 I think from my perspective the most important advances recently in terms of AI for healthcare haven't necessarily been technical advances but more within the clinical field the realisation of the potential for this field. So when we think about healthcare it has to be quite conservative. Clinicians therefore tend to act in a more traditional way. But I think what we've seen recently when we think about AI for healthcare is a number of very high-profile published papers or high-profile cases that have really illustrated the potential of the field and those have tended to be in what are probably more the kind of maybe the easier use cases to think about so things which have an established link between either an image or a diagnostic image and then an outcome so thinking about things like radiology or ophthalmology we've seen interest in retinal scans. I think the other area that we can think about artificial intelligence for health now that we kind of are beginning to see and understand the potential is when we think about these other kind of less traditional models of kind of classification so thinking about things like how would we predict epidemics, how could we predict disease from electronic health records, how could we use blood tests that are taken for other cases to predict disease and that's something that would be different from what's currently done. I think the third way that we can think about AI for healthcare and its potential benefit is to think about the non-traditional kind of non-clinical use cases so things about patient flow around the hospital about staffing which might not have the kind of instant appeal of some of those very clinical use cases but then are also a kind of hopefully a lower risk way to benefit from that technology and healthcare. Yeah so I think you know going back to the comment around clinicians and the kind of clinical community being relatively conservative and of course you have to be when you're working in that environment because the stakes are very high so the kind of the traditional pathway from an innovation to being adopted involves regulation, it involves academic studies which look at the comparative efficacy compared to other things that are published in journals like The Lancet where I work for and they fit into these phases of pre-clinical trials working through to a randomized control trial those then end up in guidelines and physicians adopt them broadly speaking. I think when you think about how that applies to AI well it's much more faster moving currently there's probably much less regulation and those are kind of potential limitations for the adoption of AI for in healthcare spaces because of the naturally kind of conservative inclination of clinicians but you know there's also lots of advantages so when you think about the scale of data we've traditionally worked with very small time consuming expensive randomized control trials but you know flip that around and think of the benefits of that model. Yeah so the the work of the focus group to offer the benchmarking tools really fits into that idea that I just explained about clinicians working within this framework of regulation and published research so the way that you would expect something to undergo certain checks and balances before it was implemented in a clinical population and really that's the benefit of the work of the focus group thinking how we could work to produce a kind of international benchmark that would be a kind of standard by which algorithms needed to pass before they could be adopted into practice or tested in wider kind of comparative studies. So the remit of the focus group to produce this benchmark is ambitious and you know many of the people who we've had giving expert advice today here at the meeting have kind of commented on various different aspects of that so I think the work of the focus group in 2019 especially given the time that it has to run is to really identify a couple of use cases that would be the most obvious to begin to apply the benchmark and then to kind of begin to set in place these operational questions around okay well how do we ask people to submit data in which form is that and how do we offer people to test their algorithm but then also understanding like how do these two communities work together the clinical community and the data science community and what's the appropriate standard or the appropriate outcome for the benchmark.