In this paper, I build on writings in psychology and philosophy to offer ecological description of jazz improvisation grounded in the analogy of navigation through a complex environment. I begin by contrasting this approach with the computational description of improvisation as algorithm of input, processing, and I continue with the ecological description itself. This description reframes many aspects of improvisation, including the (Pressing, 1984, 1988, 1998), the nature of improvisational skill, and the problem of temporal coordination between soloist and ensemble. Stylistically, I focus on straight-ahead, tonal jazz, the first style most students learn, which is foundational to much current jazz practice. The ecological description could be adapted to other styles. To bring the description to life, I offer evidence from exploratory study of improvisational errors.Because I offer something less than a falsifiable theory, I posit this paper's central premise instead as a description, which is primarily intended to encourage a shift in researchers' perspective. The result, I hope, is a picture of improvisational cognition that is richer, more complex, and harder to pin down-an opening, rather than a closing. Balancing this, I conclude with some predictions of the ecological description that could be tested empirically.The Computational View of ImprovisationThe computational view of improvisation stems from early theories of cognitive science that were inspired by the artificial intelligence research of the 1950s-60s. These theories take computerized information processing as a model for human cognition, so that information received from the external world is perceived, translated into a syntactic code of meaningful symbols, and processed according to a systematic set of rules, and body movements are mere outcomes of these processes (Maes, Leman, Palmer, & Wanderley, 2014, p. 1). In its strictest form, this view is now largely obsolete, as more recent research has revealed myriad links between perception and action, including in the realm of music performance (p. 2). But theories of improvisational psychology remain grounded in the computational view.Pressing (1984) was the first to develop a computational theory of jazz improvisation (p. 353; cf. Pressing, 1988, pp. 130-132, 1998). In Pressing's formulation, improvisation depends on two large-scale parameters. First, there is the referent: an underlying formal scheme or guiding image specific to a given piece (1984, p. 346; cf. 1988, p. 153; 1998, pp. 52-53). Second, there is the soloist's long-term memory, for like motives, scales, and arpeggios, and for techniques of combining and developing objects (1984, p. 355; cf. 1988, pp. 161, 166; 1998, pp. 53-54). Given a referent and a sufficient memory store, the act of improvisation proceeds as a self-sustaining cycle through three stages, input, processing, and output (1988, pp. 152-166). The soloist's previous actions, along with the referent, constitute the input. To process the next action, the soloist continues, develops, or alters one or more of the input's features (1984, pp. 350, 353; 1988, p. 157). At some point, the action determined by the soloist's processing crosses a cognitive Rubicon and becomes irrecoverable output. The soloist then acts without further sensory or central intervention, though there remains limited scope for local fine tuning until the action is complete (1984, p. 355; 1988, p. 153). This action, as output, in turn becomes the input for the next action, and the cycle resets.Johnson-Laird (2002) presents a different computational model. Whereas Pressing's improviser relies on a memory store of prelearned objects and processes, Johnson-Laird's has memorized no objects, instead relying on memorized rules of melodic construction within a chord progression (pp. 422, 430). The algorithm focuses on choosing notes appropriate for the harmonic context and placing them in effective contour and rhythm. …