Bill Harvey: AI is not machine learning

There has recently been much publicity about the advance of machines and their potential for replacing humans in the workplace. Ophthalmologist Pearse Keane was discussing the Moorfields collaboration with Google offshoot DeepMind on national news media only this week. This project is using software to analyse OCT data. Rapid assimilation of information should, he suggests, allow the rapid triaging of OCT scan data so that anything outside the norm might receive rapid, appropriate attention.

This does beg a few questions. Obviously, first of all, one might argue that initial primary eye care triaging is currently a key optometrist role and should require more than just OCT data. Secondly, to pass the Turing test, a machine has to show evidence of actual free and original thought. Initiating a response that may change based upon cumulative assimilation of existing information is machine learning and not true AI. Surely this is not dissimilar to, for example, the use of normative databases in current instrumentation. We have all moved away from the passive obedience to colour coded outputs from machines such as OCTs as they notoriously mislead when structural change outside the norm can be traced back to non-disease anatomical variation. Similarly, OCTs very accurately detect changes over time but rates of change can be differently interpreted.