Drinking from the fire hose: pdf free download

But it is iar semantic relationships e. These are important, but they reflect experience or personal memory. How- points on a spectrum of the vividness of subjective ever, there are also associations from vivid sensory experience.

In each case, the information content images for example, a characteristic smell to of a particular experience is determined by the episodes buried deep within long-term memory. Ongoing further increase the information content of long- visual experience of a particular object has the term memory, but the unpredictability and information content of the support for the tracker distinctiveness of that collection of links would for that object in the sensory stream.

The percep- be evident to the individual agent from its own tion of an object is much more vivid when the experience. If the memory The stored memory of a personal experience of an artificial agent is created purely through must contain far less information than the original symbolic input, or if bulk memory can be backed experience, but that amount of information must up and restored as with a disk drive, then it may still be very large, since it can apparently contain become possible to create complex individuals that some sort of sensory snapshots as well as symbolic have identical memory states, at least momentarily.

Thus, differences in 4. Why does subjective experience feel vividness appear to correspond well with differ- like anything at all? In sneaking up on the Hard Problem, we have 4. Why is subjective experience so focused on why different experiences feel different. We have not solved the Hard Problem of why sub- jective experience feels like anything at all. Like the proverbial snowflakes, no two separately- created multi-mega-pixel digital camera images are 5 ever identical, simply because of the huge number Furthermore, as Sloman and Chrisley observe [43], even bit- by-bit identity of stored representations would not imply that the of bits they encode, and the number of unpredict- experiences of two agents were the same, since the interpreta- able physical processes that determine those bits.

Kuipers When a question is this difficult, perhaps it is a Center and Periphery, and Active and Passive — non-question. As molecular biologists continue to the dynamical tracker model of consciousness pro- tease out the mysteries of the genetic code, mole- vides a specific explanation. There is no magical expressible within the dynamical tracker model. Rather, when examined closely, there is a vast 5. Qualitativeness spectrum of complexity of molecular behavior.

Perhaps the same is true of subjective experience. Every conscious state has a qualitative feel to it. Thermostats and autofocus cameras interact with. We humans are vastly more complex in terms of the numbers and The vividness, intensity, and immediacy of sub- variety of processes taking place. The intensity of subjective experi- process certain information in certain ways.

Pain ence increases with the information content of the is insistent and intrusive because its biological pur- input: from text or verbal descriptions, to viewing a pose is to attract our attention to a threat. We learn color photograph, to memories or dreams of experi- to experience higher-level sensations through their ences, to live multisensory experience.

The Easy Problem and the problem of that the entire visual field is perceived with the Intentionality both have functional import, and same high fidelity as the point of foveal attention are likely to provide plenty for us to do. Evaluating a theory of tent is determined by the number of other possible consciousness color symbols that could have been stored as a value of that attribute: at most a dozen bits or so. On the It is not yet possible to build a robot with sufficiently other hand, if a tracker is bound to a region in the rich sensorimotor interaction with the physical sensor stream, the number of bits of color informa- environment, and a sufficiently rich capability for tion streaming past, even in a small region, is orders tracking and reasoning about its sensor and motor of magnitude larger.

The remaining barriers, however, appear to be stream means that attribute values drawn from technical rather than philosophical. Although it is implausible for the entire sensory For some of these features — Qualitativeness, stream to be stored in long-term memory, at least Subjectivity, Intentionality, Distinction between some qualia e. This individuation of long-term memory.

Robots may not have the same There is a compelling argument that perception constraints. Also see the movie, Being John requires abduction [44]. There must be a process Malkovich. If this process were purely bottom— aspects of its perceptions are under its direct con- up i.

This distinction comes not However, experience suggests that there are signif- from anatomy, but from the existence of tight con- icant top—down and perhaps random processes for trol loops. Virtual reality and telepresence are sub- generating hypotheses.

Under conditions of sensory jectively compelling exactly because humans are deprivation, people tend to hallucinate, that is, to quickly able to learn novel models of senses, generate perceptual hypotheses poorly grounded in actions, body, and world from interactive experi- sensor input [45].

Symbolic logical own sensorimotor system confirm this description of theories are subject to multiple interpretations. Searle [15], p. Thus, the qualitativeness of experience control law. Unity the cognitive agent because it helps to keep the generation and refutation of perceptual hypotheses At present, I do not just experience the feelings in in balance. Subjectivity outside, but I experience all of these as part of a single, unified, conscious field [15], p.

Because of the qualitative character of conscious- We experience the audio—visual surround as a ness, conscious states exist only when they are single unified field, continuous in space and time, in experienced by a human or animal subject.

For example, the fovea has consciousness has a first-person ontology [15], p. Consciousness is experienced exclusively from a The density of color-receptive cones is even more first-person point of view.

That is, its body is coherent narrative, constructed ms after the physically embedded in the world, and Eqs.

Several mechanisms and cognitive perceptions z. By selecting a control law H i, the architectures have been proposed to explain how agent creates a causal path from its sensory input z this narrative is constructed. Within this kind of programmed the algorithms and control laws that architecture, trackers are the modules that inter- make the tracker work. This argument is vulnerable face between the sensor stream and the symbolic to a demonstration that effective trackers can be cognitive modules.

As discussed in Section remain to be answered about how the coherent 3. Other highly relevant parallel and irregular sources of input. Global Work- work on the same problem includes [33,46,48,49]. The Spatial Semantic Hierarchy [36] In robotics, the Kalman Filter [21] is often used to maps an unknown environment by identifying locally predict the most likely trajectory of a continuous distinctive states and linking them into a topological dynamical system along with its uncertainty , given map.

The ability of a symbol to refer to a distinctive a model and an irregular collection of sensor obser- state in the physical environment depends on the vations along with their uncertainties.

The tech- behaviors of the dynamical systems defined by the nical methods are different, but philosophically, the control laws, not on intentionality in the pre-exist- slightly retrospective construction of a plausible ing set of symbols.

Pierce and Kuipers [39] showed coherent narrative from irregular observations is that these control laws could be learned from the no more problematical than a Kalman Filter. Intentionality constrained by their causal connections with the environment. This feature, whereby many of my We believe that learning methods like these can experiences seem to refer to things beyond them- be extended to learn trackers for many kinds of selves, is the feature that philosophers have come distinctive configurations in the sensory stream.

The agent thus acquires intention- [14] is that strong AI commits a category error with ality of its own. The mind necessarily has intentionality the ability to refer to objects in the 5.

The Distinction between the Center world , while computation the manipulation of and the Periphery formal symbols according to syntactic rules neces- sarily lacks intentionality. Therefore, the mind can- Some things are at the center of my conscious field, not be a computation.

I can focus tracker for a high-level concept delivers: it binds my attention on the glass of water in front of me, or a portion of the current sensor stream to the sym- on the trees outside the window, without even bolic description of an object believed to be in the altering my position, and indeed without even mov- external world. The relationship of intentionality ing my eyes. In some sense, the conscious field follows from the causal connection from the exter- remains the same, but I focus on different features nal, physical world to the contents of the sensor of it [15], p.

When cer- pen. Attentional attended to later, or they may contribute to main- processes such as giving a particular tracker more taining Situatedness next section. However, they fail to be conscious because the agent.

The Gestalt structure to shift focus of attention. We do not, for example, in normal vision see undif- 5.

Situatedness ferentiated blurs and fragments; rather, we see tables, chairs, people, cars, etc. But, facts. One is, the brain has a capacity to take normally I am in some sense cognizant of where I am degenerate stimuli and organize them into coher- on the surface of the earth, what time of day it is, ent wholes. Furthermore, it is able to take a con- what time of year it is, whether or not I have had stant stimulus and treat it now as one perception, lunch, what country I am a citizen of, and so on with now as another [15], p.

When it finds it, that structure is [15], p. The findings of the clear when applied to images of objects that move Gestalt psychologists provide clues about the within the visual field, it applies equally well to properties of individual trackers, of the process tracking the location of the robot within a given by which potential trackers are instantiated and frame of reference, for example, localization within become active, of the ensemble of active track- the current enclosing room.

This concept of tracker ers, and perhaps even of the learning process by can, in turn, be generalized to track motion through which trackers for new types of objects are an abstract space such as time or a goal hierarchy. Such background situation trackers could poten- For example, interpretation-flipping figures such tially continue tracking with little or no attention.

Active and Passive Consciousness as mutual exclusion and continued competition among the higher level of hierarchical trackers, The basic distinction is this: in the case of percep- while lower levels preserve their bindings and can tion seeing the glass in front of me, feeling the be used by either competing interpretation. Kuipers 5. Mood experience. Memory can include qualia such as snapshots or fragments of the sensory stream with All of my conscious states come to me in some sort high information content.

The content of the con- of mood or other. How mood affects 6. Conclusions behavior is embedded in part in the mechanism for selecting the next control law H i. We approach the problem of consciousness from the pragmatic design perspective of AI and robotics. One of the major requirements on an embodied 5. One cognitive architec- phenomenon that for any conscious state there is ture that meets this requirement includes trackers some degree of pleasure or unpleasure. Or rather, that ground dynamic symbolic descriptions in spa- one might say, there is some position on a scale that tio—temporal regions of the sensory stream, and a includes the ordinary notions of pleasure and plausible coherent narrative that explains the unpleasure [15], p.

The Sense of Self tual images in the sensor stream. In many ways, the most pragmatically useful Qualia reflect the information density of the aspect of consciousness is the ability to observe, sensor stream.

Intentionality is plans, and beliefs. And continuous nature of conscious experience is the the sequential stream of subjective consciousness is post-hoc construction of a plausible coherent nar- a plausible coherent narrative, constructed retro- rative to explain a somewhat irregular collection of spectively by ms or so.

The Intentionality Problem applies to any Once such a narrative exists, it can be stored in embodied agent, human or robot, that interacts long-term memory, recalled, and reasoned about with the world only through coded sensor and like any other piece of symbolic knowledge.

The motor signals. We argue that there is no magic, construction, storage, recall, and manipulation of for humans or robots, whereby symbols inside the this kind of knowledge poses no fundamental diffi- mind can refer, directly and correctly, to corre- culties for computational modeling [51,43]. Drinking from the firehose of experience sponding objects in the outside world.

On the References other hand, we can exhibit early versions of learn- ing algorithms that can construct explanations for [1] Kuipers B. Consciousness: drinking from the firehose of the regularities of pixel-level sensorimotor inter- experience.

In: Proceedings of the 20th National Conf. AAAI Press; Sneaking up on the Hard Problem of consciousness. AI and consciousness: the- sentation is an entity hypothesized by such a oretical foundations and current approaches, AAAI Fall Sym- learning algorithm, that is, another internal con- posium Series. If these internal entities correspond [3] Sacks O. The man who mistook his wife for a hat and other clinical tales. The feeling of what happens. New York: Har- be able to plan and act effectively.

If not, it is court, Inc; A brief tour of human consciousness. Problem by offering relative information content [6] Crick F, Koch C. A framework for consciousness. Nat Neurosci as an explanation for why different experiences ;6 2 — The quest for consciousness: a neurobiological have different levels of vividness. This leaves approach. The society of mind. NY: Simon and Schuster; However, we do know that information [9] Baars BJ. A cognitive theory of consciousness. New York: transfer in an embodied agent necessarily corre- Cambridge University Press; Consciousness explained.

The conscious mind: in search of a fundamen- sensed. Drawing on an analogy with classic models tal theory. New York: Oxford University Press; An information integration theory of conscious- sical correlates of raw information transfer that ness. BMC Neurosci ; Seeing red: a study in consciousness. Minds, brains, and programs.

Behav Brain Sci of the agent. The empirical and philosophical study of con- [15] Searle JR. Mind: a brief introduction. The study of the brain helps us bramanian V, et al. How much the eye tells the brain. Curr understand the one implementation of a conscious Biol ;— Minds, machines and Searle. J Exp Theor Artif Intell ;— But according to our claim, consciousness is not [18] Quine WVO.

Two dogmas of empiricism. In: Quine WVO, restricted to biological implementation. The editor. From a logical point of view. Second revised essential features of consciousness can, in princi- edition, Cambridge, MA: Harvard University Press; Some philosophical problems from the sorimotor system, embodied and embedded in its standpoint of artificial intelligence. In: Meltzer B, Michie D, editors.

Machine intelligence 4. Edinburgh: Edinburgh Uni- environment. A tutorial on visual servo control. Acknowledgements [21] Gelb A. Applied optimal estimation. Cybernetics or control and communication in This work has taken place in the Intelligent Robotics the animal and the machine. University of Texas at Austin. Active vision.. A framework for representing knowledge. The psychology of computer vision. Although the review articles and highlights of the literature that appear in various journals make some attempt to look beyond the top journals for gems e.

Premise 2: scientists are laughably anachronistic in their approach to information. The internet has completely revolutionized the way that everyone accesses information — everyone except for us academics, that is, who continue to organize and locate information in essentially the same old ways.

We made lots of photocopies. Now, we publish papers in electronic journals and locate articles using citation indexes and PubMed with web interfaces although more and more of us prefer Google. We download lots of PDFs and make lots of printouts. The really technologically savvy among us might eschew the print and make digital folders on our laptops. Academic scholarship has fallen far behind the times, at least compared with such noble pursuits as online shopping.

A few fields, such as physics and mathematics, are making progress; anyone can freely read manuscripts submitted to arXiv. The devil is in the details: what flavor of technology will best satisfy our hunger for efficient access to the most relevant information?

In other words, what should a useful and practical post-filter look like? During the course of sharing our ideas on this subject, we will apply the typical straw 1 Chris Patil is at the Buck Institute for man approach and raise and dispatch a few weaker proposals.

Software text mining has come a long way and is actually pretty good; if you know that you want an article about subjects X and Y but not Z, modern data mining utilities can reliably retrieve every single article in every available database matching that description. There are ongoing efforts to make these tools cleverer using natural-language processing to evaluate the semantic relationships between the entities mentioned within an article in order to improve the quality of search results.

But no matter how finely we hone these tools, they can do little to help us sort through the large number of descriptively similar papers pertaining to our original interests in X and Y but not Z. Which ones should I read? Which ones should I delete? Search engines like Google make a good start at doing just that: search rank is based on linkage, so it potentially represents a measure of how useful others have found a document in the past.

This need not be the case, however; other users may have linked to a document for reasons other than its high quality. Furthermore, as with citations, links take a long time to accumulate. One possibility would be to allow others to spend a good deal of their own time to filter the literature within a field and share their opinions with the world. Straw man solution 2: blogs. It is similarly tempting to assign the post-filtering task to hordes of enthusiastic volunteers — intrepid, pajama-clad souls, armed only with keyboards and search engines, who would wade through the jungle of the literature and return to us only the choicest prizes.

But this is a fantasy. For bloggers to provide an efficient and efficacious post-filter service, they would have to meet an imposing list of qualifications: sufficiently well-trained to make wise judgments about the papers most worthy of attention; sufficiently idle to have nothing better to do than read papers all day; free of idiosyncrasy or agenda that might bias their choices; and willing to work continuously for free.

An Amazon for science publishing? Consider shopping on Amazon. The alternatives are ranked, and the rankings are backed up by prose descriptions of buyer experiences. We just skim their words and make a good decision informed by their past experience. Could we establish such a system for the literature? Suppose that rich metadata accompanied papers, detailing their classification — questions asked, methods used, model organism used — as well as information about the reactions of every other reader of that paper.