←back to thread

693 points hienyimba | 2 comments | | HN request time: 0s | source
1. DanielBMarkham ◴[] No.28523336[source]
There's an interesting efficiency/reverse-opportunity-cost issue here.

If you set up your ML so that it works x% of the time, you might very well have a profitable business even if you end up accidentally screwing over a bunch of folks. But no competitor can challenge you in the marketplace because the human cost of answering phones and emails to find that last little bit of efficiency is overwhelmingly disproportionate to any economic value the business would gain.

Many of us like to bang on businesses as being amoral and impersonal, but most are trying to do something people want, only better and more efficiently. ML may be providing an upper limit to efficiency by taking out any opportunity to do some serious analysis. Because in many cases removing that last 1-5% in inefficiency is the bit that leads to a completely new way of working, in many areas we may be boxing ourselves in to a very long-term status quo.

replies(1): >>28523704 #
2. quesera ◴[] No.28523704[source]
Yes.

You can build a system that efficiently serves the 98% case of "simple" customers. Then you can ignore the 2% unprofitable/complicated customers, forcing them to go to other vendors.

If you're big enough, you starve your competitors of the low-cost/simple customers. So their cost structure goes way up, which in turn prices the services out of reach of all other customers except the stupidly profitable, which is to say: gambling and porn.

(This has parallels to the USPS v. FedEx/UPS problem in the US, with the exception that the USPS is required to serve all customers, so no one is completely without service)