Comment by Futurebot | Hacker News Reader

Comment by Futurebot | Hacker News Reader

Parent Full thread

Right. Part of what's missed is that we don't need great GAI to get rid of lots of jobs; weak "AI" / deep learning / related automation techniques get us a long way there. We have existing examples in law document review, article topic summarization, etc.

Another thing which is often missed is how complementarity may play out for certain jobs: it's NOT going to necessarily be "program / robot replaces every job X", but instead "program / robot allows one person to do the job previously done by 50." You can still have massive unemployment without getting rid of every worker doing a certain job (and their job may change more to be a machine guide / manager / error corrector.)

An example I see all the time now is in supermarkets. You have self-checkout lanes that are overseen by one person. It used to be 5 lanes staffed by people. Now you only need one to intervene when something goes wrong. 5 workers have become 1 without eliminating the job "register person" completely. Instead, they've changed into "auto register checkout manager" while everyone else got the ax.

And on the topic of register checkout manager, you can easily see the natural progression towards the elimination of the profession entirely.

The registers get more self sufficient as the software matures. The security systems become cheaper and more accurate to eliminate the lackluster security effect a clerk has standing there. The clerk is removed entirely. Then the shelf stocking is automated with computer vision and maybe magnetized sticker guide rails in the floors.

And then people realize its stupid to go to a store to buy stuff when you can virtually tour a mall of everything and have whatever you want shipped to you. You order it, it goes through computers without ever interacting with a person, and a manufacturer ships you it instantly.

And that process gets automated too. The transport goes from self driving autos transporting your goods using standardized automated transfer mechanisms to your automated mailbox to a fabricator in your own home that comes from Star Trek.

None of that (besides the fabricator part at the end, thats going a bit heavy) is nothing novel or even new. It exists. It just takes market pressure and time to make it economical to implement and for culture to accept it. Because it is more efficient, and it will inevitably happen because its better in every way except the "but people aren't doing it!" angle. It does not take AI sentience to move boxes or see dirt on a floor through dictionary lookup and fuzzy logic processing.

In the developed world, I think we'll be seeing the hybrid approach for the next few decades at least, but yes, we will get past that to your fully automated vision. Also, the non-evenly distributed future effect means we'll likely see that vision in SF in, let's say, 2100, but won't see it in Flint, MI until 2150. It is coming, though.

Reply on news.ycombinator.com