I also question whether we are talking about algorithms, or the data set they are working with or the model created from that. I can forgive confusing an algorithm with its implementation (e.g. the source code), but this goes beyond that.
Machine learning is somewhat unique in the software world in that the actual useful artifacts are not necessarily strongly tied to the source code itself. You can have the identical source code running at two different companies, but by supplying them with two different training sets, you'll end up with very different outputs. That's what algorithmic destruction is targeted at--not even necessarily the source code or algorithm in the technical sense (you can't destroy "KMeans" or "convolution" as a concept, obviously), but both the data and the model weights that are produced through the use of that data that are used in perform a business action. Those weights are typically stored separately from the source code, and can be extremely expensive to re-create from scratch.
> The FTC’s new enforcement weapon spells death for algorithms
makes it sound like running quick-sort is a federal crime
Its pretty clear cut tbh. An algorithm is a set of steps to follow to produce some output. A trained model is, 'hey do these matrix multiplications with these coefficients to get an output'. The fact that the exact coefficients were arrived at via backprop, doesn't make it not an algorithm.
Indeed it is - for one thing, it allows us to see that various useful theorems and results about algorithms and computability apply as much to large programs as to small ones, such as the fact that there's no fundamental impediment to porting them between computers with different instruction sets, or running them in virtual machines.
What's not so well or usefully defined here is your distinction between hard and soft computing.
> In other words, a statistical algorithm does not point to the same category as a deterministic algorithm and an algorithm refer to the later class by default.
You appear to be under the misapprehension that the set of statistical algorithms is disjoint from that of deterministic algorithms. I strongly suspect that all the algorithms covered by the article are both statistical in terms of what they compute and deterministic in terms of how they do it.