Active Structure is essentially an analytic or planning tool, used to represent the objects and relations in some domain. To find applications that people will accept, we have continually "upped the ante" until we are now closing on high level cognitive activity, but in areas that people don't like thinking about.
Contracts and leases are written in a dense, highly clipped style, with many references to other parts of the document or to other documents. They are tedious to read, as the meaning does not leap off the page, but needs to be unravelled bit by bit. If one is unfamiliar with the particular contract, one can't jump to a relevant clause, because that clause will use terms defined in preceding or even succeeding clauses, and those definitions... - until you have to read the whole contract to be sure what it is saying.
What typically happens is that people smear contracts together in their head - they operate on a generalised contract because they don't want to spend the time going through an individual contract in detail. This leads to mistakes, and the mistakes can run into millions of dollars, but it is just not possible to force people to read these things. A lawyer may be willing to read it, but he/she will usually only have the most superficial idea of the business processes it describes, and there are enough mistakes in the typical contract to indicate that even lawyers don't like reading this stuff. The contract will usually have been cobbled together using clauses from other contracts using different defined terms, or amendments made over time - another large source of error.
By extracting the meaning structure of the document, and checking and linking all the references (a "legal compiler", in the computer language sense), the machine now has an excellent basis for answering questions about the contract. This is effectively the first time that the machine will know more about the subject at hand than the user, and the interface needs to reflect the change in "who knows what".
Specifications for complex systems themselves become complex systems. With the human limitation of between six and nine pieces of information in play at any one time, dealing with a specification with hundreds or thousands of things potentially in play is not easy. A machine does not have this limitation, so if a specification can be turned into something a machine can manipulate, fewer errors should result.
Searching through voluminous emails looking for evidence is a difficult assignment. Using a machine to read the emails and extract their semantic structure can save time and do a better job.
Medical records en masse can be used for useful research on their contents - what they mean. This is vastly different to finding key words in them, and assuming the document is concerned with those key words in a way you might assume without reading the document.
See Medical Note
The knowledge in certain types of call centres is held in text - in legislation or insurance policies - and can be quite tricky. It needs the interaction of the caller's current or future state with the knowledge in the text for decisions to be made and advice given. Queries can range across - find the context, clarify an object, return a logical value, return a number based on a sequence of steps. The questions can form a dialog, building on the answers to questions already asked.
See Call Centre Support
Why do these applications first?
Because they are ones that no-one likes to do and they are relatively easy - legal documents or specifications have reasonable spelling and grammar, a specification or contract will usually be written to minimise ambiguity, and there can be huge leverage. However, some contracts and specifications involve intention, which is usually lacking in scientific or medical text.
The Medical Note application is trivial in comparison with contracts or specifications, but being easy does not mean it is not valuable.
The Call Centre application is a continuing thorn for organisations handling complex interactions with their clients - tax, health insurance.
Other applications in medicine or genetics or general legal may encroach on what the user feels is their specialty, so there is tension about shedding some knowledge-handling activities to a machine.
For general law, see SOX presentation.