Beyond Science Fiction Tales

We have already argued that an automated archaeologist cannot understand past social actions by enumerating every possible outcome of every possible social action. The need to insert all the world within the automated archaeologist’s brain and then maintain every change about is impossible. However, if we cannot introduce the world inside the robot, we may introduce the robot inside the world. What the automated archaeologist would need then is to be situated in the past, and then using observation and attention to learn from human action, because of the complexities of the past, which resist modeling. It leads to a modification of the aphorism espoused by Rodney Brooks (1989): “the past itself should be its own best model.” Consequently, the automated archaeologist must travel to the past to be able to understand why it happened. Only by being situated directly in the past, the automated archaeologist would understand what someone did and why she did it there or elsewhere.

Visual and Non-Visual Analysis in Archaeology

As we have discussed in previous chapters, an artificial neural network is an information-processing system that maps a descriptive feature vector into a class assignment vector. In so doing, a neural network is nothing more than a complex and intrinsically nonlinear statistical classifier. It extracts the statistical central tendency of a series of exemplars (the learning set) and thus comes to encode information not just about the specific exemplars, but about the stereotypical featureset displayed in the training data (Churchland, 1989; Clark, 1989, 1993; Franklin, 1995). That means, it will discover which sets of features are most commonly present in the exemplars, or commonly occurring groupings of features. In this way, semantic features statistically frequent in a set of learning exemplars come to be both highly marked and mutually associated. “Highly marked” means that the connection weights about such common features tend to be quite strong. “Mutually associated” means that co-occurring features are encoded in such a way that the activation of one of them will promote the activation of the other.

Shape Analysis in Archaeology

In order to be able to acquire visual information, our automated “observer” is equipped with range and intensity sensors. The former acquire range images, in which each pixel encodes the distance between the sensor and a point in the scene. The latter are the familiar TV cameras acquiring grey-level images. That is to say, what the automated archaeologist “sees” is just the pattern of structured light projected on the scene (Trucco, 1997). To understand such input data is the spatial pattern of visual bindings should be differentiated into sets of marks (points, lines, areas, volumes) that express the position and geometry of perceived boundaries, and retinal properties (color, shadow, texture) that carry additional information necessary for categorizing the constituents of perception.

Problem Solving in the Brain and by the Machine

When a specific goal is blocked, we have a problem. When we know ways round the block or how to remove it, we have less a problem. In our case, the automated archaeologist wants to know the cause of the observed material outcomes of social action. What blocks this goal is a lack of knowledge: it does not know the particular mechanism that caused in the past what it sees in the present. To remove this obstacle it must learn some specific knowledge: how a causal process or processes generated the specific measurable properties determining the observed evidence. To the automated archaeologist, problem solving has the task of devising some causal mechanism that may mediate between the observation and its cause or causes. Consequently, explanatory mechanisms taken in pursuit of that goal can be regarded as problem solving. In other words, explanation is a kind of problem solving where the facts to be explained are treated as goals to be reached, and hypotheses can be generated to provide the desired explanations (Thagard, 1988).

Spatiotemporal Analysis

As we have suggested many times throughout the book, the general form of an archaeological problem seems to be “why an archaeological site is the way it is?” If we translate it into the spatial domain, we should be asking “where social agents performed their actions and work processes on the basis of the observed relationships between the actual locations of the social action material traces?,” or more precisely, “why those archaeological materials have been found here and not elsewhere?” Consequently, the automated archaeologist should infer where social agents performed their actions and work processes based on the observed relationships between the actual locations of the supposed material consequences of social action. This is the domain of application for a spatial analysis: to infer the location of what cannot be seen based on observed things that are causally related to the action to be placed. Knowing where someone made something based on what she did, is an inverse problem with multiple solutions, which can be solved using some of the methods and technologies already presented.

Texture and Compositional Analysis in Archaeology

In this section, we will consider archaeological textures as the archaeological element’s surface attributes having either tactile or visual variety, which characterize its appearance. The surfaces of archaeological objects, artifacts, and materials are not uniform but contain many variations; some of them are of visual or tactile nature. Such variations go beyond the peaks and valleys characterizing surface micro-topography, which is the obvious frame of reference for “textures” in usual speaking. Archaeological materials have variations in the local properties of their surfaces like albedo and color variations, uniformity, density, coarseness, roughness, regularity, linearity, directionality, frequency, phase, hardness, brightness, bumpiness, specularity, reflectivity, transparency, and so on. Texture is the name we give to the perception of these variations. What we are doing here is introducing a synonym for “perceptual variability” or “surface discontinuity.” It is a kind of perceptual information complementing shape information.

Computer Systems that Learn

Inverse problems are among the most challenging in computational and applied science and have been studied extensively (Bunge, 2006; Hensel, 1991; Kaipio & Somersalo, 2004; Kirsch, 1996; Pizlo, 2001; Sabatier, 2000; Tarantola, 2005; Woodbury, 2002). Although there is no precise definition, the term refers to a wide range of problems that are generally described by saying that their answer is known, but not the question. An obvious example would be “Guessing the intentions of a person from her/his behavior.” In our case: “Guessing a past event from its vestiges.” In archaeology, the main source for inverse problems lies in the fact that archaeologists generally do not know why archaeological observables have the shape, size, texture, composition, and spatiotemporal location they have. Instead, we have sparse and noisy observations or measurements of perceptual properties, and an incomplete knowledge of relational contexts and possible causal processes. From this information, an inverse engineering approach should be used to interpret adequately archaeological observables as the material consequence of some social actions.

"Automated" Archaeology

The task of this automated archaeologist will be to assign to any artifact, represented by some features, visual or not, some meaning or explanatory concept. In other words, the performance of such an automated archaeologist is a three-stage process: Feature extraction, recognition, and explanation by which an input (description of the archaeological record) is transformed into an explanatory concept, in this case, the function of an archaeologically perceived entity (Figure 1). In order for the system to make a decision as to whether the object is a knife or a scraper, input information should be recognized, that is “categorized,” in such a way that once “activated” the selected categories will guide the selection of a response.

An Introduction to Neurocomputing

Let’s build an automated archaeologist! It is not an easy task. We need a highly complex, nonlinear, and parallel information-processing “cognitive core” able to explain what the robot sees, in terms of causal factors, which not always have an observable nature. Of course, such a “cognitive core” should not run like a human brain. After all, automated archaeologists do the same tasks as “human archaeologists,” but not necessary in the same way. Nevertheless, there is some similitude in the basic mechanism. My suggestion is that an archaeologist, human or “artificial,” will perceive archaeological data and, using some basic principles of learning, as those presented in previous chapter, will develop ways of encoding these data to make sense of perceived world. Consequently, we may try to build our artificial archaeologist based on the idea of learning and the ability to adapt flexibly epistemic actions to different archaeological problems waiting for a solution.

An Automated Approach to Historical and Social Explanation

It is obvious that answering the first question is a condition to solve the second. In the same way as human archaeologists, the automated archaeologist needs to know what, where and when before explaining why some social group made something, and how. That is to say, only after having explained why archaeological observables are the way they are in terms of the consequence of some social activity or bio-geological process performed in the past or in the present, the automated archaeologist will try to explain more abstract causal processes.