David Chalmers has argued against the thermostat view, suggesting that adaptation to the environment is not sufficient. John Searle also disagrees that the current state of machine learning is capable of consciousness on the grounds that information processing is not a sufficient criterion (public lecture, 2016). Both of these philosophers emphasize clarity in their presentation of arguments and I highly recommend reading some of their books for a general introduction to current problems in consciousness and perception.
Complexity
In 1998, Tononi and Edelman published a science paper on Consciousness and Complexity. One of the main proponents, from the same lineage of researchers, links the complexity criterion to consciousness: Anil Seth and colleagues have suggested that a measure of complexity is central to explaining consciousness (pdf). His works are probably a good starting point if you are interested in a possible link between (abstract formulations that are relevant to both) artificial intelligence and consciousness.
Recurrent Processing
Although "recursive complexity" is a concept in Anil Seth's work, if you are particularly interested in the role of recurrent processing for consciousness, Victor Lamme's work on vision could be a starting point: Lamme & Roelfsma (2000)
Et cetera
For various views and perspectives, visit the most complete collection of references on consciousness online: http://consc.net/online/