Search

Why a new approach is needed in the quest for general artificial intelligence. Since the inception of artificial intelligence, we have been warned about the imminent arrival of computational systems that can replicate human thought processes. Before we know it, computers will become so intelligent that humans will be lucky to be kept as pets. And yet, although artificial intelligence has become increasingly sophisticated—with such achievements as driverless cars and humanless chess-playing—computer science has not yet created general artificial intelligence. In Algorithms Are Not Enough, Herbert Roitblat explains how artificial general intelligence may be possible and why a robopocalypse is neither imminent nor likely. Existing artificial intelligence, Roitblat shows, has been limited to solving path problems, in which the entire problem consists of navigating a path of choices—finding specific solutions to well-structured problems. Human problem-solving, on the other hand, includes problems that consist of ill-structured situations, including the design of problem-solving paths themselves. These are insight problems, and insight is an essential part of intelligence that has not been addressed by computer science. Roitblat draws on cognitive science, including psychology, philosophy, and history, to identify the essential features of intelligence needed to achieve general artificial intelligence. Roitblat describes current computational approaches to intelligence, including the Turing Test, machine learning, and neural networks. He identifies building blocks of natural intelligence, including perception, analogy, ambiguity, common sense, and creativity. General intelligence can create new representations to solve new problems, but current computational intelligence cannot. The human brain, like the computer, uses algorithms; but general intelligence, he argues, is more than algorithmic processes.

Despite the importance of effective causal reasoning, people still have trouble answering questions like: Does this chemical cause cancer? Was this economic policy the cause of economic growth? Does this drug prevent COVID-19? Did Thomas Crittendon cause the death of Jesse James? This essay aims to provide tools to help people reason effectively about causation. We use the killing of Jesse James to discuss naturalistic causal analysis, then introduce Pearl’s causal ladder and counterfactual reasoning. Methods, such as random assignment and the kind of observational methods used in epidemiology are discussed in the context of assessing evidence relative to the potential for causal relationships. We take an in-depth look at the evidence and methods that were used to establish a causal relationship between smoking and lung cancer and describe the psychological heuristics that can interfere with effective causal analysis.

Artificial general intelligence continues to be elusive in part because of a focus on a narrow range of problems that can all be described as path problems. A researcher sets up the goals, the optimization methods, and the parameters of the problem, which determine the potential solutions that a machine can learn. Machine learning is the process of finding the right parameter values that solve the problem. Formal path problems, like the games of chess or go, are fully described by their rules and the current state. They can be treated as purely mathematical processes, independent of any physical instantiation. Other, less formal problems, such as how to drive a vehicle on a busy road, do depend on the physical properties of their instantiation and on feedback from the physical world. Path problems are well defined with relatively easy to evaluate metrics, but they are not the only kind of problem that a generally intelligent agent needs to address. Other problems, called insight problems, require the solver to not just evaluate well-defined functions, but to create those functions. Failure to recognize the need to solve multiple types of problems leads people to believe that computers will at some point be able to make themselves arbitrarily intelligent (the technological singularity), potentially to the detriment of human existence. Path problem solving has provided a good model for special purpose problem solvers, but not for general intelligence. As the ancient Greek poet, Archilochus observed, “a fox knows many things, but a hedgehog one important thing.” Artificial intelligence researchers have been able to build very sophisticated hedgehogs, but foxes remain elusive. And foxes know how to solve insight problems.

The hearing sensitivity of an Atlantic bottlenose dolphin (Tursiops truncatus) to both pure tones and broadband signals simulating echoes from a 7.62-cm water-filled sphere was measured. Pure tones with frequencies between 40 and 140 kHz in increments of 20 kHz were measured along with broadband thresholds using a stimulus with a center frequency of 97.3 kHz and 88.2 kHz. The pure-tone thresholds were compared with the broadband thresholds by converting the pure-tone threshold intensity to energy flux density. The results indicated that dolphins can detect broadband signals slightly better than a pure-tone signal. The broadband results suggest that an echolocating bottlenose dolphin should be able to detect a 7.62-cm diameter water-filled sphere out to a range of 178 m in a quiet environment.

We examined the ability of Vietnam veterans with PTSD to focus attention on a primary digit detection task while concurrently viewing neutral or Vietnam-related picture and word distractors. Controlling for combat exposure, military service, and psychopathology, veterans with PTSD took longer to detect the target when Vietnam-related pictures were distractors. There were no reaction time differences when word stimuli were distractors. The latency effect was specific to trials with trauma-related pictures and did not spread to neutral trials interleaved within a mixed block of trauma and neutral pictures. Individuals with PTSD recalled proportionally more Vietnam-related words than other groups, implying differential attention to Vietnam-related words. Attending to trauma-related pictures interferes with performance of a concurrent task by individuals with PTSD.

The vocalizations from two, captive false killer whales (Pseudorca crassidens) were analyzed. The structure of the vocalizations was best modeled as lying along a continuum with trains of discrete, exponentially damped sinusoidal pulses at one end and continuous sinusoidal signals at the other end. Pulse trains were graded as a function of the interval between pulses where the minimum interval between pulses could be zero milliseconds. The transition from a pulse train with no inter-pulse interval to a whistle could be modeled by gradations in the degree of damping. There were many examples of vocalizations that were gradually modulated from pulse trains to whistles. There were also vocalizations that showed rapid shifts in signal type--for example, switching immediately from a whistle to a pulse train. These data have implications when considering both the possible function(s) of the vocalizations and the potential sound production mechanism(s). A short-time duty cycle measure was developed to characterize the graded structure of the vocalizations. A random sample of 500 vocalizations was characterized by combining the duty cycle measure with peak frequency measurements. The analysis method proved to be an effective metric for describing the graded structure of false killer whale vocalizations.

We present word and picture stimuli constituting a validated stimulus set appropriate for cognitive investigations of posttraumatic stress disorder (PTSD). Combat related and neutral words and pictures were rated by Vietnam veterans with PTSD and by three comparison groups along four dimensions: unpleasantness, Vietnam relevance, stressfulness, and memorability. There were distinctive patterns of responses by the PTSD group which efficiently discriminated the individuals in this group from those in the control groups. These stimuli have the potential to be developed as a diagnostic instrument.

Contents: Preface. H.L. Roitblat, H.E. Harley, D.A. Helweg, Cognitive Processing in Artificial Language Research. G. Bradshaw, Beyond Animal Language. S.A. Kuczaj II, V.M. Kirkpatrick, Similarities and Differences in Human and Animal Language Research: Toward a Comparative Psychology of Language. W. Bechtel, Knowing How to Use Language: Developing a Rapprochement Between Two Theoretical Traditions. E. Hunt, A Proposal for Computer Modeling of Animal Linguistic Comprehension. L. Bloom, Language Acquisition and the Power of Expression. P.L. Tyack, Animal Language Research Needs a Broader Comparative and Evolutionary Framework. J. Sigurdson, Frequency-Modulated Whistles as a Medium for Communication with the Bottlenose Dolphin (Tursiops truncatus). C.T. Snowdon, Linguistic Phenomena in the Natural Communication of Animals. R.M. Seyfarth, D.L. Cheney, Meaning, Reference, and Intentionality in the Natural Vocalizations of Monkeys. I.M. Pepperberg, Cognition and Communication in an African Grey Parrot (Psittacus erithacus): Studies on a Nonhuman, Nonprimate, Nonmammalian Subject. R.J. Schusterman, R. Gisiner, B.K. Grimm, E.B. Hanggi, Behavior Control by Exclusion and Attempts at Establishing Semanticity in Marine Mammals Using Match-to-Sample Paradigms. K.N. O'Conner, H.L. Roitblat, T.G. Bever, Auditory Sequence Complexity and Hemispheric Asymmetry of Function in Rats. W.D. Hopkins, R.D. Morris, Hemispheric Priming as a Technique in the Study of Lateralized Cognitive Processes in Chimpanzees: Some Recent Findings. P. Morrel-Samuels, L.M. Herman, Cognitive Factors Affecting Comprehension of Gesture Language Signs: A Brief Comparison of Dolphins and Humans. D.M. Rumbaugh, W. Hopkins, D.A. Washburn, E.S. Savage-Rumbaugh, Chimpanzee Competence for Comprehension in a Video-Formated Task Situation. S. Itakura, T. Matsuzawa, Acquisition of Personal Pronouns by a Chimpanzee. R.K.R. Thompson, D.L. Oden, "Language Training" and Its Role in the Expression of Tacit Propositional Knowledge by Chimpanzees (Pan troglodytes). M.R. Shyan, A.A. Wright, The Effects of Language on Information Processing and Abstract Concept Learning in Dolphins, Monkeys, and Humans. L.M. Herman, A.A. Pack, P. Morrel-Samuels, Representational and Conceptual Skills of Dolphins. M.D. Holder, L.M. Herman, S. Kuczaj II, A Bottlenosed Dolphin's Responses to Anomalous Sequences Expressed Within an Artificial Gestural Language. E.S. Savage-Rumbaugh, Language Learnability in Man, Ape, and Dolphin.

A bottlenose dolphin (Tursiops truncatus) demonstrated the ability to select the matching object in a matching-to-sample task after listening to another dolphin inspect the sample object via echolocation. The listener was prevented from inspecting the sample himself. In Experiment 1, with objects familiar to both dolphins, the listener’s performance was significantly better than chance. In Experiment 2, objects familiar to only one of the dolphins were used. On these trials, the listener’s performance was significantly better than chance only when the inspecting dolphin made a correct choice. Analysis of the listener’s responses when the inspector made an error demonstrated that this contingency was not due to the listener’s matching the inspector’s response, but was apparently due instead to inadequate information in the echo. The results suggest that the listener was able to “eavesdrop” on echoes produced by the inspector’s clicks and derive characteristics of the sample object.

The relationships between acoustic features of target echoes and the cognitive representations of the target formed by an echolocating dolphin will influence the ease with which the dolphin can recognize a target. A blindfolded Atlantic bottlenose dolphin (Tursiops truncatus) learned to match aspect-dependent three-dimensional targets (such as a cube) at haphazard orientations, although with some difficulty. This task may have been difficult because aspect-dependent targets produce different echoes at different orientations, which required the dolphin to have some capability for object constancy across changes in echo characteristics. Significant target-related differences in echo amplitude, rms bandwidth, and distributions of interhighlight intervals were observed among echoes collected when the dolphin was performing the task. Targets could be classified using a combination of energy flux density and rms bandwidth by a linear discriminant analysis and a nearest centroid classifier. Neither statistical model could classify targets without amplitude information, but the highest accuracy required spectral information as well. This suggests that the dolphin recognized the targets using a multidimensional representation containing amplitude and spectral information and that dolphins can form stable representations of targets regardless of orientation based on varying sensory properties.

Dolphins have biological sonar abilities that exceed those of any man-made system in an aquatic environment. One problem of particular importance, and for which only limited capabilities exist, is the detection and recognition of targets buried under sediment. This paper reviews dolphin echolocation capabilities and describes a system that uses a dolphin-like signal and biomimetic signal processing mechanisms to emulate the performance of the dolpin on such targets. The system employed a digitized dolphin click with a center frequency of 120 kHz and a 3dB bandwidth of 39 kHz, 50 μs duration. This signal was transmitted through seawater into mud and the echoes reflected from the objects were recorded and digitized. Two spectral estimators were used to extract a time-frequency representation of the echo. One was based on short-time fast Fourier transforms, and the other was based on an autoregressive estimator. The time-frequency representation was then processed by a separate backpropagation neural network for each estimator, designed to derive independent identifications of the targets. These identifications were then combined in a modified probabilistic neural network that used a linear transfer function rather than a binary function for its output. Finally the output of the probabilistic network was processed symbolically by a simple expert system. Three experiments are described in which the system was used to discriminate a small stainless-steel cylinder from cyliners of the same size made of hollow aluminium, foam-filled aluminium, or coral rock embedded in resin. Each of the targets was presented buried in mud at a depth of several centimeters. The system proved highly effectively at recognizing these buried targets.

How organisms (including people) recognize distant objects is a fundamental question. The correspondence between object characteristics (distal stimuli), like visual shape, and sensory characteristics (proximal stimuli), like retinal projection, is ambiguous. The view that sensory systems are 'designed' to 'pick up' ecologically useful information is vague about how such mechanisms might work. In echolocating dolphins, which are studied as models for object recognition sonar systems, the correspondence between echo characteristics and object characteristics is less clear. Many cognitive scientists assume that object characteristics are extracted from proximal stimuli, but evidence for this remains ambiguous. For example, a dolphin may store 'sound templates' in its brain and identify whole objects by listening for a particular sound. Alternatively, a dolphin's brain may contain algorithms, derived through natural endowments or experience or both, which allow it to identify object characteristics based on sounds. The standard method used to address this question in many species is indirect and has led to equivocal results with dolphins. Here we outline an appropriate method and test it to show that dolphins extract object characteristics directly from echoes.

This study examined the comprehension of English conventional indirect requests by native English speakers and Japanese learners of English. Subjects read stories inducing either a conventional or a literal interpretation of a priming sentence. Reading speeds did not vary as a function of the interpretation. Subsequent target sentences that paraphrased either the literal or the conventional interpretation of the prime sentence were read more quickly when they paraphrased a conventional interpretation of the sentence than when they paraphrased a literal interpretation. Target sentences were also read more quickly if they paraphrased the interpretation induced by the context than if they did not match. The results suggest that both native and nonnative speakers process both meanings of an ambiguous conventional request.

Empirical evidence of a relationship between combat-related PTSD and increased anger is lacking. In this study, 24 veterans of the Vietnam War with posttraumatic stress disorder (PTSD) scored significantly higher on an Anger factor comprising multiple measures of anger than did comparison groups of 23 well-adjusted Vietnam combat veterans and 12 noncombat Vietnam-era veterans with psychiatric diagnoses. In contrast, the 3 groups did not differ significantly on orthogonal factors, one of which comprised cognitive impulsivity measures and the other of which reflected motor impulsivity. Changes in heart rate in response to provocation loaded positively on the Anger factor and negatively on the 2 Impulsivity factors. Concurrent depression and trait anxiety did not have an effect on level of anger in individuals with PTSD. These empirical findings support and extend the clinical evidence regarding PTSD and anger.

AND NATURAL C ONCEPTS .... . . . . . . . . . . . . . . .. . . . . . . .... . . . . .... . . . . ......... . 691 Abstract Concepts. . . . . . . . . . . . . . . . . . . . . . . .. . . . ...... . . . . .. . ..... .. . . . ..... . . . . . . . .. 692 Categories and Concepts . . . . . . . . ... . . . . . . . . . . . . . . . . . ... . . . ..... . . . ....... . . . . . . . . . . . .. ,......... 697 Mechanisms of Categorization ... . . . . ... . . . .... . . . .... . . . . . .. . . . . . .. . . . . . . .. .... . . ..... . ... . . . . 700 C ONCLUSIONS . . .. ..... . . .... . . . . ...... .. .. . . . . .... . . . .. .... . . . ..... ......... . .... . . . . ...... .. ...... . 703

Only a limited number of species have been found capable of generalized matching-to-sample (MTS) after exposure to relatively few training exemplars. We trained a juvenile, experimentally naive California sea lion (Zalophus californianus) in MTS, using a pair of three-dimensional objects as samples. Successful matching to a criterion of 90% correct or better over 2 successive sessions was attained in 12 sessions (269 trials and 70 errors). Two subsequent “partial” transfer tests, in which each of the two training objects was paired with a novel test object, and four additional transfer tests, all with novel objects, were presented following training. An 80% performance criterion over 2 successive sessions was reached, or closely approximated, in from 2 to 4 transfer sessions for all transfer tests; errors to criterion tended to be reduced across the successive novel transfer tests and were as few as five during the final two tests; and performance on the first 48 trials of the last two novel transfers was not significantly different from a near-ceiling level baseline performance measure. Neophobic responses of the sea lion to new objects precluded an unbiased evaluation of immediate (Trial 1) transfer. The sea lion’s short-term memory for sample objects was also measured. Matching performance was maintained at a level of 78% correct responses or better for delays through to 45 sec after removal of the sample object. At a 58-sec delay, the longest tested, performance declined to 69% correct responses. These retention levels are only somewhat below levels reported for dolphins and nonhuman primates tested on visual delayed MTS, but they are above levels typically reported for pigeon subjects.

An novel network architecture, the integrator gateway network, was developed to classify multiple successive echoes from targets ensonified by a dolphin's natural echolocation signal in a naturalistic environment. The inputs to the network were spectral vectors of the returning echo plus one unit representing the start of each scan. This network combined information from successive echoes returning from the same target and reset between scans of different targets. The network was trained on a small subset (4%) of the total set of available echoes (1,335). Depending on the measure used to assess it, the network correctly classified between 90–93% of all echo trains. In contrast, a standard back propagation network with the same number of units and variable connections as in the integrator gateway network performed with only about 63% accuracy in classifying echo trains. The integration model seems to provide a better account of the dolphin's performance than a decision model that does not combine information from multiple echoes. Although it does not prove that dolphins similarly combine information from multiple echoes, in the absence of relevant neurophysiological evidence, it provides support for such a hypothesis. It also suggests the potential beneifts to be derived for pattern recognition from combining multiple samples from the same target.

An adult male dolphin was trained to perform a three-alternative delayed matching-to-sample task while wearing eyecups to occlude its vision. Sample and comparison stimuli consisted of a small and a large PVC plastic tube, a water-filled stainless steel sphere, and a solid aluminum cone. Stimuli were presented under water and the dolphin was allowed to identify the stimuli through echolocation. The echolocation clicks emitted by the dolphin to each sample and each comparison stimulus were recorded and analyzed. Over 48 sessions of testing, choice accuracy averaged 94.5% correct. This high level of accuracy was apparently achieved by varying the number of echolocation clicks emitted to various stimuli. Performance appeared to reflect a preexperimental stereotyped search pattern that dictated the order in which comparison items were examined and a complex sequential-sampling decision process. A model for the dolphin's decision-making processes is described.

Echolocating bottlenose dolphins (Tursiops truncatus) discriminate between objects on the basis of the echoes reflected by the objects. However, it is not clear which echo features are important for object discrimination. To gain insight into the salient features, the authors had a dolphin perform a match-to-sample task and then presented human listeners with echoes from the same objects used in the dolphin's task. In 2 experiments, human listeners performed as well or better than the dolphin at discriminating objects, and they reported the salient acoustic cues. The error patterns of the humans and the dolphin were compared to determine which acoustic features were likely to have been used by the dolphin. The results indicate that the dolphin did not appear to use overall echo amplitude, but that it attended to the pattern of changes in the echoes across different object orientations. Human listeners can quickly identify salient combinations of echo features that permit object discrimination, which can be used to generate hypotheses that can be tested using dolphins as subjects.

The focus of this study was to investigate how dolphins use acoustic features in returning echolocation signals to discriminate among objects. An echolocating dolphin performed a match-to-sample task with objects that varied in size, shape, material, and texture. After the task was completed, the features of the object echoes were measured (e.g., target strength, peak frequency). The dolphin’s error patterns were examined in conjunction with the between-object variation in acoustic features to identify the acoustic features that the dolphin used to discriminate among the objects. The present study explored two hypotheses regarding the way dolphins use acoustic information in echoes: (1) use of a single feature, or (2) use of a linear combination of multiple features. The results suggested that dolphins do not use a single feature across all object sets or a linear combination of six echo features. Five features appeared to be important to the dolphin on four or more sets: the echo spectrum shape, the pattern of changes in target strength and number of highlights as a function of object orientation, and peak and center frequency. These data suggest that dolphins use multiple features and integrate information across echoes from a range of object orientations.

A novel network architecture was developed to classify multiple successive echoes from targets ensonified by a dolphin echolocating in a naturalistic environment. The inputs to the network were spectral vectors of the echo plus one unit representing the start of each scan. This network combined information from successive echoes from the same target and reset between scans of different targets. The network was trained on a small subset (4%) of the total set of available echoes (1,335). Depending on the measure used to assess it, the network correctly classified between 90% and 93% of all echo trains. In contrast a standard backpropagation network with the same number of units and variable connections performed with only about 63% accuracy in classifying echo trains. The integration model seems to provide a better account of the dolphin's performance than a decision model that does not combine information from multiple echoes. >

The echolocation capabilities of dolphins and small whales exceed those of current man-made sonars. Dolphins, beluga whales and false killer whales can perceive small targets presented over 110 m away, can classify target shapes independent of internal target reverberation, can discriminate wall thickness differences in targets of less than .2 mm, and can operate in high noise environments. Recent natural observations indicate that several species may also detect and choose targets buried in sediment. These tasks are accomplished through the use of range-gated clicks that tend to be broad band with peak frequencies exceeding 100 kHz. The short (50 microsec) pulses can have amplitudes exceeding 220 db and bandwidths exceeding 60 kHz. This paper provides a short review of animal echolocation capabilities, methodologies used to examine them, and potential uses of neural networks and other signal processing techniques to understand and perhaps duplicate those animal capabilities. >

Whales and dolphins have evolved in a sensory world very unlike our own. Although one can guess what the perceptions of a cetacean might be, it is impossible to “get inside a dolphins head” and experience what it must be like to hear sounds over 100kHz and discern fine details of the environment, conspecifics, and prey via echolocation. It is possible, however, using psychophysical techniques first developed for human measurements, to characterize some properties of cetacean perception and accurately measure the acoustic sensations that are experienced by whales and dolphins.

The central feature of intentionality for philosophers is "aboutness." Something, some symbol, idea, or mental state has intentionality when it is about something else. For example, I may believe that baseball is fun. My belief is a mental state and specifically it is a mental state about the relationship between baseball and fun. Paradigmatic examples of intentional ideas include needs, wants, desires, and beliefs. The problems for philosophers have been in part to explain how symbols obtain this property of aboutness, and how intentional symbols can be causally effective. In a standard mechanical system, one item affects another by virtue of the causal laws of physics. Physical properties of the items affect other physical properties of the items in ways that are determined by the physical properties of the items and their relations. In an intentional cognitive system, on the other hand, one mental state affects another because of their content, that is because of what they are about. The physical structure of the symbol does not figure into its causal function. The problem of translating thought into action has a long history in philosophy as well as in psychology. The problems inherent in intentional systems have long resisted solution by philosophers and have long haunted the foundations of psychology (e.g., Bechtel, 1988). Some psychologists, most notably Skinner (e.g., 1950), have attempted to solve these problems by describing psychology solely in terms of its physical characteristics, neglecting thereby as unnecessary, any consideration of the intentional aspects of behavior. Other investigators (e.g., Fodor, 1975) have attempted to solve the problems by focusing almost exclusively on mental contents, neglecting thereby as uninteresting, any consideration of how those mental contents are causally effective regarding behavior, or how they may be different from other kinds of items. Still other investigators, notably Pylyshyn (e.g., 1980), have argued for a kind of demarcation between those mental events that are intentional and those that are merely the product of the "mechanics" of the "cognitive architecture." The latter processes obey the rules of physical causation, whereas the former processes are characterized by their content and obey the rules of intentional causation (without any explicit characterization of what those rules might be). I raise these points about philosophical intentionality to place the sense of intention on which Lewis has focused in a larger perspective. Desires or intentions in Lewis's sense (I'll call it specific or colloquial intention or intentions; see Bechtel, 1988) are intentional in the philosophical sense (I'll call it general or philosophical intention or intentiong) because they are desiresfor something. They are about a certain kind of relationship between the actor and the thing or state the actor desires. Placing the problems of specific intentionalitys in the context of philosophical intentionalityg helps to show why it has been so far a particularly intractable problem. One seeks to explain intentionals behavior on the basis of the intentionalg sense rather than on the basis of specific computational or mechanical bases. That is, one seeks to explain a behavior in terms of the individual's desires rather than in terms of reflex or physical happenstance. The problems inherent in philosophical intentionalityg are necessarily, therefore, inherent in intentionality5, but any particular solutions that one adopts to explain how an individual comes to intends may not generalize to the problem of how mental states are about their referent. In fact, one explanation for intentionalitys is that it is not an example of intentionalityg, but rather the results of specific mechanical or computational mechanisms. Lewis seems specifically interested in the means by which organisms use desires and plans to guide their behavior. In this interest, he is not far different from Tolman (e.g., 1932, 1959). Tolman argued that behavior is goal directed, controlled by its outcome. When a hungry rat seeks food, it is behaving as if it has a purpose or goal. He also argued for a molar analysis of behavior, described in terms of actions, rather than a molecular analysis based on movements or muscle twitches. Because of its adaptiveness and variability, purposive behavior can be made of many alternative kinds of basic units that vary from episode to episode, but all perform the same function. The molar analysis emphasized features of behavior that are consistent from one episode to another involving the same goal and deemphasizes those variations that are different from episode to episode and not strongly related to the goal. The particular pattern of movement by which an organism reaches a goal is less important in this view than the goal-seeking aspects of the behavior. Notice that Tolman used the term "purposive" rather than "purposeful" in describing the rat's performance. Tolman did not claim that rats had a purpose in mind when they behaved, merely that they acted as if they had purpose. The notion of purposive rather than purposeful suggests a solution to the dilemma of how an organism might develop intentional behavior without previously having demonstrated any. Behavior can be goal directed (intentionals) without the organism having to reflect on its goal orientation. It need not have any mental state about its goal (intentionalityg) to behave relative to its goal. The purposive approach is very similar to that taken by Lewis. Hunger is a problematic but enlightening case in point. Certain artifacts such as thermostats or heat-seeking missiles appear to be intentionals in the sense of being goal directed. the goal of the thermostat is to regulate the temperature within a narrow range about its set point. The goal of a heatseeking missile is to contact its heat-source target. Both mechanisms operate in an adaptive way relative to specific goals, but it would probably be erroneous to claim that either device had one of these goals in mind. If we describe hunger as a desire for food, then it is clearly an example of a mental state about food (intentionalityg). If on the other hand, we describe hunger as a result of the brain's response to certain neural activities in the hypothalamus that operate to maintain body weight within a narrow range about its set point, then it

This study reports the use of unsupervised, self-organizing neural network to categorize the repertoire of false killer whale vocalizations. Self-organizing networks are capable of detecting patterns in their input and partitioning those patterns into categories without requiring that the number or types of categories be predefined. The inputs for the neural networks were two-dimensional characterization of false killer whale vocalizations, where each vocalization was characterized by a sequence of short-time measurements of duty cycle and peak frequency. The first neural network used competitive learning, where units in a competitive layer distributed themselves to recognize frequently presented input vectors. This network resulted in classes representing typical patterns in the vocalizations. The second network was a Kohonen feature map which organized the outputs topologically, providing a graphical organization of pattern relationships. The networks performed well as measured by (1) the average correlation between the input vectors and the weight vectors for each category, and (2) the ability of the networks to classify novel vocalizations. The techniques used in this study could easily be applied to other species and facilitate the development of objective, comprehensive repertoire models.

Most of the papers presented in this volume are related to such topics as sensory anatomy and physiology, echolocation, psychophysics of sensory systems, and vocalization and acoustical communication in aquatic mammals. All of them provide useful information on the way in which these animals process information. The aim of this contribution is to extend this approach by presenting additional data about a slightly more psychological phenomenon: the conceptualization of information. In the first section we will examine the special status of symmetry in pattern recognition, followed by a review of symmetry studies in other animals. Finally, the third and fourth sections are devoted to abstract concepts in dolphins and to the question of whether dolphins can detect and conceptualize symmetry.

In delayed matching-to-sample (DMTS) the subject is presented with a sample at one time and must then pick a matching comparison stimulus from a set of alternatives. The choice the subject makes in discriminating among the comparison stimuli, is contingent on the identity of the sample that preceded them in the trial.

Although the most striking of cetacean sensory abilities is their acoustic echolocation skill (Nachtigall and Moore, 1989), they are also endowed with sensitivities for a number of other modalities. Among the senses considered at the conference are included magnetic, tactile, electroreceptive, pressure, and gravitational senses. Further, in addition to the animals’ sensitivity to these various modalities marine mammals are also endowed with the cognitive capacities to make use of this sensory information to represent their environment and the events within it.

Byrne & Russon propose an account of imitation that mirrors levels of behavioral organization, but they perpetuate a tendency to dismiss imitation by members of most species as the result of more primitive processes, even though these alternative phenomena are often poorly understood. They argue that the prerequisites to program-level imitation are present in great apes, but the same prerequisites appear to be present in a broad range of species. The distribution of imitative capacity across species may be more limited by research methodology than by cognitive ability.

Dolphins have been observed foraging for prey that bury themselves into a sandy bottom. The dolphins swim about 1 to 2 m above the bottom and scan in a circular motion or swim immediately off the bottom, scanning from side to side with their beams pointed approximately normal to the bottom. The dolphin has been emulated by using dolphinlike sonar signals (120‐kHz peak frequency, 39‐kHz bandwidth, 50‐μs duration) in order to classify proud and buried targets in real time. The transducer was attached to a bottom‐crawling remotely operated vehicle. Target echoes were received via a cable and digitized at 1 MHz. Short‐time Fourier transform and the Morlet wavelet were used to obtain time‐frequency representations of the echoes. Echoes were processed in a hierarchical neural network system to perform target classifications. Six targets (cast iron pot, stainless steel sphere, glass jar, concrete tile, and coral rock) were placed either on the ocean bottom or buried into the bottom. Echoes were separated into three target categories: (1) cast iron pot; (2) stainless steel sphere; and (3) remaining four objects. The neural network was able to correctly identify 74%, 97%, and 88% of the category 1, 2, and 3 test echoes.

The Biomimetic Sonar invention ensonifies submerged objects, digitizes acoustic images reflected from the ensonified objects, and classifies and stores the digitized images using electronic processing derived from that believed to be used by Tursiops truncatus , the bottlenose dolphin. The invention also provides a probable identification of an ensonified object based on comparison of an ensonified object with templates in a library of acoustic images.