Re: real challenges still facing human civilisation – regulatory compliance.
I doubt machine learning can deal with those high level requirements that I just referenced
Whilst I'd agree with that view, don't forget that IBM will be selling this as a service. If you're a company operating in a heavily regulated market, you can expect multi-billion fines with some regularity, even if you make a halfway decent attempt at complying. The complexity is too great to avoid breaches, and then you often find that the rules were written by the same people investigating, prosecuting, and judging. And even where they don't often benefit from the fine, they often use this as a metric of their "success" in regulating their sector.
Against this backdrop, a machine learning approach becomes more attractive - you can do things like (for example) scan all sales calls with VR software looking for patterns and indicator words, or (with enough grunt) an attempt at interepreting the language into its spoken meaning. That's great for IBM - so long as the system can flag up enough convincing cases to investigate, it will be seen as a credible purchase by the client company. From their directors point of view, although the business case will assume the elimination of non-compliance, they will know this is just tokenism that won't catch the worst egresses. But what it offers those directors is a fig leaf to show the regulator. And in most regulatory enforcement models, when you get fined, the size of the fine is greatly affected by things like keeping records, having adequate systems, using systems to find and target fraud and non-compliance etc.