Extrapolation and direct matching mediate anticipation in infancy

Why are infants able to anticipate occlusion events and other people's actions but not the movement of self-propelled objects? This study investigated infant and adult anticipatory gaze shifts during observation of self-propelled objects and human goal-directed actions. Six-month-old infants anticipated self-propelled balls but not human actions. This demonstrates that different processes mediate the ability to anticipate human actions (direct matching) versus self-propelled objects (extrapolation).     

Perceiving the Present: Systematization of Illusions or Illusion of Systematization?

Mark Changizi et al. (2008) claim that it is possible systematically to organize more than 50 kinds of illusions in a 7 × 4 matrix of 28 classes. This systematization, they further maintain, can be explained by the operation of a single visual processing latency correction mechanism that they call “perceiving the present” (PTP). This brief report raises some concerns about the way a number of illusions are classified by the proposed systematization. It also poses two general problems—one empirical and one conceptual—for the PTP approach.     

Perceiving the present and a systematization of illusions

Over the history of the study of visual perception there has been great success at discovering countless visual illusions. There has been less success in organizing the overwhelming variety of illusions into empirical generalizations (much less explaining them all via a unifying theory). Here, this article shows that it is possible to systematically organize more than 50 kinds of illusion into a 7 × 4 matrix of 28 classes. In particular, this article demonstrates that (1) smaller sizes, (2) slower speeds, (3) greater luminance contrast, (4) farther distance, (5) lower eccentricity, (6) greater proximity to the vanishing point, and (7) greater proximity to the focus of expansion all tend to have similar perceptual effects, namely, to (A) increase perceived size, (B) increase perceived speed, (C) decrease perceived luminance contrast, and (D) decrease perceived distance. The detection of these empirical regularities was motivated by a hypothesis, called "perceiving the present," that the visual system possesses mechanisms for compensating neural delay during forward motion. This article shows how this hypothesis predicts the empirical regularity.     

Learning about occlusion: initial assumptions and rapid adjustments

We examined 6-month-olds abilities to represent occluded objects, using a corneal-reflection eye-tracking technique. Experiment 1 compared infants' ability to extrapolate the current pre-occlusion trajectory with their ability to base predictions on recent experiences of novel object motions. In the first condition infants performed at asymptote ( approximately 2/3 accurate predictions) from the first occlusion passage. In the second condition all infants initially failed to make accurate prediction. Performance, however, reached asymptote after two occlusion passages. This is the first study that demonstrates such rapid learning effects during an occlusion task. Experiment 2 replicates these effects and demonstrates a robust memory effect extending 24h. In occlusion tasks such long-term memory effects have previously only been observed in 14-month-olds (Moore & Meltzoff, 2004).     

Nanotoxicology and Ethical Conditions for Informed Consent

While their strength, electrical, optical, or magnetic properties are expected to contribute a trillion dollars in global commerce before 2015, nanomaterials also appear to pose threats to human health and safety. Nanotoxicology is the study of these threats. Do nanomaterial benefits exceed their risks? Should all nanomaterials be regulated? Currently nanotoxicologists cannot help answer these questions because too little is known about nanomaterials, because their properties differ from those of bulk materials having the same chemical composition, and because they differ so widely in their applications. Instead, this paper answers a preliminary ethical question: What nanotech policies are likely to contribute to society’s ability to give or withhold free informed consent to the potential risks associated with production and use of nanomaterials? This paper argues that at least four current policies appear to jeopardize the risk-disclosure condition that is required for informed consent. These are the funding problem, the conflict-of-interest problem, the labeling problem, and the extrapolation problem. Apart from future decisions on how to ethically make, use, and regulate nanomaterials, this paper argues that, at a minimum, these four policies must be modified. Government must spend greater monies on nanotoxicology; ensure independent nanotoxicology research; label consumer products containing nanomaterials; and avoid assuming that nanotoxicological properties are based merely on mass and chemical composition. Otherwise free informed consent to these new technologies and materials may be jeopardized.     

Applying Deutsch’s concept of good explanations to artificial intelligence and neuroscience – An initial exploration

Artificial intelligence has made great strides since the deep learning revolution, but AI systems remain incapable of learning principles and rules which allow them to extrapolate outside of their training data to new situations. For inspiration we look to the domain of science, where scientists have been able to develop theories which show remarkable ability to extrapolate and sometimes even predict the existence of phenomena which have never been observed before. According to David Deutsch, this type of extrapolation, which he calls “reach”, is due to scientific theories being hard to vary. In this work we investigate Deutsch’s hard-to-vary principle and how it relates to more formalized principles in deep learning such as the bias-variance trade-off and Occam’s razor. We distinguish internal variability, how much a model/theory can be varied internally while still yielding the same predictions, with external variability, which is how much a model must be varied to predict new, out-of-distribution data. We discuss how to measure internal variability using the notion of the Rashomon set and how to measure external variability using Kolmogorov complexity. We explore what role hard-to-vary explanations play in intelligence by looking at the human brain, the only example of highly general purpose intelligence known. We distinguish two learning systems in the brain – the first operates similar to deep learning and likely underlies most of perception while the second is a more creative system capable of generating hard-to-vary models and explanations of the world. We make contact with Popperian epistemology which suggests that the generation of scientific theories is a not an inductive process but rather an evolutionary process which proceeds through conjecture and refutation. We argue that figuring out how replicate this second system, which is capable of generating hard-to-vary explanations, is a key challenge which needs to be solved in order to realize artificial general intelligence.