This is a presentation of some of the basic points I plan to make in a chapter on "phonetics in phonology" for the forthcoming new edition of Blackwell's Handbook of Phonological Theory, edited by Goldsmith, Yu and Riggle. Some of you will have heard some of this a couple of years ago, when I tried out ideas for a reply (never completed) to Port and Leary's article "Against formal phonology" in Language, but the context is new, and I'd be very grateful for feedback. The chapter is overdue but still not finished, and the presentation will be pretty informal.
The basic idea is that IPA transcription (and, presumably, alphabetic writing more generally) led linguists and phoneticians in the first half of the 20th century to idealise the stream of speech as a string of "phones", even though many of them were aware that there are many things wrong with this idealisation. (For example, Pike worried in print about the idealisation, and Bloomfield rejected it out of hand.) The "phone" idealisation then took on a life of its own when it became a central part of theoretical phonology (think of allophones and complementary distribution), complete with the idea that there is a specifiably finite universal language-independent set of phones available for use in language (cf. Chomsky's "systematic phonetics", which was simply a formalisation of mid-centiry practice in phonetics and phonology). Many theoretical ideas and research questions make sense only if phones are real (e.g.: "biuniqueness" in classical phonemics; built-in "feature detectors"; a universal constraint set that deals with phonetic output; the idea that babies are born with the ability to perceive "all possible" phonemic distinctions and gradually lose the ones they don't need; etc.). However, accumulating evidence that these ideas don't make sense has led some people to doubt the idea of the phoneme as well (in so-called exemplar models), which I think is a clear case of throwing out the baby with the bathwater.