Day: April 20, 2023

The Consciousness Compass

Musings
section-ca3cc10

The Consciousness Compass

Edited by GPT 4. Illustrated by MidJourney. What does this mean? See Afterword below.

 

"What the hell is ‘a common denominator of consciousness’”? you ask me.

We’re in a café in the city on a rainy day. You’ve just opened my new paper on your laptop, recently published in the peer-reviewed Nature Portfolio journal Biology Communications. I crack my neck, pause for a beat, and then ask you a question.

“Would a ‘consciousness compass’ make more sense?”

You shake your head no. What the hell is he talking about, you’re thinking. Okay. I’m ready for that.

“Let’s put these on,” I say, and I pull two virtual reality head-mounted displays along with haptic suits and place them on the table.

“Seriously?” you ask.

“It’s easier to show you,” I say.

We both strap the haptic suits and VR headsets on, and in a flash, we’re suddenly whisked away to … somewhere. The landscape is naked rocks in every direction and the sky is a cloudless blue. It looks sweltering hot, you think, and you’re glad we’re only in VR. Occasionally, a shrub or a small cactus dares to peak out from behind a boulder. Why are we here, peaking out too?

“Joshua Tree?” your avatar asks me. Bingo. It’s a National Park we’ve both visited a dozen times. But off trail, everything starts to look the same. Rocks and boulders in every direction. I pull a compass from my pocket.

“If we want to get back to the campsite, we just need a map and a compass.”

“I know how a compass works,” you say, rolling your eyes.

I pull the simple gadget from my pocket. It renders nicely in VR, wobbling a little bit as I steer it around in my hand. I point in the direction of the compass needle.

“So this way is north, right?”

You cock your eyebrow. “Well, yeah. I don’t need to be a boy scout to know that.” What am I getting at?

“It’s almost north,” I say, “but not quite. Magnetic north isn’t true north. There’s a bias.” I snap my fingers and away we go again.

We’re on the edge of a fjord, glaciers blanketing the land around us, icebergs drifting by. You didn’t bring a jacket, and you’re still glad that you can't feel the temperature in VR.

“We’re in Nunavut, Canada. Can you get us to the north pole?” I ask, handing you the compass.

“I get it,” you say, “I know. Earth’s magnetic pole doesn’t align with the geographic pole. The compass is less useful now because we’re already so close to the location where the two poles diverge.”

https://i0.wp.com/advancedconsciousness.org/wp-content/uploads/2023/04/wherearewe.jpg?resize=891%2C557&ssl=1

“You got it,” I said. “And what’s more,” I add, “the north magnetic pole is constantly drifting by many kilometers each year. It’s worse than just being a little off from the north geographic pole: it’s a moving target. If you use it as an approximation of the north geographic pole, the error isn’t constant: it’s always changing.”

You do a sarcastic slow clap. Even in VR, your avatar is expressive. What does this have to do with the Communications Biology paper?

“Let’s try one more place,” I say, and with a quick nod, we’re back in the desert again. But this time, the desert is different. The soil is much redder and entirely devoid of vegetation. The sky is clear again, but now a surreal pink rather than a serene blue. You look toward the horizon, and it almost looks like the moon is rising … but it’s much too small, and rising much too fast, with a strange potato shape, not a glorious lunar disc.

“We’re on Mars?” you ask.

“Yes, time for Phobos to rise,” I say, referring to the larger of Mars’ two moons. “I would almost say it’s my favorite time of day … but actually, it happens twice a day.”

“What are we doing here?”

I hand you the compass. Your spacesuited avatar takes it and gives it a hard look.

“The needle is taking a while to settle. But if the moon is rising—I mean, if Phobos is rising—then that direction must be east,” you say pointing toward the Phobos-rise.

“Not so fast. Phobos orbits faster than Mars rotates—it rises in the west and sets in the east.”

“Fine, west then. But anyway, why won’t the compass work? It’s still just spinning.”

“There’s the rub,” I say. “There’s no global magnetic field on Mars. Nothing to directly tell us which way is north. Plenty of iron rocks to confuse the needle though.”

“Okay, you got me. What’s your point?”

I take the compass from you and toss it in the Martian dirt. “What we need,” I say, “is a universal compass. A compass that points north under all circumstances, no matter where we are. A compass to guide us just as reliably in Canada as in California, just as reliably on Mars as on Earth.”

You shrug. “I think a normal compass works just fine most of the time. How often are we really going to be exploring the Artic or crawling around on Mars?”

“But that is exactly when we need a compass the most,” I retort. “When we are off exploring, in the unknown. When the world is familiar, we can get our bearings easily. But on the frontiers, there are few things worth more than a reliable compass.”

“Okay, I get it. When we’re in a new place, we can’t rely on familiar landmarks.”

“Yes,” I say, and I know there’s no smile rendering through my opaque spacesuit visor, but you can feel my grin. “And so it is too, with consciousness. When brain dynamics are familiar—the same old familiar activity that neuroscientists and neurologists see each time electrodes are placed on the scalp to record electrical brain activity or EEG—that’s when you hardly need a compass. After all, you know that large, highly synchronized, slowly oscillating brain waves indicate deep sleep or unconsciousness. And you know, equally well, that low amplitude, desynchronized, fast brain waves indicate consciousness, either as wakefulness or dreaming.”

You nod and take off your space helmet. “I’m glad we’re in VR,” you say, looking around with your bare eyes. “Why are we wearing these things anyway? We don’t need oxygen.”

“Are you paying attention?”

“Yeah, just a little distracted. I mean, we’re on freakin’ Mars.”

I snap my fingers, and again, we’re in a beautiful place again, but this time teaming with life and people. We gaze out the window from the fifth floor of an office building in Santa Monica. Palm trees dot the busy street outside. Off in the distance, to the south (yes, the compass works again!) a plane lands at LAX, and further west, you see the U.S. Bank Tower, Wilshire Grand Center, and other jewels in the skyline of downtown LA. Santa Monica Beach is somewhere just around the next city block. Mars, on the other hand, is now over 100 million kilometers away.

“Where am I?” you ask. Behind us, a young man in a plaid shirt steps out of his office. “Who are you?”

“I’m Nicco Reggente, welcome to IACS!” he answers with a smile as he grabs a kombucha from the fridge. You slowly scan the office space and catch a glimpse of me—a year or so younger—coding away on a laptop.

“Institute for Advanced Consciousness Studies,” I clarify. “This is where I wrapped up the project, sitting over there, running surrogate data simulations.”

You look lost, and I realize I’ve gotten ahead of myself.

https://i0.wp.com/advancedconsciousness.org/wp-content/uploads/2023/04/santamonica.jpg?resize=891%2C557&ssl=1

"So, back to consciousness," I continue, "just like with the compass, when we venture into unfamiliar territories, we need a more reliable way to measure consciousness. The familiar landmarks and patterns of brain activity can only take us so far. In cases of severe brain injury, anesthetic drugs, or genetic disorders, we need a 'common denominator of consciousness' to guide us."

You nod, taking in the view of the bustling cityscape. "Alright, I get it now. You're saying that we need a 'consciousness compass' to help us navigate these uncharted territories where EEG brain waves lose their usual meaning. Once the brain is lesioned or challenged by drugs, EEG waves begin to behave strangely and all bets are off.”

"Exactly," I say. "The 'common denominator of consciousness' is a compass that can guide us through the complex landscape of profoundly altered brain activity. In a nutshell, what we want to know is whether a person is still experiencing themselves or the world around them, regardless of whether they have the capacity to respond to tasks or questions.”

You take a seat in a chair, contemplating my idea. "Okay, I think I got the gist of it,” you say with a nod.

I lift my finger and say “Let’s consider some of the traumatic brain injury patients at the UCLA Medical Center”

 “Can we not teleport this time? I just sat down.”

“It’s just a bit down Wilshire actually, but okay, as you wish,” I say with a grin, also taking a seat across from you. “The patients in the intensive care unit at the hospital are often unresponsive, so we can’t simply ask them ‘are you conscious?’. And because of their severe brain injuries, it’s not clear what their EEG oscillations mean. A doctor might see slow waves on a patient’s EEG and assume this means the patient is unconscious But these slow waves could also be the ‘loud scream’ of a focal brain lesion, appearing everywhere in the EEG recording as electrical signals conduct from this focal lesion to distant regions of the skull and scalp!”

"Right,” you say “I get it, looking at slow waves in these patients might be like relying on a magnetic compass to find north on Mars. But you don’t know for sure which patients are conscious, so how can you develop a better compass from their data?”

“Ah, that’s a big problem with some earlier studies,” I say. “Even if we throw machine learning at lots of data, we often lack the ‘ground truth’ which tells us which patients are conscious and which are not. So in my new paper, we don’t look at those patients at all.”

“You don’t?”

“No—instead, we look at children with rare genetic disorders during sleep and wakefulness. We know that these children are conscious when they are awake, just like you and me. And we know that they are probably experiencing nothing during the ‘non-REM’ stage of sleep, where vivid dreams are unlikely. That’s our ground truth. But, a bit like other neurological patients, these children have unusual EEG patterns. In one disorder, called Angelman syndrome, the children show slow EEG activity resembling sleep during wakefulness. And in another disorder, called Dup15q syndrome, the children show fast EEG activity, almost resembling wakefulness, during non-REM sleep.”

I see the lightbulb flash in your eyes, and finally, this long trek through VR feels worth it. “So whatever EEG patterns reliably indicate consciousness in these children, despite their abnormal EEGs, those patterns are the common denominator of consciousness we might want to use as our compass!”

“You got it! And lastly, we also look at healthy, typically developing children to make sure that those patterns generalize to them, also indicating consciousness in normal EEGs.”

"All right, so what did you find?" you ask, genuinely intrigued now.

"We found that a particular family of measures, called entropy, were the most reliable indicators of consciousness in both the children with genetic disorders and typically developing children," I explain.

“Entropy … isn’t that something from physics?”

“Right, physicists like to talk about entropy as the number of possible ways to arrange a system or signal, kind of like ‘disorder’. In EEG, entropy measures show us how complex the signals are. It’s low when the signal is highly regular and predictable, and it’s high when the signal is irregular and unpredictable. Overall, entropy appears to be much more reliable in identifying conscious brain activity than traditional EEG measures based on amplitude and frequency.”

You lean back, processing the information. "So, in a way, you've discovered a 'compass' that points to consciousness, regardless of the brain's individual quirks or injuries. It's like a universal compass that works on Earth and Mars, even when the magnetic poles are shifted or absent."

"Exactly," I say with a smile. "This 'common denominator of consciousness' could help doctors and researchers identify consciousness in unresponsive patients, even when it’s masked by unusual EEG patterns. It's a step toward better understanding and treating severely brain-injured patients, like those in the intensive care unit at UCLA, and making sure that we don’t misdiagnose someone with a ‘vegetative state’ just because they can’t respond to our questions. Ultimately, it might also reveal something about the mechanisms of subjective experience, as many neuroscientists think that neural complexity is foundational to consciousness itself.”

The sun begins to set outside the window, casting a warm glow across the room. You take a deep breath and look back at me. "Well, it looks like I've got some reading to do. This paper of yours just got a lot more interesting."

With that, we remove our VR headsets and haptic suits, returning to the familiar surroundings of the café. But as my eyes focus again, I see the café manager walking toward us, looking none too happy.

“Excuse me," he says sternly, "this is a café, not a VR gaming center. You’ve knocked over enough glasses and chairs. Please pack up your equipment and take your drinks."

Embarrassed, we quickly apologize and start to gather our things. As we sip our now lukewarm coffees, you glance over at me, a mirthful smile playing at the corner of your lips.

And so, we continue our search for the common denominators of consciousness and life, one spilled coffee at a time.

https://i0.wp.com/advancedconsciousness.org/wp-content/uploads/2023/04/spilledcoffee.jpg?resize=891%2C557&ssl=1
section-1ea6e20

Afterword: Using generative AI as a copilot

Generative artificial intelligence (AI) assisted with the production of this blog post, from GPT 4 suggesting what should come next in the story, to Midjourney version 5 generating custom illustrations to accompany the text. The past several months have seen enormous advances in AI and, as of this writing, the future role of generative AI as a “copilot” in activities ranging from writing to coding remains uncertain. Five years from now, transparency statements such as this one may be completely unnecessary, just as you feel no obligation to disclose having used the spell checker in your word processor. Or, perhaps society will shun generative AI tools for spoiling creative work and taking fulfilling jobs away from humans. Either possibility feels plausible at this point. We are truly in terra incognita.

So, how exactly did generative AI assist with the production of this blog post? Midjourney’s role is obvious—I have no artistic talent in visual art, and so AI art programs like Midjourney are really the only option for me to illustrate a casual blogpost with no budget. But the player you’re probably really wondering about is GPT 4—what role did this quasi-AGI play in my blog post?

For the first half of my writing, the answer was, none whatsoever. I began writing this idiosyncratic present tense short story to explain my recent Communications Biology publication one morning in February this year when inspiration struck. Then, as tends to happen, I got busy with other things and didn’t revisit my story until two months later.

When I returned to the story in April, the characters had just teleported in VR to Santa Monica. I knew what I wanted to have happen, but writer’s block wouldn’t let me put it down in words. So, I pasted what I had written thus far into ChatGPT and let the GPT 4 model continue my story. Here’s what it suggested:

"So, back to consciousness," I continue, "just like with the compass, when we venture into unfamiliar territories, we need a more reliable way to measure and understand consciousness. The familiar landmarks and patterns of brain activity can only take us so far. With new technologies, artificial intelligence, and novel states of consciousness, we need a 'common denominator of consciousness' to guide us."

You nod, taking in the view of the bustling cityscape. "Alright, I get it now. You're saying that we need a 'consciousness compass' to help us navigate these uncharted territories in neuroscience and beyond."

The GPT 4 story then continues for several more paragraphs, but the dialogue gets increasingly vague, misdirected, and shallow. Still, I like the transition, so I used the first two sentences from GPT 4’s suggestion. And buried a few paragraphs down, there’s a gem that I decide to also use later in my story.

The sun begins to set outside the window, casting a warm glow across the room. You take a deep breath and look back at me. "Well, it looks like I've got some reading to do. This paper of yours just got a lot more interesting."

But after my character finishes explaining the methods of the paper to “you”, I called upon GPT 4’s help again. This time, the dialogue was much more on the mark from a technical perspective.

"Alright, so what did you find?" you ask, genuinely intrigued now.

"We found that a particular measure, called the Lempel-Ziv complexity, was the most reliable indicator of consciousness in both the children with genetic disorders and typically developing children," I explain. "This measure quantifies the complexity of the EEG signal, and it appears to be much more reliable in identifying conscious brain activity than traditional EEG measures like amplitude and frequency."

Wow, I really did use Lempel-Ziv complexity (among other entropy measures) in my paper! And since I hadn’t mentioned Lempel-Ziv yet in the story, this appears to be GPT 4 drawing on its knowledge of the scientific literature to do just what it was designed to do: predict what words should come next in a text (yes, everything else you might use ChatGPT for, like debugging code or tutoring you in a foreign language, is just a side effect). I felt that Lempel-Ziv was a bit too technical for this post—there wasn’t any need to introduce jargon or specific algorithms for estimating entropy—but I nonetheless tweaked some of GPT 4’s dialogue, revising it to fit my vision of the story, and incorporated it into the blog post.

I then called on GPT 4 for help one last time at the conclusion of the story, when the characters reenter the real world. Again, I knew what I wanted to have happen: the characters have wreaked havoc on the café and are duly expelled. After a bit of back and forth with GPT 4, I got the prose that I was looking for, including the last line of the story. I asked GPT 4 to “Make the final line thought provoking and provocative.” The baby AGI gave me five different options, but I took the first one, with just some slight tweaking.

And so, we continue our search for the common denominators of life, one spilled coffee at a time.

So, what is the future role of GPT 4 and large language models in creative writing? Will these generative AIs serve as our virtual editors, suggesting new directions for a story and penning a few of our best lines? Or will we eventually see these tools as the artistic equivalent of performance enhancing drugs—the writer’s version of doping?

Alas, this is one question that GPT 4 cannot answer.

section-a8ead51
Read more