Here is my reading list over the last week:
- Giuseppe Sollazzo has one of the best jobs, and best newsletters. He is head of AI skunkworks at the NHS and has a newsletter 'a quantum of sollazzo' https://buttondown.email/puntofisso/archive/451-quantum-of-sollazzo/. In this piece he is interviewed about whether the NHS can do AI?
- tl;dr yes, but cautiously
2. This is a repot on a Turing institute conference about data science and AI in the age of COVID
It's longish, and a multi-column PDF :(
- COVID is the first pandemic to occur in the age of AI
- Many people working in AI did things related to COVID
- There was a conference run by Turing, it had many people contribute to it
- New data sources were created, including making available patient records for 58M patients
- existing databases were expanded to be useful
- genomic sequencing happened at scale (490K genomes at the time of writing of the report)
- Data on movement of people was made available allowing interesting models on transmission to be created
Those were some of the broad - covid specific - highlights
- Data access remained a problem
- data linking and standardisation was a problem (plus ca change)
- This lead to recommendations around data sharing that are - frankly - the same recommendations that have been present in this debate for some time.
Inequality and exclusion were highlighted in the workshop
- This feeds into sampling bias
- ONS data is perhaps not as robust as we would like
- (on a side note during the pandemic myself and my wife were thinking about whether you could even determine what the population of London is at any level of accuracy, and we realised that this is a hard hard thing to do, which effects things like estimations of incidence rates. We though perhaps you might be able to make some determination by measuring volume of poo going through the sewerage system a poopulation study if you will. That doesn't work because while people might work from home but still poo at the same amount as when they were in the office,
Lack of transparency around policy making was identified as an issue. I don't even know what to say about that.
3. On a totally different topic - the technology we use today is a little bit magical, and how it works under the hood is a little bit hidden from us. This blog post covers how some naive assumptions led to a large bill around deleting things from the cloud: https://www.cyclic.sh/posts/aws-s3-why-sometimes-you-should-press-the-100k-dollar-button
4. I've be fascinated by the idea of mob programming for a while, but have never had the chance to try it out. This is an interesting write up on how it has been going at eLife - mob programming - Invisible to the eye. There is not enough in this write up to help you determine whether you should do this in your team or company, as indeed there shouldn't be, but there are some interesting perspectives here to consider.
5. How do you measure value? In science? In software? In open source? The github policy team have raised the last question - https://github.blog/2022-01-20-open-source-creates-value-but-how-do-you-measure-it/, or should I say have raised a battery of questions. I don't recommend reading this blog post, it will leave you none the wiser, but it is interesting that they are asking these questions. I look forward to checking in on them when they have some answers.