Using AI for psychological therapy appears a ‘no-brainer’, no need to disclose anything, readily accessible, little or no cost. Given that NHS Talking Therapies has failed to demonstrate superiority over any active control condition, AI is an attractive option. AI’s treatment integrity, i.e faithfullness to a CBT protocol is no less evident than the philandering with protocols in NHS Talking Therapies. It is likely that AI will perform no better in terms of outcome than the tip-of-the iceberg recovery rate in NHS Talking Therapies [Scott (2018)].
An editorial in the British Journal Of Psychiatry by Shafran et al (2026) suggests that NHS Talking Therapies data could be usefully mined by AI. But fails to point out that the data has first of all to be ‘meaningful’. A test score [PHQ-9 or GAD-7] by itself has no meaning, it is just a number. AI is not Alladin’s magic lamp that can be ‘rubbed’ to transform numbers into meaning. The numbers per se are not credible units of analysis, the input into AI has to be something that the ordinary person would recognise, e.g becoming able to see at a distance following treatment. Numbers demand an explanation, NHS Talking Therapies data is ‘vapour-ware’.
Dr Mike Scott