Hacker News Reader

Comment by trabant00 | Hacker News Reader

An important point to make imo is that algorithms will not be close to human decision making until we hit AGI. No matter how complex they get until that point you can always cheat them in absolutely ridiculous and more importantly reproducible ways. Meaning scams scale and can do way more damage than a human error.

> Maybe this means web-scale sites like Amazon's marketplace aren't viable. I'm fine with that implication. Stores like Target and Walmart seem to have the same prices with human-curated inventories

Not only are they viable, they are hugely more profitable without the human expenses. This inhuman model - yes, I think that is exactly what it is - will displace everything else that simply can't compete in runaway capitalism.

View on HN

>Not only are they viable, they are hugely more profitable without the human expenses.

I don't believe this. The reputation loss alone will cost you a lot of potential money. Stadia is basically dead because the Google brand is completely untrustworthy. B2B customers want to avoid Google as much as possible because it's just a liability that threatens to ruin their business.

I think this is key. When Amazon becomes widely known as untrustworthy, then it will fail and fail quickly. There is already a _lot_ of counterfeiting on Amazon, and a lot of people avoid it for anything safety critical or areas where it is most prevalent. If that becomes more widely known and widely practiced, it would make a dent. Small places naturally verify their stock. Amazon doesn't.

McDonalds has a reputation for serving crap food, yet its franchises are everywhere, and they print money hand-over-fist.

I don't expect Amazon to go anywhere, even if people's impression of drops.

In a non-dystopian world, though, the algorithms can make life way better. We could have filters that catch more scams and abuse than typical humans do... and then reasonable humans behind that are responsive and fix the carnage for where they go wrong.

Of course, that's not what we've got...

Human-equivalent AGI is not sufficient for doing better than human moderation, despite all the problems with current generation AI content moderation.

The reason it isn’t sufficient to be human-level is because individual humans have all kinds of biases and blindspots which scammers design their scams around.

I’m kinda thinking it may be a necessary prerequisite, as scams could involve anything a human wants, but it isn’t enough.

It doesn’t have to displace everything else, it simply has to be regulated by law. A landlord cannot come to your physical store and remove few products. The same logic must apply here, with the platform providing sufficient means for interaction with local authorities in place of registration of the business.

> you can always cheat them in absolutely ridiculous and more importantly reproducible ways

Given that's possible with human moderation too (cf tax codes), what makes you think AGI will solve the problem?

> Not only are they viable, they are hugely more profitable without the human expenses

I agree with you, but what I meant by "viable" was more from a consumer perspective.

> This inhuman model - yes, I think that is exactly what it is - will displace everything else that simply can't compete in runaway capitalism.

While I think you're mostly right, Amazon is eventually going to burn enough buyers that they go back to the more trustworthy brick-and-mortar stores.

Most people I personally know have already done this. They may search for something on Amazon, but they will go and buy it somewhere else.

I'm obviously part of a very specific subset, but it's going to hurt Amazon eventually, especially if the US takes a more consumer-friendly regulatory approach than it has for the past few years (which it will).

Reply on news.ycombinator.com