The article doesn't explain what ELIZA effect means: the tendency to unconsciously assume computer behaviors are analogous to human behaviors; that is, anthropomorphisation.

https://en.wikipedia.org/wiki/ELIZA_effect

This is more specifically the assumption that a computer understands and more importantly cares about what you are saying, and maybe also you as a person. This assumption is obviously wrong for the computer, but also usually wrong when interpreting responses from an actual therapist, or any other human, so no difference really.

Why do you mention computers? In what way is it that actual therapists are like any other human? What does that suggest to you?

I see what you did there ;-)

Tell me more about what you did there.

I built a simple chatbot that uses GPT-3, which can be very convincingly human at times, so I can understand the tendency.

Related: I didn't see mentioned on the page, there is also a visual novel game called Eliza, named after the same subject, and the in game Eliza follows a very similar script to the text screenshot shown on the page. It's an interesting game if you like visual novels, and it explores a few other concepts beyond the human machine relationship.

https://en.m.wikipedia.org/wiki/Eliza_(video_game)

https://store.steampowered.com/app/716500/Eliza/

Eliza was a staple in every basic book and magazine ever published in the 80s and there were always copies in the computer labs, I’m sure I entered in this by hand at some point - https://everything2.com/title/ELIZA+source+code+in+BASIC

I knew a guy who wrote an irc bot called “Petunia“ that wasn’t even as sophisticated as Eliza - it would just make random statements and every time someone sent it a message it would be added to the list of statements. It would post everything sent and received on a web site - including a mountain of nsfw images received - would take some people 15-20 minutes to decide that Petunia was fake.

Awesome! Someone wrote Eliza in golang with a web interface https://github.com/mattshiel/eliza-go

I was wondering if this was going to be a commentary on GPT-3 (it's not), since much of the latest hype has a similar feel as ELIZA. Of course the verdict is still out as to what extent GPT-3 "understands", but it's clear that the spooky realism we feel when interacting with it is partly due to our imposing anthropomorphic interpretations, just like ELIZA.

Yeah and we have better training material for AI now. For example it could be possible to create a program that mimics Hackernews comments. There's more than enough training material on HN to create plausible-looking comments that aren't a bunch of nonsensical gobbledygook and that also pander to people's emotions.