 Felly, yn y cyfnod bwysig, ysgandl yn bohemio, yna'r Llyfr Gwladd, o'r Hwmau Sholach yn ystafell, yn ystafell y cyfnod ymwysig i wych i fyfodd y ddau ar gyfer ychydig. Ond yw'r cyfnod yma, mae'r ddaerol yn psychodraeth, yna'n mynd yn ystafell i'r ddau, i fy nghymru, roedd byddwn yn cael ei meddwl yn ystafell ar gyfer y ddau. Ond mae'r ddau. Yn y gallu gwlad, mae ydych chi'n ei gwybod o'r cyfnod ychydig yw ddau yn y form of electronic health records this data must be used to develop knowledge to improve patient care. But how should we be doing this? Now there's been so much hype about big data especially when paired with artificial intelligence that it's hard to think about how we cut through this hype to get to what is possible now and what will be possible in the future. At the Institute of Mental Health UCL we aim to fully understand the pitfalls of this data so that we can answer questions about what will and won't be possible. Artificial intelligence has excelled at prediction, shopping choices, movie choices, stock trading algorithms and there are times in medicine when we're particularly interested in prediction and times when we want to actually understand cause and effect which might be more complex. So in terms of prediction we are beginning to develop models to predict which adolescents might be at risk of depression which people with schizophrenia might be at risk of unplanned hospital admissions and which of an approved set of drugs might be most suitable for a person with bipolar disorder. There will be times certainly in the future when a machine will read someone's previous electronic health record and pick out the salient features like the comorbidities or illness characteristics and recommend a treatment to a clinician based on the chance that that person will respond to the treatment and the chance that that person might experience adverse effects. But the premise that all of these predictive algorithms will lead to better treatments is flawed. For example the prediction of adolescent depression doesn't tell us anything about the risk factors for depression or how to modify that depression. For this sort of thing we need to integrate causal inference techniques into machine learning, into artificial intelligence and this is more complex. But doing this we have been able to recreate medication trials in electronic health records which therefore we have people that are representative of the total population experiencing the disorder rather than people that are highly selected, these populations that tend to get into trials. We've also been able to identify drugs in common usage for physical health problems such as statins which might potentially be repurposed to treat psychiatric symptoms and because of the scale of the data we've been able to identify rare but serious adverse effects of some of these treatments. These data are therefore very powerful in a way that goes beyond their initial clinical usage of recording what happens between a doctor and their patient. They become even more powerful if we become able to augment this data and this augmentation might come from genetic data, biomarker data, data from wearables or other methods of passive data capture. We begin then to look at whole countries as laboratories to think about improving mental health care. Given the stage things are at globally with electronic health records the industry and researchers have the potential to shape this field for the benefit of all. As it stands most of the electronic health data comes from high income countries and it's unclear how insights generated here will transfer into other national contexts. We're just beginning to test some of these algorithms in other data sets from the data where the models were developed. The other thing is we've seen in other contexts that AI can mistakenly integrate racial, social or gender prejudices and where this to happen in medicine it could be particularly detrimental. If you want to take a truly global approach to population mental health we must consider how we approach variable health care systems, administrations and how digital technology is spreading. But the data science wave that transformed industry and commerce is now reaching mental health care. The potential benefits of substantial so exploiting such data is beginning to improve our understanding of mental illness causation and classification and beginning to contribute actionable analytics to health and health care. But the major challenges need to be addressed to realise the potential more fully. Thank you.