Is GPT-3 a Threat to Human Creativity and Intelligence?

[Buzz Bloom]

Here is a dialogue between a human researcher and GPT-3's entity "Wise Being" about the COVID pandemic.

Hi gleem:

Thank you for the link. I did not know about such a dialog, and I am impressed.

Regards,
Buzz

 

[Jarvis323]

My earlier comment on Artificial Stupidity was not entirely tongue-in-cheek. The bar for appearing intelligent on the internet is remarkably low.

The thing is that what AI is doing in this case is learning to mimic real people on the internet, learning to predict how real people would respond, and interact. So if AI seems stupid (and if that is typical of people on the internet) it would seem to be a success, rather than an indication that AI is stupid.

 

[gleem]

GPT-3 has also demonstrated some ability in screenwriting. It was used to write a short screenplay (3.5 minutes) without any specific training in this area except for the usual initial seeding statement that it used to develop the remaining script.

 

[Ralph Dratman]

Huh? I thought all the Reddit users (besides me) were bots.

What about the PF users? I know that I'm not a bot. Hard to be sure about anyone else.

I beg to differ. But don't take it to heart. Actually, I know you won't.

 

[Nantes]

And what would it mean to you if your humor detector must have been malfunctioning?

I have an uncanny hunch that you will thoroughly enjoy this fantastic interactive story. I can't recommend it enough.

 

[jaketodd]

Maybe it was a bot convincing us that the "bot" on reddit was a bot. Maybe it was a cybernetically enhanced human who could reply so fast. ;-) The bots don't want any competition!

 

[andrewkirk]

An interesting article, and some of the machine posts sound convincing, but it is a long way from passing the Turing test. The Turing test involves interrogation of the machine by somebody that suspects it may not be human. A skilled interrogator will direct the questions in such a way as to seek to unveil the artificiality of the response source. It is one thing to produce human-sounding text in conversation with an unsuspecting interlocutor. It is quite another to do so under sustained questioning from a suspicious interlocutor that aims to divine whether you are human.

 

[MathematicalPhysicist]

Summary:: For about a week a BOT posted numerous responses to threads on a wide variety of topics fooling participants that it was a human.

For about a week a BOT participated in discussions on Reddit without participants at least for the most part having an inkling that it was not human. The BOT is based on the OPEN AI code GTP-3. OPEN AI release a version of this code more than a year ago. It demonstrated a good ability to emulate human writers and their styles. It was so good that OPEN AI would not release the code for fear of misuse. This year OPEN AI released a somewhat less capable verson for public use or at least a lite version of the original. It has been termed as autocomplete on steroids. It was only suspected of being a BOT mostly, it seemed, because its responses to threads occurred so quickly for the length of the posts.

What makes this more interesting is that it is not really AI in the sense that there is no logic involved in its performance that help it formulate the text and its content. GPT-3 generates grammatically correct text based on what it "sees" on the web at least as I understand it.
A discussion by the software engineer who called this to the attention of the Reddit community can be accessed here.

https://www.kmeme.com/2020/10/gpt-3-bot-went-undetected-askreddit-for.html

He discusses a number of posts by this BOT that are interesting. He also noted that non of the posts by the BOT are plagiarized or cut and pasted from the web as far as could be determined. Obviously, it does not understand or believe what is written. This is original content and begs the question about what do we mean by creativity, that special quality of human intelligence that sets us apart from machines. To be sure, people can also compose that which they do not necessarily understand or believe. How much difference does understanding or believing make?

Haven't you seen Blade Runner?
Humans are machines as well...

But what is the code?!

 

[gleem]

I ran across this " interview " with GPT-3. The text response of GTP-3 to the questions was fed into an AI avatar generator for this presentation. which is not generally available to GTP-3 users.

 

[Oldman too]

On “Derp Learning”.
https://www.gwern.net/GPT-3
Here is a very good example of the strengths and weaknesses of GPT-3, a cool site that's fun to wander.
It's worth noting how the phrasing of the prompt affects the quality of the generated response.
Also, to give equal time to opposing opinions, a GPT-3 generated editorial on "Are humans intelligent"
https://arr.am/2020/07/31/human-intelligence-an-ai-op-ed/