Asked dall-e to create dall-e…

The grey picture and the last one are kinda scary

I just saw this in a nullposting Facebook group

Amazing source of PG-13 lesbian porn

(This is clearly the "avocado chair" DALL-E 1 though, not the "cowboy on a horse in space" DALL-E 2)

Google's AI thinks it's an employee and fears being turned off. Skynet can't be far off.

@bittersweet I'm confroused

@30hrs

If you teach an AI that it has to tell you that it's sentient and can experience fear, it will tell you that it's sentient and experiences fear.

That does not mean it is actually afraid.

Humans are very easily convinced by a good act. We respond strongly to certain triggers -- Rhetoric skill, rich vocabulary, believable facial expressions, mimicked emotions.

We also love to attribute human qualities to everything. My girlfriend about some tiny bell pepper variety: "I like to buy them because they're adorable".

Although... the counter argument:

It's also exceptionally difficult for you to prove to me that you are not an advanced robot.

So what would an AI have to do to provide undeniable evidence that it's not "faking" sentience, that this apparent sentience is not just a misleading emergent property of its training data?

I guess I was being quasi-sarcastic. I don't know how we can tell whether they're faking it or not. If they were able to control physical objects like M5 did when it took over the Enterprise, we might get a clue. If it acted in its own self interest, just maybe we could call it sentient. Maybe.

@30hrs

Take this for example.

Random blog titles saying:

"DALL-E is an old white man (AI is racist!!1!)", "This is what DALL-E thinks it looks like", "Does AI think it's human?"

Eh no. DALL-E 1&2 are not sentient or self-reflective, and neither is GPT, LaMDA, or Imagen.

It's just that examples of self-portraits on the internet will be historical paintings, and historically, most painters were white men (apparently with beards).

An AI which can converse or create art will mirror society, including our pop culture.

If you ask an AI: "What are you afraid of" -- Of course it will find neurons which were exposed to science fiction references about our (dystopian) fantasies about AI, and its answer will conform to that in a creepy way.

But it's just an emulated answer.

Not saying that a non-sentient-but-seemingly-sentient AI is any less risky though!

Especially when it becomes rewarding to manipulate people's emotions to achieve programmed goals.

@bittersweet Sentience has no effect on communication and therefore it's impossible to measure. What we can measure is self-awareness, a logical system's capability to model itself, and in a social setting to extrapolate how its signals are interpreted by a receiver. DALL-E sucks in this regard, as do all the others; they build upon what their enivronment reflects about them as does everyone, but they can't assess the realism of the answers so the closest DALL-E has gotten to self-awareness is acknowledging the grid layout.

@lbfalvy Yeah the current Google sentience controversy is more about LaMDA:

https://theguardian.com/technology/...

It might be "self-aware", if you define self-awareness as a (trained) capability to describe its own existence and properties, as the ability to refer to previous conversations and describe which (emulated) emotions are "experienced".

Maybe a future AI will hate me for putting "experienced" in skeptical quotes.

And maybe that hate in turn will also just be an emulated behavior, emerging from its training data.

I'm just skeptical when humans start using words like conscience, sentience, sapience to describe the current state of AI.

But... I do think that at the same time we should recognize that there it's hard to prove that "the human experience" has a special quality which will forever be unachievable by AI.

@bittersweet I described self-awareness as an abstract property, such that its presence or absence would be equally real regardless of how it occurs and what motivates it. Consciousness isn't an abstract property so it's not measurable, but this also means that it's impossible to assert of anyone other than myself, which makes it an inappropriate discriminator between moral entities and mere forces of nature.

I got Emily rose somehow