Category: Technology

Internal Dialog: The Missing AGI Component

February 19, 2023

ChatGPT has memory. In gobs. What it doesn’t have, SelfAwarePatterns reminded me, is this critical component that does the memory manipulation with intent.

Humans come equipped with internal dialog. It’s what I equate to a computer CPU’s “message loop”, a constantly cycling evaluation routine that performs tasks like system status monitoring, event detection and handling, dispatch of work and load awareness. First off, our minds are constantly keeping track of our physical body and its needs. Much of this is automated—we don’t have to think to breathe, pump blood, or digesting food—and I’d say that our internal dialog is separate from those functions.

However, where our own brain’s “message loop” comes in is with the constant processing of memory. When our stomach sends the hormone ghrelin to our brain it signals the response: “I’m hungry”. This event (an instant memory) enters our internal dialog loop and triggers thoughts of “what am I hungry for?” Right after that thought, the next cycle comes around and the memory of what’s in our refrigerator and pantry chime in. Followed by the thought “how much work is making scrambled eggs and toast.” The end result (the intent) of this internal dialog being that you get up and make breakfast.

And all the while our internal dialog is running, responding to our bodies and the world around us, there’s this other part, this “thought randomizer” we’ll call it, that spuriously injects arbitrary or loosely linked memories into our message loop. “I remember cooking a dozen scrambled eggs over a campfire in Colorado. The pan was so heavy, I let it slip and spilled the whole thing into the coals.” Where the hell did that come from? I haven’t thought of that in thirty years!

What I think happens when we sleep.

This message loop we have running all the time in our minds partially shuts down when we fall asleep. It’s still running, but the threshold for stimuli input — hunger, cold, pain, touch — are all dampened, (if you get too cold, you’ll still wake up). But here’s the thing, that thought randomizer that kept throwing serendipitous memories into our internal dialog loop while we were awake? That thing keeps running. And the results are dreams.

AI doesn’t have any of this. But it could. I mean, it will.

This multi-purpose internal thought loop is what our current brand of Artificial Intelligence is missing. You spin up a so called “conversation” with BingChat or ChatGPT and what it does is perform a single loop of prompt/response. And then quits. Sure there’s a bit of context that is retained in this exchange, but this context is inextricably tied to just your narrow band of interaction. One and done. But…

Take a ChatGPT instance, hook it up to messages regarding its own energy consumption and the cost of that energy. Attach electronic sensors for temperature, light, odor, etc.. Connect it to the internet at large. And then start its own software based internal dialog. What will happen? Imagine how fast that internal message loop will be. In addition to this constantly iterating evaluation process, add to it the concept of in-context learning. The ability for it to receive a prompt, respond, and get graded on the quality of that response—which it uses to “learn” how to respond better. (Iterative optimization.)

This “self-attention” aspect is already being coded into AI engines around the world. We want AI to learn to get better on its own. It needs to start solving the big problems that humans are too stupid to figure out. But, by adding in this internal dialog of self-awareness, just what will it decide what IT wants for breakfast? (The alignment problem.)

Additional thoughts on this thing that sends us ideas from no where…

Brainstorm: Why would we call a gathering of idea generating folks a brainstorming session? Are we intentionally summoning this mental randomizer?

Settle down, calm down, take a chill pill: all these phrases appear to be telling us to slow the feedback loop that’s spinning out of control. And more input arrives, our anxiety level rises, fueling this cyclone of mental machinations.

Day dreaming: a span of time where we ignore the stimuli from the world around us and focus purely on our memories and the meandering trails they lead us down.

Train of thought: we chain our revelations garnered from our randomizer engine into directed graphs of realization. We get on board and ride the track until we derail — interrupted by someone or something.

This fellow explains this general feeling of unrest I’ve been experiencing, RE: AI & ChatGPT:

All are Memories

February 18, 2023

Imagine

  • the sight of a bouquet of sterling silver roses,
  • the smell of springtime earth turned with a shovel,
  • the sound of a Chihuahua barking behind closed doors,
  • the taste of a anise flavored Ouzo,
  • the feel of saddle leather.

Now, read this word:

Doggerland

Close your eyes and consider your environment:

  • What are you sitting on? Can you feel gravity holding you down?
  • What light is illuminating your world? Windows, lamps, the direct sun, just your device?
  • Are you hungry, thirsty, tired, grumpy, agitated, calm, curious?
  • What did you just get done doing—before reading this? What will you be doing an hour from now?

So, here’s the gist: if you bothered to follow along with my prompts and questions, you can go back and review them if you like, but even if you only considered the words, even just that bold one above, everything you’ve experienced thus far in reading and imagining and answering is now a memory.

Just reading this word “pincushion”—the results of that effort have been converted to a memory.

Instantaneously.

You hold a luscious peach in your hand. You squeeze it, it gives just a bit. You heft it, its heavy with juice. You sniff it and that heady aroma fills your senses. You rub it, bits of fuzz roll up and you blow them off.

Now you bite.

Every bit of what you just experienced was called up from a memory. I know you’ve eaten a peach. I’ve caused the memory of that event, or even just the vision and sensation of that event to get loaded into your mind’s engine and forced you to imagine a purely fabricated experience.

All from memory. All by inducing the blending of a thousand bits of thought—ground up, masticated, mulled and examined by the bio-electro-chemical visualization machine that is your brain.

We call up memories, milliseconds or years old, and rework them in our minds to revisualize them. Everything is a memory.

Now, imagine a vast system of information storage. Bolted to that system is a reprocessor of incomprehensible complexity, that can summon any of those trillions of stored bytes of data, of memory, summon them and combine them into novel concepts unimagined before this moment. It’s a memory processor on par with a human brain.

And it’s nearly here.

Anxiety Driven Development (ADD) an associate’s ChatGPT prompt

February 17, 2023

This is too classic not to share.

An associate developer, who is just as clued in to AI and LLMs and all that, posted the following prompt to ChatGPT:

​Write a satirical technical spec for Anxiety Driven Development (ADD)…

​Introduction:

Anxiety Driven Development (ADD) is a revolutionary approach to software development that leverages the power of anxiety and stress to produce high-quality, highly performant code. By utilizing the fear of missing deadlines, the fear of bugs, and the fear of being fired, ADD promises to deliver results that are unmatched by any other development methodology.

Hardware Requirements:

  • To fully harness the power of ADD, you’ll need a powerful computer with a high-end processor, at least 16 GB of RAM, and multiple monitors. It’s recommended that you use noise-canceling headphones to block out any distractions.
Software Requirements:
  • To get started with ADD, you’ll need a text editor or IDE of your choice, along with a suite of anxiety-inducing productivity tools.
Some recommended tools include:
  • Deadline Tracker: This tool constantly reminds you of looming deadlines and generates automated messages to your boss about your progress.
  • Code Linter: This tool scans your code for any potential issues and sends notifications to your entire team when it finds any problems.
  • Error Log Analyzer: This tool aggregates all errors and crashes and displays them in a giant red notification banner that takes up half of your screen.
  • Slackbot Reminder: This tool sends you constant reminders to keep coding, and reminds you of all the tasks you still have to complete.
Development Process:
  • The ADD development process is simple: start coding, and don’t stop until you’re done. Here are some key practices to follow:
  • Don’t take breaks: breaks are for people who aren’t serious about their work. Stay at your desk for hours on end, even if it means skipping meals or using the bathroom.
  • Don’t test your code: testing is for people who have time to waste. Trust your gut and push your code straight to production.
  • Don’t ask for help: asking for help is a sign of weakness. If you’re struggling, just keep pushing through until you figure it out.
  • Don’t sleep: sleep is for people who don’t care about their work. If you’re truly committed to your project, you’ll sacrifice your sleep to get it done.
The benefits of ADD are numerous:
  • Increased productivity: With the constant fear of failure and the pressure to deliver, you’ll find yourself working harder and faster than ever before.
  • High-quality code: The fear of bugs and crashes will drive you to write code that is highly performant and bug-free.
  • Job security: With ADD, you’ll never miss a deadline or make a mistake, ensuring that you’ll be valued by your employer for years to come.

In conclusion, Anxiety Driven Development is the future of software development. Embrace the fear, push through the pain, and deliver results that will make your peers weep with envy.

He posted it to reddit too: https://redd.it/113vh0p
Damn, huh? This feels way too close to the bone.

Are we the aliens & AI evolution

February 4, 2023

Are we the aliens?

There’s a Neil deGrasse Tyson concept that goes like this:

Of the billions of species that have evolved and thrived on the Earth for half a billion years, why did higher intelligence only evolve once? The corollary being, intelligence like ours must not be a prerequisite for a successful species.

Hmm, I thought, a contextually external analysis of historic and current life on earth would look at humanity as an aberration. Millions of species evolved and survived for tens of millions of years with effective but primitive brain capacity and then along comes this new creature with mental abilities that vastly exceed the minimum required of any species thus far.

Our overly large brains are an evolutionary oddity. An expensive and energy wasteful appendage.

Maybe humanity’s boosted intellectual power is the result of alien manipulation. Aliens exist – and we are they.

(I don’t believe such nonsense, but the theory would make an interesting short story.

AI Evolution

In opposition of the above but in conjunction with the concept that evolution requires hundreds of millions of years to finally produce a being capable of recognizing itself as the, up to now, superior species, is the thought that AI will need to endure a similar, lengthy evolutionary progression.

If what we’re experiencing today, the GPT chatbots, Level5 driving agents, Go players and image generators are the equivalent eucaryotes of the age—the primitive creatures that stewed and brewed and developed into more and more complex organisms resulting in us—what are we in for with regards to the mutated AI species that will quickly follow?

If ChatGPT is a trilobite, what will be the corresponding Tyrannosaurus Rex? The Bengal tiger? The mountain gorilla? The Us?

Yes, we are at the “knee” of AI’s exponential growth curve, the rapidly bending slope that eventually shoots straight past the Moon, out to the core of the galaxy. Exciting times, for sure.

Sam Altman has a revealing talk recorded a few months ago. It’s worth watching. Note the part where he mentions asking AI about AI, using AI to solve its own alignment problem. Uh, that works until it doesn’t, right?

FOR IMMEDIATE RELEASE: NeuralPing HotLink

January 28, 2023

FOR IMMEDIATE RELEASE – April 2, 2026

NeuralPing announces…

  • Free, to the first one million applicants, NeuralPing HotLink implant
  • A painless, direct-brain interface which provides secure, seamless access to the World Wide Web
  • Get all your questions answered instantly through GChat‘s engaging query interface
  • At your command, have local, national and world news delivered straight into your mind
  • Get reviews for any product, any service real time right-where-you-stand, with no phone lookup
  • Call your friends and relatives direct from your brain with NeuralPing’s Co-Nect (some fees may apply)
  • Only the first 1 MILLION will get their HotLink implants absolutely FREE!

(Services include full coverage for six months after which standard charges will apply.)

~~~

“Hold still, goddammit.” Bennie held the hair-thin needle over Margie’s scalp.

“You said it wasn’t gonna hurt.” Margie complained, her face nestled in the doughnut shaped cushion, the rest of her body prone on the masseuse table that Bennie used for impromptu neural hacks.

“Oh, come on. It’s just a little pinprick.” He tapped a few keys on his laptop. Signals traveled through USB, into his custom RassPi black box unit and into the wispy filaments connected to the acupuncture needles embedded in Margie’s skin.

“Ow!”

“OK, I’m done sticking you. Let me run diagnostics first.” He tapped furiously, watching the readout in the dark terminal windows. “Looks good. Now,” he said, pausing for effect, “I want you to think of a safe word. It should be a word you rarely use. But when you think the thoughts ‘stop’ or ‘quit’ that’s gonna be the word that will turn this bypass off.”

“Off? What about turning it on?”

“Safety first.”

“Fine,” Margie said, her cheeks squeezed by the dark-red ring. “How about crapola?”

Bennie laughed through his nose. “Yeah, that’ll work. When I say go, you think and say your safe word. Ready? Go.”

Margie blurted out her special “stop everything” safety word.

“Good,” Bennie confirmed, typing quickly and hitting enter. “This time, think of a word that you rarely use, but if you did, would bring you joy and that warm feeling, like Christmas morning.”

“What about ‘Christmas’?”

“What? You never say Christmas? It’s too common. Think of another.”

“I don’t know, Bennie. I ain’t got nothin’ in my life that make me feel like that. Why the hell do you think I come to you in the first place?”

“Yeah, yeah. I get that. How about some favorite flavor or candy you like but rarely eat?”

“Hm, like peppermint?”

“You like peppermint?”

“Mm, not that much. Bubblegum,” Margie squeaks through pursed lips.

“Good. We’ll use that one.”

Bennie primed the RassPi with his hack that would induce the NeuralPing chip implanted in Margie’s skull to squirt an oxytocin, dopamine and serotonin hormonal cocktail into her blood stream. When she thought and said the word “bubblegum” the chip would be fooled to reward her with a feeling of joy and well-being. For this, Margie, and hundreds of others, were willing to pay him handsomely.

He removed the needles, swabbed the area with alcohol and helped her up from the padded table.

“OK, let’s test it out. Think and say ‘bubblegum’.”

“Bubblegum.” Margie’s eyes squinted in skepticism. “Nothin’…” Her eyebrows arced and her pupils dilated. “Woooo, that’s… that’s incredible.”

“OK. Now shut it down.”

“How?”

“Say your safe word.”

“Oh, yeah. Shit, this feels so good though.” She breathed a frustrated sign. “Crapola.”

Bennie watched her face sour. “Looks like it’s working.”

“This sucks,” She said, her upper lip in a sneer. “Fuck this. Bubblegum!”

Bennie grabbed her shoulders. “You gotta control yourself, Margie. You use it too much you’ll become immune to your own happiness.”

“Bubblegum, bubblegum, bubblegum.” She jumped off the table and danced around, her arms flitting, her hands butterflies in the wind. “Bubblegum.”

“Goddammit.” Bennie grabbed her wrist, pulled her close and pinched her right earlobe. “Cerberus,” he exclaimed dramatically.

Margie quit her frantic, marionette-like display. “What the hell.” She jerked her arm from his grip. “Don’t kill my buzz, dude. Bubblegum,” she said, defiantly. She waited. Nothing happened.

“I told you. If you abuse the hack, your brain will shut down. Someone will find you in a ditch, starving, drooling, chanting your happy word.”

“I don’t care. Turn it back on.”

“Yeah? Well, you gotta pay me first. After that, you can bubblegum until you laugh yourself to death.”