|
Author
|
Topic: Can intelligence be described scientifically?
|
Erik
Member
Member # 160
|
posted 22. February 2003 16:25
I'm not sure if this board may be used as an ask-an-ID advocate forum, but I'll try and see if my post stays or disappears.
To a good approximation, science is all about finding maximally accurate, general and simple descriptions of observed and unobserved data (a description of unobserved data is a prediction). For instance, Ohm's law, which relates the voltage over a resistor to the current flowing though it, is a good scientific model because it allows us to compress long lists of observed voltages and currents through a resistor to a single parameter value (called the resistance). Other considerations are probably also involved, but to a good first approximation science theorizing is just a sophisticated excercise in data compression and curve fitting.*
Just as one can describe the behaviour of steam engines, the planets in our solar system, gene frequencies in biological populations, gene expression in a developing organism, etc., it seems that one should also be able (at least in principle) to describe the behaviour of intelligent agents. Yet ID advocates seem to insists on a mystified view, where intelligence is described by words like "non-mechanistic" and is considered fundamentally different from the rest of the universe. And despite similar mystifications of computers and information technology (which, despite the fact that any such system can be described by a sufficiently complicated finite-state machine**--if digital--or a system of differential equations--if analog--are considered teleological and fundamentally different from other types of systems), it is insisted that AI will never be genuine intelligence.
My question is: Exactly what is the ID position on intelligence? Can intelligence be described by a sufficiently sophisticated differential equation, finite-state machine or any other kind of descriptive model (deterministic or not)? Or is intelligence considered to be a process which cannot be described in any more compact way than simply listing all our observations about it? To clarify, I'm not asking about our current state of knowledge, but about what ID advocates think is possible in principle and with future knowledge.
Erik
* For the record, I do not adhere to so-called "methodological naturalism", which is either question-begging (if "natural" is defined as "precisely that which is amendable to scientific study"), poorly expressed or false when considered as a dividing line between science and non-science. Science can be done (and is done) without any ontological commitments beyond the assumptions that an external world exists and that there are patterns to be discovered in our observations of this external world.
** For those who don't know, a "finite-state machine" is essentially a flow-chart. A vending machine is a typical non-abstract example of a finite-state machine.
[Edit: One of the two tactless terms was removed.]
[ 22. February 2003, 17:42: Message edited by: Erik ]
IP: Logged
|
|
Moderator
Administrator
Member # 1
|
posted 22. February 2003 17:00
Erik, I think that this is a perfectly legitimate thread. My only concern is the use of the "romanticized and mystified" terminology; the point could have been made more tactfully. Otherwise, I think your question is ripe for discussion.
[ 22. February 2003, 17:07: Message edited by: Moderator ]
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 22. February 2003 17:25
I posted a question about the definition of "information" that seems highly relevant to this discussion also, on the thread Shapiro on the Genome that Eric referred to.
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 26. February 2003 16:25
Excellent question, Erik! How would you describe “intelligence”?
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 26. February 2003 17:12
Here I'll expand a little on my first post by providing two examples of how intelligence has been (crudely) studied scientifically. For the purposes of this discussion I am not particularly concerned with whether the conclusions in these studies are well-supported, but rather whether or not the basic approach is considered in line with the views of ID advocates.
quote:
Wakeling J. & Bak P. (2001) "Intelligent systems in the context of surrounding environment", Physical Review E, 64:051920
Abstract: We investigate the behavioral patterns of a population of agents, each controlled by a simple biologically motivated neural network model, when they are set in competition against each other in the minority model of Challet and Zhang. We explore the effects of changing agent characteristics, demonstrating that crowding behavior takes place among agents of similar memory, and show how this allows unique "rogue" agents with higher memory values to take advantage of a majority population. We also show that agents' analytic capability is largely determined by the size of the intermediary layer of neurons. In the context of these results, we discuss the general nature of natural and artificial intelligence systems, and suggest intelligence only exists in the context of the surrounding environment (embodiment).
Wakeling & Bak argue that "intelligence" is essentially the same as the level of success with which an agent reacts to stimuli. As such, "intelligence" only makes sense given a particular environment. So if someone asks "how intelligent is agent A?" the proper response is, according to the authors, "it depends on the environment".
quote:
Zak M. (2000) "Dynamics of Intelligent Systems", International Journal of Theoretical Physics, 39(8):2107-2140
Abstract: Recent advances in nonlinear dynamics demonstrate a remarkable complexity of patterns outside of equilibrium, which are derived from simple basic laws of physics. A class of mathematical models has been identified providing a variety of such patterns in the form of static, periodic, or chaotic attractors. These models appear to be so general that they predict not only physical, but also biological, economic, and social patterns of behavior. Such a phenomenological reductionism may suggest that, on the dynamical level of description, there is no difference between a solar system, a swarm of insects, and a stock market. However, this conclusion is wrong for a very simple reason: Even primitive living species possess additional non-Newtonian properties which are not included in the laws of Newtonian or statistical mechanics. These properties follow from a privileged ability of living species to possess a self-image (a concept introduced in mathematical psychology). In this paper we consider the existence of a self-image as a postulate to be added to classical physics for modeling behavior of living systems.We show that self-image can be incorporated into the mathematical formalism of a nonlinear dynamics which evolves in probability space. We demonstrate that one of the basic invariants of living systems is their ability to predict the future, which is associated with intelligence.
Zak M. (1999) "Physical invariants of biosignatures", Physics Letters A, 255:110-118
Abstract: Biosignature is one of the most important evidences of life available to researchers. However, many complex physical and chemical phenomena can mimic prints of life so closely that special methods are required to make the distinction. In addition to that, life, in principle, can be composed of components which are fundamentally different from those known on Earth. That is why identification of biosignatures should be based upon some phenomenological invariants. Such invariants, within the framework of Newtonian formalism, are introduced and discussed.
Control theory is a theory of how to best control a system's output signal by giving it appropriate input signals. That is, an input variable and a model of the system's response to different input signals are given and the objective is to find out which input signal to apply in order to get certain output. It can be shown that "Adaptation Implies Internal Model", i.e., in an abstract sense, a successful regulatory systems contains a model of the system and signals it regulates. (I'm not sure how interesting that result is, because regulation can often be successful even when it is based on a quite poor model of reality, but it is suggestive.) The above articles by Zak are somewhat hard to follow, but I think the basic message is similar in the sense that he argues that one of the defining characteristics of living systems is an internal model of the external world. He provides a toy model for describing some process associated with intelligence (although I must admit that I am not sure exactly which processes are being modelled--I'll need to reread the articles).
To sum up, I have provided two specific examples, one in which intelligence is treated (or concluded to be) something that only makes sense in a given environment and another in which the dynamics of intelligent systems is described by a mathematical model. Is there, according to ID advocates, something objectionable about these approaches? If so, how ID advocates suggest that intelligence is studied and captured in scientific models?
Erik
PS. The article linked to in my signature may also be of some interest! DS.
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 28. February 2003 13:19
I’m not an “IDer,” Erik, but if you’ll permit, I have two very basic questions: What is a “model” and what is an “environment”? Since much of our research seems to be based on the intuitive notion of the relation of models to environments, it might be helpful to understand what is intended by those two terms.
Much research in the cognitive sciences, very broadly, seems to presume that “intelligence” is “model” or “representation” based. The science then becomes modeling models of models. The whole enterprise becomes a sort of strange foray into the infinite regress. LOL
The idea becomes a lot more vague and implicit in biology. Take the notion of “environment:” Does an environment begin outside a cell wall or membrane, e.g.? Or does what is inside the cell count as an environment? Is there a way to non-arbitrarily bound an environment for the purposes of analysis? A genome “lives” in some immediate “environment” that is largely of its own making. And often we consider that “intelligence” stands in some interactive relation to its environment—it is not merely shaped by its environment, but shapes its environment. Does this mean that a genome is definitively “intelligent”?
(Also, the closely related concept of “adaptation” is often identified with intelligence. But theorists sometimes seem to exclude arbitrarily “evolutionary adaptations.” I’ve never been able to figure out why exactly this is an important restriction for them, but I suspect its so we can say that “evolution” is not an “intelligent” process. Whatever!)
I think both terms, model and environment, stand in interesting relation to what one of the authors you cited is talking about. E.g., if models are in some sense “self-referential,” (and I think they are) then the model becomes an element in the environment and the implicit distinction becomes even blurrier. (I believe that Hofstadter practically made a career from this “self-reference” paradox.)
But to be contentious (and “mystifying”) I can imagine, on engineering design principles alone, that organisms do not contain elaborate, detailed models of their environments, but contain only “methods.” E.g., I might subsume the “environment” under a great “uncertainty” and include only a very generalized model of uncertainty, i.e. a “method,” rather than a model strictly speaking, for structuring it and adapting to it. (The “method” then becomes itself a sort of “model.” But models presume methods and methods presume models. So all I keep doing is going ‘round in circles.) This relates to the "bounding of the environment for the purposes of analysis," which is sometimes done in interesting ways in engineering.
Its conceivable to me that this might suffice instead of a model such as biologists seem to imply. I can also see me many advantages to this approach, if I was going to invent an intelligent life form. Models are not necessarily necessary, and sometimes are a positive liability. E.g., if there is a demand to adapt, continuously update a model, a minimized model suggests itself, simply for computational reasons alone. (And also suggests a trade-off: representation/computation.)For the very same reason, if we require a model of some level of complexity we may want it to still be compact, requiring compression, which is one of the reasons why I suspect that there is going to be no straightforward information-theoretic measure of the relation of a model to its environment, even if it is only a model of the environment. Scientists, quite intelligently, do this all the time (minimize and compress).
In design theory the concept “model,” as you might well know or imagine, has been extensively studied and becomes a little more subtle, esp. with the “internal model principle,” e.g. The principle does not require that the model be a sort abstract representation (like a program), but can be incorporated directly into actual processes and structures, if you know what I mean. Designs may not contain “models” but they always are models.
As a general class, and in very broad terms, “intelligence” seems to be essentially non-linear, involves complex mappings or processes, is highly heterogeneous and changeable in composition, is embedded in a rich time domain, unpredictable at some level, unbounded (See Burtt’s The Metaphysical Foundations of Modern Science for an intriguing argument), etc., and therefore for all these reasons those aspects of it that are really interesting must be to some extent inaccessible, at present, to science.
I believe this is the very reason for the “mystification” you mention (and I commiserate completely, while at the same time, no doubt, contributing to the “mystification”)--of which scientists are themselves the principal purveyors! As Ashby might say, “mystification” is a way of coping with complexity. LOL
It might be helpful to suggest, as some AI researchers do, with some somewhat axiomatic “skills” or “attributes” set and work from there. What is it that we think is minimally (definitively) “necessary” for “intelligence”? What would we identify as intelligent behavior, e.g.?
Just some thoughts from the “non-ID” perspective.
(I also read recently a critique of “agent,” i.e., some system that relates to its environment according to a model, and I’ll see if I can dig it up and link to it.)
IP: Logged
|
|
gedanken
Member
Member # 594
|
posted 28. February 2003 15:07
quote:
Much research in the cognitive sciences, very broadly, seems to presume that “intelligence” is “model” or “representation” based. The science then becomes modeling models of models. The whole enterprise becomes a sort of strange foray into the infinite regress. LOL
Janitor, you seem to have taken something I said in another thread to heart, and extended it too far.
I have a veiw of “modeling” that I find is a little different from a that of many scientists and presenters on ARN and ISCID. Many that I have talked to seem to make a distinction between an equation such as F=M*A from physics, and a complex representational model such as a stellar evolution model.
I consider F=M*A to be a “model” because it (along with the description of how to apply the equation and additional aspects like F=G*m*M/r^2) forms a description of behavior that can be simulated. Curves can be derived from this “model” and checked with real-world events such as trajectories of artillery shells or apples, and planetary motions. What we note is this is an abstraction of how we think the physical world operates. We don’t include every detail -- for example a simple use of the model may ignore friction and other bodies like far away planets, moons, solar systems, galaxies, etc., becoming successively less influential and less relevant. Various degrees of modeling can take into account greater degrees of the complexity of the real world.
But an aspect of all models is that they are not perfect descriptions that embody all cases and all conditions. We recognize that the classical model F=M*A is not only incomplete in that it is difficult to model all classical forces, but furthermore we have learned that new models, like quantum mechanics and relativity, have better conformance with reality in greater numbers of cases. A key point is that the new model (e.g. QM, relativity) does not invalidate the old model when the old model has significant applicability in a large range of applications. We recognize that both the old and the new model are incomplete -- that they are necessarily Western reductionist abstractions and not completely accurate and full descriptions of reality.
So for me a model is any (necessarily) simplified abstract description of how a portion or aspect of the real world operates. A usual characteristic of models is that they are testable by observation of reality, at least in controlled circumstances.
Other models, like evolutionary models have relationships, just as the F=M*A relationship specifies interrelatedness of parts of the model. But these relationships may be less mathematical or deterministic. For example “descent with modification” is a “model” in my view, because it describes a relationship (when described with a body of detail of what the term means and how to apply it). It is a very abstract model, and much more detailed models have been formed to get at greater detail of how evolutionary processes seem to work in real nature. Here the relationships may or may not be specified in mathematical terms (they often are in evolution models), however they may be in the form of constraints or of stochastic nature. We note that the QM model of physics is of a stochastic nature, for example, though the relationships are more predictive of exact behavior than are some evolutionary models. Other evolutionary models restrict their domain to very narrow and focused aspects of nature, and may in fact have a greater numerically focused predictivity than some example of a QM case in which events can vary widely.
Now the case of modeling of models that Janitor mentions is simply a case of a greater degree of abstraction of reality. Both are exctly the same character, they are approximations of how a part (abstract) of reality can work. Often they may be derived by logical argument of the consequences (such as by mathematical proof or calculation) of some aspect of another model. This does not, as I mentioned, change their status as approximations and necessarily incomplete.
There is no infinite regress because there is no intention of scientists to take these levels of abstraction to inifnite levels -- though in fact there are an infinite number of greater levels of abstraction possible in any model! I think that some ID promoters have in fact been responsible for the appearance of a regress of abstraction by claiming that certain properties apply (such as “No Free Lunch theorems”) and thus those properties must be true of the physical reality that the model was attempting to describe. The failures to conform to logical consistency that might be described in such an application of logic (e.g. proof from “NFL” theorem) does not imply that reality is wholly inconsistent with aspects of that model precisely because all models are incomplete. For example the application of NFL theorems may indeed show that models that conform to the predicates of the NFL theorems must necessarily not exhibit characteristics that the NFL theorems show must be absent. That does not mean that evolutionary models must therefore exhibit those conditions, because other evolutionary models may not fall under the predicates of the NFL theorems, and thus they do not apply to these other models. (And in fact the class of models, for example that NFL theorems apply to, might be extremely small and irrelevant.)
An aspect of this that I think is important:
Any “model” of intelligence must be viewed as one model, and not as the model. I think that intelligence has been shown to have a great variety of different expressions, and thus will defy a single description (or model). Thus the discussion should move to classes of descriptions or models, rather than trying to define the definition or model of intelligence.
These different classes may have different logical properties. For example some may distinguish between possible activities of “forethought”, wherein an evolutionary model builds then tests, a mental computational system may “model,” then simulate, and “test” in simulation, and only then test in the physical by build and test. The distinction of forethought may distinguish kinds of intelligence, or whether a process is “intelligent”.
[ 28. February 2003, 15:09: Message edited by: gedanken ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 01. March 2003 17:55
Janitor@MIT, since I don't want to take this thread too far off track (it's intended as an inquiry about ID advocates' view of intelligence, as opposed to a presentation of my, or the mainstream science, view) I will only reply in brief:
1. "How would you describe “intelligence”?". Since I do not share the ID movement's view of intelligence as an either-or property fundamentally distinct from what can be found in the "natural world", I am not in the same strong need of a scientific model of intelligence in order to do science. The ID movement relies on the existence of a sharp distinction between intelligence and non-intelligence in a way that mainstream science does not.
Just as I think the concept itself is fuzzy, I also think the real-world phenomena to which it refers are fuzzy. I think there are more than one phenomena which is called "intelligence" and that there are entire spectra between "intelligence" and "non-intelligence". Therefore I think a proper scientific study of intelligence would introduce distinctions between different types of "intelligence", collecting data about each, and trying to capture the patterns discovered in the data in scientific models.
2. I mean "model" in the statistics sense. A model is a (possibly perfect) approximation of a system's input-output relation*. Consider, for instance, a resistor. The input is the current flowing through and the output is the voltage over it (or the other way around, if you prefer!). Ohm's law--the statement that the output is proportional to the input--is an approximation of the real behaviour of the resistor and is thus a model.
To say that a "system contains a model of X" is to say that the system contains a subsystem which is a model of X.
3. How we divide the universe into agents/systems/environment is sometimes arbitrary and at other times there is a single natural way to do it. Exactly how the partition is done depends on exactly what we are trying to study. I'm afraid whatever point you were trying make about this did not get through to me.
Erik
* "Input" and "output" should be taken very generally. Any (set of) observable variable(s) can qualify.
IP: Logged
|
|
|
|
gedanken
Member
Member # 594
|
posted 03. March 2003 00:06
I posted a summary and analysis of "information" definitions on Shapiro on the Genome thread, and some relations to associations of communications theory to biology. This relates somewhat (but not directly) to the definition of intelligence from ID perspective, because consideration is given to Dembski's definition of "information".
I'm not saying direct relevance, just tangential.
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 03. March 2003 00:40
Erik:
You wrote:
Quote: "1. ... Since I do not share the ID movement's view of intelligence as an either-or property fundamentally distinct from what can be found in the "natural world", I am not in the same strong need of a scientific model of intelligence in order to do science. The ID movement relies on the existence of a sharp distinction between intelligence and non-intelligence in a way that mainstream science does not."
It seems to me this is circular. If you don't need a model of intelligence to do science, then ID is false. I would think a better way to put it would be:
"1. ... Since I do not share the ID movement's view of intelligence as an either-or property fundamentally distinct from what can be found in the "natural world", I __*believe* that I__ am not in the same strong need of a scientific model of intelligence in order to do science."
As for your initial question about how ID would model the "I" in "ID", the question stems from a misconception about ID. ID is not so much a new way of doing life science as merely a recognition of (i) how life science has always been done, and (ii) the evidence we have from the life sciences.
(i)
I'll repeat this simple example. When Linus Pauling was theorizing about the structure of DNA, he hypothesized the nucleotides to be sticking out radially for easy access and reading, rather than sticking inward as we now know. This is the design inference at work and it is used all the time in the life sciences. In this case, the scientist was wrong.
So when you ask, how does ID model the "I", there is no new, clever approach that ID is proposing beyond what has always been done.
(ii)
ID is a recognition of the evidence we have from the life sciences which indicates that the theory of evolution is not sufficient to explain what we know.
--Cornelius
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 03. March 2003 12:02
quote:
So when you ask, how does ID model the "I", there is no new, clever approach that ID is proposing beyond what has always been done.
(ii)
ID is a recognition of the evidence we have from the life sciences which indicates that the theory of evolution is not sufficient to explain what we know.
Is that all? I thought that ID was meant to provide a non-materialistic alternative to science not merely the finding that the present theory of evolution is not sufficient. That's hardly surprising in a science which is still uncovering new evidence and proposing new pathways and hypotheses. In fact we already KNOW that the present theory of evolution (RM&NS) is not the only mechanism of evolution, genetic drift for instance does not rely on natural selection. Similarly horizontal gene transfer may or may not be relying on selection.
Science does not need ID to point out that in this age of new scientific methods to explore the genome, new tools and perhaps even additional theories are needed.
Perhaps this is a good issue to explore: Is ID merely a recognition that our scientific understanding of evolution is still evolving? Is this surprising knowing that much of the genome is yet to be explored? Does such a recognition even require ID? Is this the limited role of ID or are there other roles ID may be able to play and what would be required?
Certainly as far as Cornelius' statement about ID is concerned it seems obvious that ID is in that aspect not different from how science operates in that it realizes that our understanding of evolutionary science is still growing. But that does not require ID nor does it provide for an alternative to materialistic/naturalistic science. Certainly it does not seem to "... pursue the theoretical development, empirical application, and philosophical implications of information- and design-theoretic concepts for complex systems."
ID would hardly be as controversial if it merely was limited to a recognition that we don't know everything there is to know yet about evolution.
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 03. March 2003 12:41
Frances:
Frances wrote: "ID would hardly be as controversial if it merely was limited to a recognition that we don't know everything there is to know yet about evolution."
But, of course, this is not ID. ID is a recognition that evolution (the species arising via strictly naturalistic means) is not sufficient; not that we are still refining the theory of evolution.
--Cornelius
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 03. March 2003 13:07
Cornelius wrote quote:
I'll repeat this simple example. When Linus Pauling was theorizing about the structure of DNA, he hypothesized the nucleotides to be sticking out radially for easy access and reading, rather than sticking inward as we now know. This is the design inference at work and it is used all the time in the life sciences. In this case, the scientist was wrong.
This is not the "design inference" as Dembski conceives it, which is an inference to allegedly non-naturalistic causes from probabilistic and eliminative premises. This example is a "design inference" by functional analysis. Roughly, it says that If the function (not "purpose") of the nucleotides implies that they be readily accessed and read by other biochemical entities in the cell, and if there is relative selective advantage for easy availability for that, then it is likely that they're sticking out radially. There's no necessary teleology or big I Intelligence about that reasoning. It is a hypothesis, or conjecture, based on a line of reasoning starting with what Pauling conceived to be reasonable naturalistic functional grounds. The most interesting thing about it is that is was testable: those clever biochemists could actually discover the structure and thus directly test Pauling's conjecture. There is a real empirical research program associated with the question of the structure of DNA that goes on to this day.
To interpret Pauling's reasoning as embodying something more than 'a structure suggested by functional analysis of the hypothesized operating context' is to load it down with connotations it can't bear. It's quite true that functional analyses, with associated conjectures and hypotheses about biological structures and processes, are used all the time in biology. To imply that functional thinking like that entails or presupposes big D "Design," though, is deceptive. Every evolutionist I know is fully aware that biological structures can be described in terms of little d "design." All of them that I know accept - indeed assert - that natural physical processes can produce that little d "design."
Cornelius asserts that quote:
So when you ask, how does ID model the "I", there is no new, clever approach that ID is proposing beyond what has always been done.
If that's the case, then why all the hoorah? Why the talk of transforming science? Why the political engagement to change the teaching of biology in secondary schools? If that's what has always been done, then ID has won; indeed, it has been knocking at an open door all these years.
RBH
[ 03. March 2003, 13:11: Message edited by: RBH ]
IP: Logged
|
|
Erik
Member
Member # 160
|
posted 03. March 2003 14:09
Some points in response to Cornelius G. Hunter:
1. I stand by my original version. Your suggested modifications change nothing in terms of content, but merely transforms a statement about matters of fact into a wimpy statement about my beliefs. Whenever I make a statement that I'm prepared to defend, I ditch wimpy qualifiers like "I believe". They serve no useful function.
2. Mainstream science does not need a scientific model of intelligence, because many--indeed, most--phenomena are studied and described in ways that completely independent of whether the systems being modelled are intelligent or not. Ohm's law holds regardless of whether resistors are intelligent (in some sense) or not. Newton's law of motion holds for, say, cars regardless of whether cars are intelligent (in some sense) or not. There are many phenomena that can be described scientifically without including intelligence as a variable. If you still disagree, I'd like to know precisely how it is circular for physicists to model (e.g.) electrical components without having a model of intelligence.
3. I agree with you that ID is not new and that it is just a piece of negative campaigning against the theory of evolution. However, not everyone agrees with this. As an example, I note that a current front-figure of the ID movement, William A. Dembski, claims to have formulated a statistical method for inferring that an event was due to intelligent design. Clearly, a necessary condition for such a statistical method not to be worthless is that we have some idea of what it means to be "intelligently designed". If I tell you that I have a reliable method for determining whether or not an object is a skirnob, you will want to know what a "skirnob" is before you bother with trying apply my method.* If the ID movements' attempts to establish that certain objects were designed by an intelligent agent is to be of any scientific value we need a scientific model of intelligent agents.
Erik
* A similar point has been made in an article by Elsberry & Wilkins.
IP: Logged
|
|
|
Printer-friendly view of this topic