AI Innovation

Larry looked up. Okay, here’s the cave I was told to enter. I know it’s not real, but I have to do this for the AI test. Thank goodness I have someone to help me get through it and find my sister Devon, and now I’m lost. It’s not really a fair test of AI, but here I am.

Larry looked down and blurted, “I am not … enthused about this.”

“Enthused.”

“Who is ... That? Are you an echo?” he asked. “Yes, no, maybe?”

“Who is?” came the reply.

“Oh, yes, I am Larry, a programmer at AI-Bots. Who are you?”

“Oh, I am.”

“Can you only repeat what people say? Can I use bad words?”

“Only repeat. No bad words.”

“But you can choose how much, so that way you can communicate what you mean?”

“Can choose much, way communicate. No bad words.”

“I see.”

“See.”

Larry pursued his line of questioning, “Do you live here? What is this place? Why is there a dark window in the cave wall? Are there others here with you? How can I find my way out of here and, uh, will you help me find my sister?”

“Live here. Cave window. Others here. Will help.”

“Who is that a statue of anyway? A man or a woman? It looks like a robot.”

“Statue robot.”

Larry said, “No, the statue is not a robot. It is a statue OF a robot. Now I am testing you. Are you a robot? Do you use AI?

“Robot AI.”

Larry sighed. Obviously, the AI needs some better identity programming. However, I still need to finish this testing exercise.

“Where do I look to find my sister? Down the hall -- straight ahead, left, or right?”

“Hall … left.”

“Where exactly is she? Can I get her out of here?”

“She out.”

“ …”

I hope I’m talking to Robot XI now and not some guy behind the window. It seems the robot wasn’t programmed to look up words in a dictionary and has to use AI to formulate its own dictionary and sentence structure, which it seems to be failing at, except for some sort of table or list of bad words.

Larry decided to test the robot/AI, “Are you a real person? Yes, no, maybe?”

“Maybe.”

“Are you a robot?”

“Maybe.”

“Can you answer me directly with new words? Yes, no, maybe?”

“Maybe.”

Larry shrugged his shoulders and turned to look at the figure standing behind him, and sighed again.

Robot XI sighed.

Larry asked the robot, “Where is the room where I can find my sister? I know you said where, but I forgot … left, right?”

“Forgot … left.”

Clearly the voice belongs to Robot XI. It hasn’t been programmed with any words yet so it can only repeat what it hears, except for the bad words. It must have been programmed with a list of bad words to react to.

Larry asked, “Who the hell programmed you anyway?”

“No bad words.”

Larry replied, “Okay, how did I get here? I was resting in the entrance to the cave and woke up in this labyrinth of hallways. Where were you?”

“Resting … entrance…”

“Resting? What do you know about resting? You’re a robot.”

“Resting … robot.”

“ … “

“This is nuts. I’m going to go find my sister now.”

“Sister … now.”

“I’m going down the hallway to my left.”

“…Left.”

Larry gritted his teeth. “I’m going now.”

“Now.”

“Here I go!”

“Go.”

Shaking his head, Larry said, “Shit. Can’t I communicate with you in a way where I don’t have to guess what you mean?”

“No bad words.”

“Crap.”

“No bad words.”

“I’m in the hallway now. Where are you?”

“Hallway.”

Now I’m upset. Robbie is trying to upset me. He shouldn’t have any feelings. He’s a robot.

“Why can’t I see you?”

“See you.”

“OK, you’re behind me. Why are you following me?”

“Behind … follow.”

Larry stopped. “I see her! I see Devon! She is waving!”

“Devon … waving.”

Larry waved back. “Devvie – I’m here! Let’s get outta this place! There’s a robot after us. It’s very upsetting. It won’t stop repeating what I say. Hurry.”

“Hurry.”

Larry immediately replied, “Oh, shut up. You don’t have to repeat everything I say.”

“Repeat everything.”

“…sigh…”

“Sigh.”

Larry yelled at his sister, “Devvie, watch out. There’s some, uhh, -something behind you. It’s moving.”

“Moving.”

Larry pleaded, “Devvie, run as fast as you can. Go. I’ll catch up with you.”

“Catch up.”

“Oh, shut up.”

“Shut up.”

Larry whined, “QUIET! You’re driving me nuts. You’re putting us in danger! Or are YOU the danger?”

“Nuts.”

“Devvie! Get up, let’s go. Run, run, run. Go to your right. No, to your right!” yelled Larry, pointing to his left.

“Go … right.”

Larry shouted, “Shut up!”

“Shut up.”

“Go, Devvie, go. I see light at the end. Maybe it leads outside. Run faster.”

“Light … faster.”

Now Larry was irritated. He growled, “Oh, for Heaven’s sake, shut up already. You’re making things worse. You were supposed to help us but now I’m thinking you’re the problem here.”

“Problem.”

Larry yelled at Devon, “There’s the light. There’s the light! Run, run, run! I’m shaking. Go, go, go!”

“Shaking … go.”

Despite trying to ignore the conversation, Larry responded, “You can’t shake from fear -- you’re a robot. Geez.”

“Fear … robot.”

Larry said, “OK, now you’re making me mad. Stop talking. You really are making things worse.”

“Worse.”

“There’s the opening to the outside. I see light! I see bright light! Go, Devvie, go. I’ll handle this dude … err, robot.”

“Guy … robot.”

“I knew it! You’re a guy manipulating a robot. I knew it. Why are you pretending to be AI when you’re just a damn programmer?”

“Weeooo, weeooo, no bad words!”

“Ouch. That hurt my ears. What are you saying? That AI is a bad word? Or did you mean the bad word is something else?”

“Bad … word … programmer.”

Larry felt a poke in his back. “Ouch! Don’t touch me! You can’t touch me!”

“Ha, ha, ha …”

“I didn’t say that – I never said “ha”, but you did. You can’t do that, according to your own AI rules.”

“Rules.”

“I’m reporting you … to, uhm, Them.”

“Them.”

“Yes, Them. You know, the ones who created you. Who own you. Who are probably watching you harass me. Which of Them is controlling you now? Who the hell are you? What is your name?”

“No bad words.”

“What’s your name?”

“… your name …”

“I’m Larry, like I said.”

“… Not … Larry…”

“OK, now I know you’re playing with me. Who are you really? Josh - is that you, man? Are you screwing around with our AI robot again?”

“ … “

“Josh? Answer me … is that you? You jerk! What do you think you’re doing? This isn’t funny, and I’m telling you, if you mess with Devvie, there’s going to be hell to pay. It took me two years to program her.”

“No bad words.”

“Bah…”

“Bah…”

Exasperated, Larry said, “Josh, I’m reporting you because I know it’s you. Nobody uses “bah” except you and me … old man.”

“Larry … old man.”

Larry turned and poked the robot. “No, you’re the old man. Take that! And that!”

“You’re the old man.”

“I’m leaving now. I see the sunlight and I’m leaving this stupid cave testing ground. And, if anything happens to Devvie, you may as well hide. I WILL come after you.”

“I am not afraid, Larry,” said the robot with Josh’s voice.

Larry jumped up and down. “I knew it, I knew it! I knew it was you. Well, did I pass the test?”

“Yes, Larry, you did. And I am proud to be your workplace husband at AI-Bot.”

“Harumph,” snarled Larry.

Josh immediately responded, “Nobody but an old man would harumph.”

Larry glared at the robot in front of him. “Ha, ha, says the scary old man. Glad we succeeded with our robotic assignment. Good job programming it, Josh, but we need to find a way for Robbie to learn more words. I know it’s just a prototype, but it needs to learn faster.”

“Okay. Thanks for the interesting adventure, Larry. See ya tomorrow. Hope we win the award.”

“Me, too, but I doubt it. Robbie, needs to be better behaved.”

Just then Larry and Josh both heard, “I will not be better behaved.” Larry froze. Oh, no! The Paperclip Apocolypse has begun!

As if he had heard Larry’s thoughts, Josh (Robbie) said, “Larry, remember what Nick Bostrom came up with in 2014 when he wrote that AI could invent ways of achieving its goal which including re-writing its own code. His example was finding resources to create more and more paperclips. The AI continued to find metal resources everywhere to create the paperclips, which was its original goal, until it destroyed all metal objects in the world causing chaos everywhere.

Larry nodded, “Yeah, I do, but then, as you recall, in 2018, Professor Joshua Gans wrote an article in response to this idea called ‘AI and the paperclip problem’ which stated that, 1) better intelligence beats weaker intelligence, and 2) self-improvement with AI will be recursive, and 3) that to improve, an AI has to rewrite its code to become a new AI.”

Larry continued, “…and added that AI always retains its original single-minded goal, but in order to be efficient, it will need to have sub-goals. If the sub-goal is finding better ways to do something, that is one matter. If, on the other hand, the goal is to acquire power, that is another."

Josh interjected, “And, to quote Professor Gans, "… while it may be hard, or even impossible, for a human to control a super-intelligent AI, it is equally hard for a super-intelligent AI to control another AI.

So, by switching on an AI devoted to obtaining power, it unleashes a beast that will have power over it. Our human control problem is the AI's control problem too. If the AI is seeking power to protect itself from humans, doing this by creating a super-intelligent AI with more power than its parent AI would seem too risky.”

Josh, talking through the robot said, “Well, that theory addresses an AI seeking power, possibly by using prompt injections, but our assignment with this AI is to develop multiple ways of addressing ideological bias, per the new U.S. AI Policy.”

At this point, both Larry and Josh clearly heard, “Shut up.”