82,853
edits
Software-as-a-service: Difference between revisions
no edit summary
Amwelladmin (talk | contribs) No edit summary |
Amwelladmin (talk | contribs) No edit summary |
||
| Line 26: | Line 26: | ||
Configuration problems tend to be ones that attend the use of [[artificial intelligence]]. Again, [[thought leader]]s<ref>See {{author|Daniel Susskind}}’s {{br|A World Without Work}} for the classic case of this.</ref> see the ''abstract'' appeal of [[artificial intelligence]] while disregarding the challenges of the ''particular''. Neural networks need to be ''trained''. This is a slow, laborious, painstaking job. To extract a [[Rehypothecation|re-hypothecation]] formula from a database of thousands of random [[prime brokerage agreement]]s, for example, is doable — if you have on hand a specialist with a sophisticated understanding not only of legal language and market practice but also of regular expressions and PHP, who is prepared to spend weeks training the neural network to get it started. | Configuration problems tend to be ones that attend the use of [[artificial intelligence]]. Again, [[thought leader]]s<ref>See {{author|Daniel Susskind}}’s {{br|A World Without Work}} for the classic case of this.</ref> see the ''abstract'' appeal of [[artificial intelligence]] while disregarding the challenges of the ''particular''. Neural networks need to be ''trained''. This is a slow, laborious, painstaking job. To extract a [[Rehypothecation|re-hypothecation]] formula from a database of thousands of random [[prime brokerage agreement]]s, for example, is doable — if you have on hand a specialist with a sophisticated understanding not only of legal language and market practice but also of regular expressions and PHP, who is prepared to spend weeks training the neural network to get it started. | ||
Not many people do. Certainly this is not the [[general counsel]]’s expectation when, during the pitch, {{sex|he}} is presented with a magical [[machine learning]] application which, he is told, will solve his every problem. | Not many people do. Certainly this is not the [[general counsel]]’s expectation when, during the pitch, {{sex|he}} is presented with a magical [[machine learning]] application which, he is told, will solve his every problem. | ||
Note two ironies here: first, the lion’s share of the effort, expertise, and experience required to make a [[neural network]] ''work'' comes not from the product itself, but from those inside the client who painstakingly train it. Given how long the training process will take, this is no small cost — yet, being “sunk”, it will fall off the costing projections when the business case is put together. | |||
Second, this [[subject matter expert]] training is ''exactly'' the sort of thing that — if a client permits it — the [[vendor]] can harness to improve the product for its other clients. To be sure, there is a ''[[quid-pro-quo]]'' here: the client, too, will benefit from the training the product receives from the hands of the vendor’s other clients, but this only sharpens the irony: the value the [[vendor]] itself provides is minimal: merely an application interface on top of open-source [[neural network]] technology. What turns a public utility with a glossy front-end into gold-dust is the distributed training the application receives, ''from the clients''. Yet, the [[vendor]] gets to bill the clients, and not the other way around. That is ''truly'' “indistinguishable from [[magic]]”. | |||