Early detection of Alzheimer's disease is vital for developing novel treatments. Attempts to identify the intermediate state between normal cognition and dementia have evolved over the past 50 years. Current taxonomies of mild cognitive impairment (MCI) may be criticised for their imprecise operationalisation. With the advent of biomarkers such as amyloid-beta positron emission tomography imaging in established Alzheimer's disease, much research has focused on establishing which factors predict progression from MCI to Alzheimer's disease dementia. In this review, we discuss the historical context of MCI before reviewing the literature of MCI subtypes and their risk of progression to Alzheimer's disease dementia. Finally, we summarise the literature and discuss limitations and weaknesses of how the construct is operationalised and implemented, before offering suggestions for development of the concept of MCI. We conclude that MCI must be empirically defined for the sake of its predictive validity to identify Alzheimer's disease before dementia develops.