some way to have a common set of concepts for discussion. But those would be the concepts of contemporary philosophy and cognitive science.
On one specific point: KN and Walto would know better, but I don’t think that Quine’s actual viewpoint on analytic /synthetic is a consensus view in philosophy. Rather, what most agree on is the importance of his analysis to the debate. I don’t know what you mean by analytic laws of nature, but in terms of concepts I understand, it reminds me of Carnap’s ideas on logical positivism as well as modern ideas on the role of the a priori in science*
On philosophy in general, this was quoted in this Rebecca Goldstein profile:
“Wigner once quoted somebody—I can’t remember who—to the effect that philosophy is the creation of a jargon which it then proceeds to abuse.”
——————————
*ETA: Besides discussing various philosophers concepts of the a priori and pre-conditions for a scientific research program, Stump’s book also has specific chapters on conventionalism in Poincare, the role of math as a priori in science, and in what sense F=ma is both a priori as well as empirical.
My problem is that I don’t really know why you agree that I have a roughly correct understanding of your ideas. In particular, why you use the word “measuring” with respect to what organisms do rather than what scientists and carpenters do (and do).
If you ever wanted to do an OP on one of your ideas, I think an OP devoted to solely your use of that one concept of measuring in organisms would be a good place to start.
As for PP: my hand waving on that is that PP involves a general learning and action/perception mechanism. PP can be modeled using a Bayesian hierarchy of priors and hyperpriors which are updated through action/perception. Evolution hard codes both the mechanism and the top level of hyperpriors as innate features of organisms.
This paper tries to provide a framework to discuss PP and innateness:
https://link.springer.com/article/10.1007/s11229-017-1427-7
“Bayesianism offers not a vindication of either nativism or empiricism, but one way to talk precisely and transparently about the kinds of mechanisms and representations underlying the acquisition of psychological traits without a commitment to an innate language of thought.”
That could be taken in many ways….
(quote mining intentional!)
I’m thinking about that. Actually, I see measuring as a special case of categorization. So it’s really categorization that would be the issue.
Okay. But I don’t agree at all.
If “learning” is taken to mean “acquiring justified true beliefs”, then maybe PP can be taken to be general learning. But, for me, “learning” means what children do as they develop. It means what I do when I study a new area. And that has very little to do with acquiring justified true beliefs. That we may acquire some justified true beliefs is a side effect of learning. It is not the core of learning. So, to me, PP seems misguided.
I’ll modify an earlier statement. The problem with Bayesian methods, is that they require prior knowledge. Bayesian methods cannot even get started without that prior knowledge. So my concern has been about how we acquire that prerequisite knowledge.
I looked at that. It is currently open in a browser tab. It talks about “psychological traits”, but I have no idea what it means by that. Personally, I don’t have a problem with the idea that psychological traits (what I would mean by that term) are innate. But the paper is talking about something different.
For me, here is the issue of innateness. I grew up in Australia. I have well formed concepts for Australian flora and fauna. My conceptualization of Eurasian flora and fauna is a bit more shaky. My parents and grandparents were also Australian. But if I go back much further, my ancestors were European (Scotland and Germany, I think).
If concepts are part of my innate makeup (from my DNA), then I should do better on European flora and fauna and worse on Australian flora and fauna. So it looks to me as if concepts could not possibly be innate.
Beliefs cannot even be formed without concepts. So the prior knowledge that I am talking about, is knowledge of concepts (or concept formation). From my perspective, learning is concept acquisition, not belief acquisition. Neither PP nor Bayesian methods can get started before there are concepts.
At least, as I read him, Locke’s empiricism is about concept formation. But by the time we get to Hume, empiricism has changed into a story about belief formation.
I would not take it that way at all.
I’d take it instead as a term of art in the science of animal psychology. Perhaps stimulus-response learning is its most basic form — that is just off the top of my head; I’d want to review some animal psychology in more detail to commit.
I interpret a lot what you say about concepts to refer only to the subset of linguistically based concepts, which only we humans have (until Artificial General Intelligence exists). (ETA: But I am not sure if that is what you think, or if that is what you think I think)
Two points on concepts and language
First, when scientists attribute concepts to animals, they have to start with purely behavioral definitions of a concept, eg a disposition to certain behavior in a given context detailed in terms of their niche (de Waal emphasizes danger of laboratory testing ignoring niche).
Second, current linguistics studies embodied language and concepts, grounding both of them in our bodies and our sensorimotor systems. Example: abstract ideas of time and space are based in our embodied experience of movement.
That allows evolutionary continuity. Perhaps it can also be modeled using Bayesian hierarchies. I understand Eliasmith ‘s “semantic pointers” as a similar idea, although he talks in terms of control theory math, not Bayesianism. I have a vague idea that the two are related, but I don’t have the details to support that vague impression.
https://www.psychologytoday.com/us/blog/hot-thought/201306/the-new-synthesis-in-cognitive-science
I’ve only had a quick read of that article. If the article does not define in more detail, I would take “psychological traits” to refer to aspects of the mind studied by the science of psychology.
One way to explore your ideas would be to pick some philosophical or scientific article and have a thread to discuss how to understand it and how you would criticizeit using your concepts. Are you interested in doing that for this article?
Your view of innate concepts seems similar to Fodor’s thinking about innate concepts. He thinks we have a Language of Thought in which all concepts are represented and that we are born with a set of concepts represented in that language. I don’t think that at all — neither the language of thought part nor that innate concepts are linguistic, abstract ones such as elements of flora and fauna*.
Instead, as I recall, the paper I linked talks about our innate(?) perception bias to assuming light comes from overhead to be something that could be encoded innately in a hyperprior. That might be the case; studying that possibility would mean using Bayesian models in understanding human developmental psychology. I’d have to revisit the paper to see if it cites that kind of work.
——————–
ETA: OTOH, some animals may have an innate behavioral concept of eg food sources — at least, that is how we might describe that concept.
I am more inclined to see perceptual learning as the most basic form.
I disagree already.
Concepts are not based on language. If anything, language is based on using concepts that we already have before we acquire language.
But maybe I should call those “categories” rather than “concepts”.
Of course, once we acquire language, we can use that to acquire additional concepts that would be difficult or impossible without language.
We do much the same with humans, except that we include linguistic behavior.
I see this as a big mistake already.
You are talking of attributing human concepts to animals. But why do that? We should expect an animal to have concepts appropriate to its lifestyle, rather than concepts that are suited to us.
Most published articles don’t really fit.
Neil,
That’s faulty reasoning. You’re implicitly assuming that
1) concepts involving the local flora and fauna would fall into the “innate” category, and
2) they should be “implantable” in just a few generations.
Why make those assumptions?
Except that Fodor sees concepts as innate, while I see them as acquired by learning. So, if anything, Fodor and I have opposite views.
I really don’t understand why you, and many others, are so hooked on Bayesian models.***
Bayesian methods cannot account for Newton’s laws. Science is a system of learning that is mostly done in public. So any account of learning has to be compatible with how science progresses.
*** footnote: Actually, I do understand why people look to Bayesian models. It’s because they are stuck in a regime of what I am calling theistic and dualistic thinking.
They want to believe that there is a reality that is independent of us. And I agree that there is such a reality.
But they also want there to be a true description of that reality, such that the true description is independent of us. That is to say, they want there to be a “god’s eye view”. That’s where my theism comment comes from.
But it cannot be. Learning is creative. Cognition is creative. Perception is creative.
Learning is not just copying reality. It is inventing ways of looking at and interacting with reality.
Because these are creative, we do not all create in the same way. Subjectivity exists precisely because we have each created our own ways of looking at the world.
The so-called “hard problem of consciousness” cannot be solved because it presupposes that perception is just a kind of copying. It does not allow for the creativity of perception.
Just fyi I agree with Neil about this much: a lot of cognitive science does want there to be a true description of the world that is epistemically accessible to us independent of the cognitive system that we’re describing, and that this involves taking a god’s-eye view not only on the world but also on ourselves.
For this reason I am very interested in Heinz von Foerster. Foerster was a founder of “neocybernetics” or “second-order cybernetics” in which the lessons of cybernetics are applied to ourselves as we are doing science. This pushes cognitive science away from realism, or better, beyond realism and anti-realism. Enactivism beginning with Maturana and Varela has been developing this insight, and also deepening the connections between enactivist cognitive science and Madhyamaka Buddhism.
What we need is the right sort of overall coherence where the kind of pragmatic realism/anti-realism that is ‘downstream’ of enactive, embodied cognition is compatible with the presuppositions of scientific inquiry.
It would be nice if we could avoid the Lange-Nietzsche Dilemma (as I call it), in which empirical psychology undermines itself because it starts off by assuming realism but ends up entailing idealism.