Colour scheme generation by software seems to be as random as everything else (see Tim Head’s work). Colour theories are cultural products and random colour is no exception. For random colour to be interesting the viewer must either see some intention into the work, or enjoy exercising their taste unchallenged. A program that works with colour in a directed way would be more interesting.
This program would colour chords to aesthetically satisfy a set of technical or referential demands, making colour schemes that have a relationship to external conditions (visually referential, expressive, social) using a set of discrete colour spaces, a colour ontology, an emotional ontology, a compositional ontology, a relational slipnet and a set of referential palettes (natural colours, city colours, clothes colours, human colours, art colours).
Valence the ontologies and start building, weigh colour proportions and relations, vary colours and backtrack to get things right for each set of conditions. Colour graphical microdomains with the colours. Start 1D (stripes) then go to 2D.