Everything you see these days is AI enabled in some way, or according to the marketing they must be. Software, fridges, cancer detection and lots of other examples are all based on some kind of AI implementation. Google, Microsoft and all the big players are heavily invested in at least the buzzword, but the proof of delivery as promised is elusive.
- One notable example, just like those old cold fusion claims, concerns a number of expert level doctors now criticising Google, and others, to demonstrate some of the big claims in AI research and share the code and models so that others can replicate the work. The response to date has been a firm "no".
- The latest round started with a paper in Nature from the Google Brain team. It claimed that its artificial intelligence was better than humans at detecting breast cancer in mammograms. This was widely reported by the media but to date, doctors in the field have not seen the evidence to back up those claims. Without the code and data, it is not possible to replicate and would take months to independently duplicate in any way. Also key in such systems, but not provided, are the hyperparameters and the training components. The doctors who wrote a critical paper concluded that Google's "publication of insufficiently documented research does not meet the core requirements underlying scientific discovery".
- Google responded that it was "early stage research" and that "we intend to subject our software to extensive testing prior to its use in a clinical environment". It also described the doctor's observations as a "thoughtful contribution". A cynic might think that Google was holding back information for commercial reasons despite their comment that "because liability issues surrounding artificial intelligence in healthcare remain unresolved, providing unrestricted access to such technologies may place patients, providers, and developers at risk". I'll leave it up to the gentle reader to draw their own conclusions on the trustworthiness of Google.
- My new Onyx Boox Note Air E Ink device has arrived and it is impressive. Spec wise it has a 10.3-inch, 227dpi, 4096 pressure sensitive levels screen. It comes with an 8-core CPU, 3GB of RAM and 32GB of ROM. It has great WiFi support and Bluetooth 5.0. It is essentially an Android 10 greyscale tablet with some special software for note taking and e-book reading. The first thing I did was to install the Kindle app that synced to my Amazon Kindle account. The handwriting recognition works well and response is fast and smooth when using the device. The new pen works well but you can probably use your favourite pen if you already have one for another device. Battery life seems impressive so far and I'll post a report when I've tried all the bits and pieces out.
- How do you fight something like the new trend #DEBUNKED? This is now a pre-emptive tactic by some groups to stop discussion of important issues, because who wants to read something that has already been debunked. People rarely check further and see who debunked it so they may never find out the truth. Poor fact checkers now include Snopes, Wikipedia, Huff Post and others along the same line. The debunking itself can be quite sneaky. Let's say someone claims that Fred smashed his phone. If it turns out that Fred's aide did the actual smashing, the claim is #Debunked and flagged as a conspiracy theory. In fact, Fred's phone was indeed smashed, at the direction of Fred to an aide, so this is a sneaky and functionally inaccurate debunking but it is pushed as such by many. Be very wary of any #Debunked flags and always check into the facts, especially the context, for yourself.
- Seven countries, Australia, Canada, New Zealand, USA, UK, Japan and India, in interests of "public safety", have asked technology companies to design their products so they offer access to encrypted messages and content. While on the one hand praising strong encryption for personal security and intellectual property, they also claim that it can be used for more nefarious purposes and so enforcement agencies should be able to view those encrypted communications. The perfectly reasonable targets are areas like child exploitation and violent crime. They also claim they will let technology companies know if their terms of service have been violated if they find such materials in the decrypted content.
- This idea not new with the UK Snooper's Charter, the USA's CLOUD Act and Australia's Assistance and Access Bill all targeted at a similar purpose. The combined statement from the seven would indicate that these are proving to be ineffective as once the access is granted, they then run into the problem of the encrypted content. The discussion of privacy versus access is a much wider one than can be covered here and I expect this will continue to be a hotly divided topic for some time to come.
James Hein is an IT professional of over 30 years' standing. You can contact him at jclhein@gmail.com.
