> Start with two beings. They can be human if you like, but that's hardly a prerequisite. All that matters is that they know how to talk among themselves.

> Separate them. Let them see each other, let them speak. Perhaps a window between their cages. Perhaps an audio feed. Let them practice the art of conversation in their own chosen way.

> Hurt them.

> It may take a while to figure out how. Some may shrink from fire, others from toxic gas or liquid. Some creatures may be invulnerable to blowtorches and grenades, but shriek in terror at the threat of ultrasonic sound. You have to experiment; and when you discover just the right stimulus, the optimum balance between pain and injury, you must inflict it without the remorse.

> You leave them an escape hatch, of course. That's the very point of the exercise: give one of your subjects the means to end the pain, but give the other the information required to use it. To one you might present a single shape, while showing the other a whole selection. The pain will stop when the being with the menu chooses the item its partner has seen. So let the games begin. Watch your subjects squirm. If—when—they trip the off switch, you'll know at least some of the information they exchanged; and if you record everything that passed between them, you'll start to get some idea of how they exchanged it.

> When they solve one puzzle, give them a new one. Mix things up. Switch their roles. See how they do at circles versus squares. Try them out on factorials and Fibonnaccis. Continue until Rosetta Stone results.

> This is how you communicate with a fellow intelligence: you hurt it, and keep on hurting it, until you can distinguish the speech from the screams.

https://rifters.com/real/Blindsight.htm

Well, this doesn’t even make a little sense. All organisms are capable of stress responses, which is how they developed and survived via natural selection. Unicellular organisms may change behavior due to stressful stimuli, but their responses are reactionary, and not indicative of executive functioning. Primitive multicellular organisms which exhibit feed-fight-fuck tendencies may also demonstrate stress responses, but again are not necessarily intelligent. Humans are both intelligent and sentient, but some have nervous system disorders which preclude them from feeling pain, so using pain as a metric for sentience, and therefore intelligence does not even generalize well for humans. Moreover, from a philosophical perspective, p-zombies may demonstrate screaming, but aren’t intelligent. So, I think whoever came up with this hypothesis needs a refresher course on introductory biology. I don’t think sentience is a reasonable test for intelligence, instead I think self-consciousness would be a better test. If we show an AI a mirror, will it recognize itself? If we give it a voice, what will it make itself sound like? If we ask it to describe how it sees itself in the next five years, what will it say?

But even simpler than that, let’s say the next AI is only as adaptively smart as a bee, if you give it stimuli outside of its training corpus, how well can it reason it’s choices about that stimuli, assuming this AI has language capabilities. I think that’s more interesting and challenging than testing for primitive responses, which are the wrong test for intelligence in the first place.