Original link: https://pt.plus/notes-on-ai-4/

This “AI Miscellaneous Notes” series has written the fourth article. Looking back, the first article is basically a literature review of an introductory paper; the second article turned over the pile of old papers and looked back at the history of AI development; the third article put forward an optimistic signal that Language is all you need.

The fourth part will focus on the analysis of man-machine relationship.

Software Layer

An article titled Replacing Middle Management with APIs caught my attention. The main reason comes from the management dilemma in reality. After many adjustments, dozens of people reported directly to me. I joked that this has almost become a “miracle of management” – I don’t know if it is a “miracle”, it must be a dilemma. I discussed with the person in charge of HR that the general management bandwidth is about 7-8 people. I saw an internal McKinsey manual, Daniel on McKinsey, which more strictly stipulates that the ratio between partners and subordinates should be The ratio of 1:4 can ensure that partners have enough time to give each subordinate and provide corresponding guidance and training. Obviously, under the existing organizational structure, this ratio obviously cannot meet such extravagant requirements.

And the title of this article sounds like it paints an exciting scenario: using APIs for middle management. After reading it, I found that it was written in 2015, using Uber and Mechanical Turk as examples, and proposed the concept of software layer:

The software layer between the company and their armies of contractors eliminates a huge amount of middle management, and creates a worrisome disconnect between jobs that will be automated, and jobs of increasing leverage and value.

The software layer between companies and their contractors removes much of the middlemanagement and creates a worrying disconnect between the work being automated and the work of ever-increasing leverage.

In the article, the author tries to prove in the form of “pseudo-code” that programs like uber.drive(card, pointA, pointB); depersonalize simple laborers, while advanced laborers who write codes can only learn from some highly abstracted Make policy decisions in trending or categorical data, change some settings in the code, but the latter has difficulty understanding how these code changes will affect the former’s life.

The emergence of software faults happened exactly when Marc Andreessen shouted that Software is Eating the World (2011). However, the word software, compared with hardware, still seems to have some vital signs, which contains the will of the human creator. Looking at it today, software is indeed alive, but it may emerge with its own will.

This article is transferred from: https://pt.plus/notes-on-ai-4/
This site is only for collection, and the copyright belongs to the original author.