Abstract:

After defining instructional objectives and set of variables, designers often creatively combine teaching strategies, methods, tools and everything is useful to promote learning. This “creative” process is ineffective in organizational contexts that require standards and teams of interchangeable designers. Moreover, it can rarely be adapted to different situations, standardized and reproduced, according to objective criteria. Finally, it often results in the trivial contrast between face-to-face learning and online instruction, the confusing fragmentation of environments, tools and methods, or the teachers’ free improvisation.

In contrast to this naïf vision, we propose a systemic interpretation of instructional methodology based on four key assumptions (A) and a hypothesis (H):
(A1) the methodological option of a training program – the output of this phase - can and should be systematically represented through a multidimensional vector associated to each learning unit, always including the following 13 components:
(1) type of training activity,
(2) learning environment,
(3) physical communication tool,
(4) duration,
(5) studying/learning scheme,
(6) interaction/engagement methods,
(7) media and communication channels,
(8) feedback management,
(9) instructional plan,
(10) communication style,
(11) type of teaching/tutoring,
(12) digital and
(13) analog educational material;

(A2) in order to generate the corresponding output, the instructional designer should process a vector of input data with 15 dimensions including the following key components of a training program:
(1) observable behavior (instructional objective),
(2) condition and degree of
(3) accuracy,
(4) time limits and
(5) quality,
(6) knowledge level,
(7) content type,
(8) evaluation method,
(9) assessment tool,
(10) evaluation type,
(11) motivation level,
(12) inhomogeneity level,
(13) learning environment,
(14) number of participants,
(15) maximum learning time;

(A3) the states of each input or output variable should be standardized and configured in a discrete way through tables of predetermined values, without leaving any freedom to “creatively” interpret the insertion of conditions and/or the encoding of results. Obviously, these tables should effectively represent the typical situations of training contexts;

(A4) input processing and output generation are based on the integration of the instructional designers’ community experience and the theoretical framework of reference. This process should be standardized and reproduced by a software that simulates the “ideal” designer’s steps from the objective to the instructional methodology definition.

(H) Based on the complexity level of this step, we formulated the hypothesis that this process could be simulated by an architecture based on Artificial Neural Networks (ANN), trained through supervised (Multi-Layer Perceptron) and unsupervised (Kohonen Maps) learning. ANN training could be supported by a dynamic database of “effective” design examples to train the synaptic connections on the best functions between input and output.

To demonstrate this hypothesis, we implemented a software system – AID 0.1 Alfa – that can simulate this process: it reproduces the instructional designer’s flow of reasoning, starting from thousands of positive experiences wired in the synaptic weights of its artificial brain. Like an artificial instructional designer, AID 0.1 Alfa supports the designers’ team.

Keywords:

Artificial Intelligence, instructional methodology, Kohonen Maps, Multi-Layer Perceptron.