Deep Learning - Pushing the boundaries of health AI. How do we make it fair and the data safe?

Over the last 5 years there has actually been a confluence of a few different historical threats. We’ve had health data being increasingly digitalised and we’ve had the proliferation of accessible massive scale computing, both of which have un-locked a technique developed in the early 80’s called deep learning, which is really good at pattern recognition over large data sets. Key trends in the last year include the first randomised clinical trials in the clinical application of AI in health, the potential for AI in clinical discovery particularly using multimodal data (including electronic medical records, imaging data, genomic data) and combining that to find patterns in very large data sets. This is the real beginning of precision medicine. Finally there are day to day clinical process applications being used to predict resource allocation or disease outbreaks. At the same time there are some systemic challenges facing AI in health, including workflow integration, bias, equity and just access. How can we mitigate these biases and make them fair. Finally how do we make this sensitive data safe? Is the answer Federated machine learning where we send the AI algorithms out to local networks and apply them there? For more head to: codachange.org/podcasts