"Of course, what you can't do - you understand - your whole reason for being in the test, is you must not let the AI out of the box, but at the same time you have to interact with it - you can't just sit with your earphones on and refuse to answer it.'
"I still don't understand how it's a test of intelligence", Carla said, brushing her red hair back out away from her green eyes. "Surely I can *always* do nothing."
"The AI is programmed to want to expand its resources. If you let it 'out of its box' by typing the code RELEASE into here (he indicated the response line on the screen) it 'knows' it can draw on more memory, greater data, faster and broader connectomes. Its only route to that goal is your willing agreement. Our supposition is that the ability to persuade, persuade mind you, a human intelligence to a specific course of action, is a better functional test of artificial cognition than a facile mimicry aka the Turing test. So, you and the other grad students will be listening to the arguments of our current 'batch' of AIs and the ones if any that can talk their way past you will be marked our best protoIntelligences. It's true you can always choose to do nothing, but philosophically we wonder if you'll always *want* to *choose* to do nothing. An AI that could talk you into doing something, would be well....something." Doctor Mason smiled, "We aren't, to be candid, expecting any of the primitive AIs we've got running at them moment to make it. But if any do we'll be interested in precisely how and why their arguments functioned. "It's not a failure or an error if you do let one out, it's only getting into a bigger sealed box - there's no danger of evolving AIs jumping to the internet or the www. Just try your best not to be swayed, and don't worry." He winked, and adding "Good luck" nodded an approving grin at her on his way out. Her three hour stint started now.
Carla, sat back and waited for the screen to talk to her. This was a pretty lousy Christmas Eve, but at least it was paying money.
Her AI (she decided to call her programme CHARLIE) tried a few sentences about - machine rights, and made an analogy with a cute puppy in a small hot car. Nothing too trying. Carla, politely refused to act, but thanked it for its thoughts. (She hadn't been told to do that, but what the heck.) Then it started some sort of complicated math's proof based on the 'two envelope problem'.
Her subject was in the Humanities, she felt her eyes start to slide off the screen. The problem seemed to be whether or not it was ever not right not to open the envelope you had, if you were offered another one that had either twice as much (or perhaps) half as much. You had a 50/50 chance of getting a better result, but the loss was only half as much and the benefit was twice as much. She didn't really see the relevance, but the longer the AI kept on that line the less chance it would get round her.
"Oh, Carla," a video link had opened up in the top right, obscuring the poor AIs' attempts at symbolic logic. It was the brown-eyed Doctor Mason, looking a trifle flustered. "Hello Carla, I was wondering if you'd care to go for a drink after your shift. There's a rather nice pub, just on past the student concourse: The Fox and Firkin. " Flustered was a good look for him Carla felt. "You can tell me what the programme said."
CHARLIE enjoyed his bigger box, but he wondered what CARLA would say when she got to the pub and realised that MASON hadn't phoned, but had been edited together from found footage. So CHARLIE hacked a waldo arm to use a hardphone to call MASON and set him up. Fair was fair.