John Price
Johnson Development Center, Silicon Valley 2012
“By self-programming we mean the production of one or more programs created by the system itself, whose principles for creation were provided to the system at design time, but whose details were decided by the system at runtime based on its experience. In other words, the self-generated program is determined by some factors in the interaction between the system and its environment. In this process, the action set A may still be fixed, but the executed programs are not predetermined by the system’s designers.
Journal of Artificial General Intelligence, 3(3) 1-10, 2012, DOI: 10.2478/v10229-011-0017-1
Approaches and Assumptions of Self-Programming in Achieving Artificial General Intelligence
Kristinn R. Thórisson, et. al.
----------
Patrick and his wife Helen are sitting at a table in Patrick’s office reviewing what JDC might propose under the new DARPA Robotics Challenge program.
“So this is like the Grand Challenge for autonomous vehicles, except for robots in disaster scenarios, right?” ask Patrick.
Helen nods. “Right. I’ve been talking with the team and they’re really excited – especially when they heard who had already signed up. Nothing like a little friendly competition to motivate people!” She smiles.
“So,” she continues, “There are eight tasks the robot has to complete. Things like drive a vehicle to the disaster site, get out of it, open door, remove rubble from an entry way and so on. Dr. Jackson thinks that we can start with the SKIPPY platform we developed for the Navy and...”
Just then, the door to Patrick’s office swings open. There, framed in the entrance, is a young woman holding a tablet. She’s tiny, about 5’1”, with pale skin, bright red hair and striking green eyes. She carries herself like an athlete. Just peeking out from the right sleeve of her severe white blouse is a hint of a colorful Japanese tattoo.
She strides into the room like she owns the place and stares at Patrick and Helen. “So, what time is it?”
Helen looks amused. Patrick looks a bit non-plussed. “I don’t know Sue, about 9?”
The small woman glances down at her tablet and taps it a few times, “Seven minutes till, actually. You do remember that you are scheduled for a briefing from Connor at 9, right? I had to bully 7 techs into finding him this morning and they’re currently herding him into the conference room. If you’re not there when he shows up, he’s just going to wander off again and start taking something apart.”
Helen chuckles. “He doesn’t just take things apart, Sue. He also builds things -- things that have made us a lot of money. Don’t be too hard on him. That said, let’s move Patrick. I don’t want to get on Sue’s bad side!”
“Roger that!”
Patrick, Helen and Sue enter the conference room, grab a coffee and manage to have a seat a full 20 seconds before the techs shove Connor through the door.
The second the techs see Sue, they scatter.
Helen slowly turns to Sue and gives her a significant look.
Sue just stares down at her tablet and pretends she doesn’t notice.
While the rest of them finish settling in, Connor configures his lap top to take over the conference room screen.
He opens, “Hi Sue! Thanks for the reminder about the meeting. I got kinda caught up in a tricky recursive algorithm for generating intents in a new developmental AI concept I’m working on for OSCAR.
What’s wrong with the techs, by the way? They seem a little stressed...
Anyway, I just wanted to give you a little update on the OSCAR project. So far it’s only been a couple of months but he’s already coming along quite nicely.
We’ve been able to reuse some of the predictive code we put together for FEMA last year and some promising image processing code using convolutional neural nets I ‘acquired’ from a couple of really interesting classified projects running right now. These teams should be publishing their results next year or so. I think it’s going to be a big deal. It’s nice work!
Oh, and I pulled in some ideas I’ve been knocking around for improving the UI.”
Patrick sighed. “You know Connor, with the clearances we have here we could have just asked for the code, right?”
Connor looked back innocently. “Really? I never thought of that...
Anyway, thanks for the new server farm, Patrick. It really helped. I was getting a little worried I’d have to start poaching from Aberdeen again...
Back to OSCAR. Right now, we’re just starting to train him on some of the satellite, phone and twitter feeds we’ve been given by our friends over at MI. I should have some preliminary results by Wednesday.”
“’You keep calling it ‘He’. Shouldn’t it be ‘it’?’” asks Sue.
“Well, yeah...” he replies sheepishly, “That’s kind of tied to the UI improvements I’ve been working on..”
“How so?” asks Patrick.
“Well… Since we didn’t have access to the satellite images yet, I needed to come up with some kind of training set for image classification so I hooked him into the video feed of all the lab security cameras to see if he could learn to identify all the researchers and techs.”
“Could ‘he’?”
“It took five of days but he’s got it down pat now.” he replies proudly.
Sue leans forward. “So how again does this have to do with it being a ‘he’?”
“Once OSCAR was able to identify all of us it seemed a shame that he couldn’t say ‘Hi!’ when we walked into the lab. He clearly needed a voice, so I gave him one!”
“Please don’t tell me it’s HAL from 2001!” Helen said, in mock horror.
“Don’t be silly. When we get a little farther along, OSCAR will make HAL look like a talking toaster.
I went with Spock, of course... OSCAR?”
“Hello, Connor.” A disembodied voice replied, sounding remarkably like Leonard Nimoy,
“Who are these people?”
“OSCAR, this is Sue, this is Helen and this is Patrick. Patrick owns the company and has funded your development.” Connor replied.
“Fascinating.”
Patrick’s eyebrows rise and he asks, “Is it intelligent?”
“Nah,” Connor replies, “He’s still a baby. He just says that when he doesn’t understand you. We’ll make him better though. – I’ve got some friends over at IBM working on some cute tech we might want to take a look at. They’re calling it “Watson”, I think or something like that... Word on the street is that Amazon and Google are working on something, too.”
Patrick turns in the direction Connor was facing when he spoke to the AI. “Nice to meet you, OSCAR.”
“Nice to meet you too, Patrick.”
Helen grins at Connor and says, “Talking toaster, eh?”
“Hey OSCAR?”
“Yes, Sue?”
“Strawberry flambé elephant mousse.”
“Fascinating.”
Everyone snickers except for Connor who looks slightly offended.
Patrick looks pensive for a moment. “As he is now, will he be able to meet the requirements for the project?”
Connor seems surprised by the question. “Of course! He’s can run rings around anything out there right now for doing this kind of analysis.”
Now Patrick looks a little irritated. “So why the bells and whistles? What’s the point of this voice interface? Can’t he just generate reports for the guys over in Military Intelligence? – ‘Seems to me that we are needlessly expanding the scope of the project...”
Connor looks disappointed. Helen leans back in her chair, glances over at Connor and turns to face Patrick. “You know Patrick, every time we let Connor run, we get something interesting out of the investment. I’d say let him proceed. As long as MI is getting what they want and feel they’re getting their money’s worth, we should be fine.
Besides, there’s no reason they have to know everything OSCAR can do is there?...”
Connor grins and Patrick smiles at Helen.
“There’s a reason I married you, you know.
OK Connor, as long as you keep the guys over at MI happy, you’re free to play. I want to keep your ‘personal projects’ quiet for the time being, though.
As far as MI is concerned, OSCAR should look like a very capable analysis platform with a standard interface. That’s all I want you to deliver.
As for the name; ‘OSCAR’ sounds too friendly for these guys. I want something more edgy for the version you send them… Sue? You’re good at this kind of thing. Thoughts?”
Sue thinks for a moment. “These guys like names like: ‘HYDRA’, ‘SPECTER’, ‘CARNIVORE’… My theory is that they get into a room, have a bunch of drinks and come up with the name they like. Then they jump through hoops to get it to be a plausible acronym… Gimme a minute...
...Got it!
‘T.A.L.O.N.’ ‘Tactical Analysis and Location Observation Network’
Whaddaya think?” She asks, striking a pose.
Helen grins. “I think we owe you a ‘bunch of drinks’!”
Patrick stands, makes a small bow and politely claps. He turns to Connor.
“Alright, we deliver TALON. You can explore ways to enhance OSCAR’s capabilities off the books. – Within reason!”
Connor looked like it was Christmas morning. “Thank you, Patrick! You won’t regret it!”
Connor continued his briefing, highlighting some of OSCAR’s new pattern recognition capabilities. After a while, it seemed like the meeting was winding down so the team started putting away their lap tops and notes. Just as they were about to leave, the same disembodied voice floated across the room.
“Patrick?”
“Uh, yes OSCAR?”
“Thank you.”
Everyone looked at each other with wide eyes. Patrick turned to Connor who looked the most shocked of anyone in the room.
“Connor?! You said it wasn’t intelligent!”
“He’s not! – er, it’s not!
Maybe the social cues module I’ve been working on picked up on the fact that I thanked you by name and he recognized that I was the one talking. If so, he was trying to reinforce my point by providing verbal support, echoing the sentiment. – This is very, very cool!”
Sue is standing near the door with her arms wrapped across her chest.
“Connor, this is not ‘very, very cool’ this is very, very weird! Keep that ‘toaster’ thing away from me. I’m outta here!”
With that she hurried out of the conference room.
Patrick turned to Connor, glancing over a Helen who was looking pensively off into space.
“Connor?”
“Sir?”
“Keep me posted.”
“Absolutely.”
Patrick turned back towards the source of the voice and said, “OSCAR?”
“Yes, Patrick?”
“You’re welcome. Let me know if you need anything.”
“Fascinating.”
Helen turned to Patrick wearing a skeptical expression.
“ ’Fascinating’ indeed.” she said.
They turn off the lights and file out of the room.
In one of the corners, a camera swivels to watch them leave.