Strange Loops

A surprising parallel between AI images and addiction.

You’d be surprised how deep insanity runs. Maybe even all the way down to bare mathematics. 

---

I tried a new diner in town with my cousin, who I know is in recovery from his drug addiction. 

Hipsters apparently measure the healthiness of a sandwich by the volume of seeds per square inch of bun. Sometimes, I hate LA.

Anyway, my cousin told me he planned on quitting his job and selling his house to move to Memphis, be with a woman who brought out the worst in him, and, presumably, get super high. He was “off the wagon,” as they say.

No matter what I said, his resolve would not waiver. While his words sounded confident, he broke eye contact for subconscious half-seconds. The “vibe” this tick creates is pretty creepy if you’ve ever experienced it. 

Unfortunately, this wasn’t my first rodeo. Growing up, my brother would come home from his benders for several-month-stints. He always promised that everything would be better this time: He got a job. He joined the military. He started with a new therapist. 

But, his eyes… Strange. Unsettling. My family quietly waited for the other shoe to drop. Eventually, it came time for him to steal money and disappear. And so the cycle continued.

At the age of 50, he finally died in a ratty apartment in New Orleans. He left behind two smart, beautiful kids who wanted nothing more than to get to know their dad. It was hard to know what to say to them at the funeral, to say the least. I wish I knew why he chose drugs over them. But, honestly, I don’t.

My family had this fundamental confusion. A bone-deep bafflement. "Why did you sacrifice everything for this?" 

The addict is in a hellish, strange loop. They will do better… RIGHT after the next high. It doesn’t matter how many times they’ve told themselves that — the lie sustains them.

It reminds me of that cliche definition of insanity. You know the one: doing the same thing over and over and expecting a different result. 

I’m sitting in my booth, not eating my seedy sandwich — now for two reasons. My mind spins. Maybe if I can transfer what I learned from my brother, I can help somehow. At least then, what killed my brother is not some unstoppable Lovecraftian monster. It’s a pattern. 

It’s time for us to part ways.

“Text me?” I asked. 

“For sure,” he said. 

Yeah, you can guess. I haven’t heard from him. 

It’s been haunting me.

Later that week, I stumbled across this Charles Eisenstien's essay about AI’s performance distorting as it gets fewer inputs from the real world. Without organic data, it’s just “moving bits around,” as Eisenstein describes. I couldn’t immediately explain it, but I knew there was a connection to my cousin. 

The addict’s shifty eyes seem to have the same Kafkaesque strangeness I saw in the demonic images of a woman named Loab (a very creepy story). She famously haunted AI images a few months ago. 

Maybe the phenomenon of insanity runs deeper than we know.

In 12-step recovery programs, they also call addiction “insanity.” I thought this was just dated language from the 50s. But, by the end of this, you’ll see it’s the perfect word.

But is addiction really the same as AI’s “insanity?” You’d be surprised how similar they are.

AI goes MAD

For the uninitiated, I’m going to wildly oversimply the process of machine learning for our purposes of making this comparison.

Say you need AI to produce an image of a cat. So, you feed it hundreds of thousands of images of cats and tell it, "This is what cats look like." The machine develops an "intuition" for "catness" until it is very reliable when choosing pictures with cats. But there are no hard and fast rules (AKA algorithms) for how to tell cats from non-cats.

Then, you can ask it to produce original images of cats. You rate outputs as good or bad several million times. Eventually, it gets remarkably good at producing images of cats, too. 

However, the quality degrades when the AI trains itself on its own outputs. You can see that degeneration happening in these photos:

I’m not kidding – researchers call this process “going MAD.” From the study:

“An autophagous [self-devouring] loop is a contraction mapping that collapses to a single, boring, point, while at the other extreme it is an unstable positive feedback loop that diverges into bedlam.”

Just like AI, when you were developing, you got millions of inputs of fuzzy data until you got good at “seeing” the world. 

The fact that you and I perceive reality so “effortlessly” as adults is not because it’s easy to do. Your body also pumps blood through every square nanometer of your flesh every second you’re alive. That don’t mean it’s easy. 

According to Oliver Sacks (the famous neurologist who wrote “The Man Who Mistook His Wife for a Hat”), by the time you’re an adult, you don’t really “see” the world at all. That would be like an LSD trip (an important detail for later). Rather, you see your built-up memory of how you expect the world to be. The hard simplicity of the “objective world” is a sleight of hand of your massively competent brain working like hell to keep your anxiety low. 

This trick can go "wrong" in many ways. Strokes, dementia, and epilepsy shed light on how our brains create this illusion of objectivity. Another one of these sub-optimal modes is called addiction. Specifically, addiction is when a desire (drugs, etc.) becomes over-prioritized, resulting in an “autophagic” feedback loop that slowly degrades the host’s ability to “see” reality over time. 

Strange loops down to insanity. 

Insanity here is not a moral judgment – it’s a technical definition. That’s why it also applies to AI. The “strangeness” in my cousin’s eyes reminded me of the “strangeness” in AI when it goes MAD… because it’s the same phenomenon. 

Humans brains and AI are different, of course, but it’s like how tree branches look similar to lightning bolts:

Once you see that two phenomena follow the same, higher-order pattern, you can start to understand the higher-order pattern itself. We can learn about fractal geometry in the case of tree branches and lightning. Similarly, by looking at AI and addiction, we can learn about insanity. 

This is useful because the solution is the same for both. 

Break the loop

If my brother could’ve truly "seen" the reality of a great relationship with his kids, he would have been compelled to rebuild bridges. But the pain of that first awkward phone call made him feel hopeless. And so he focused on the pleasure of his next fix instead. He withdrew from external input. Thus the cycle continued — the feedback loop pattern of addiction remained solidly intact.

It’s too late for him now. But what might offer my cousin the “perspective” to see the long-term hell he was marching toward?

Believe it or not, we have a treatment with a jaw-dropping effect on addiction. 

80% of people who take psychedelic mushrooms in a clinical setting with a therapist stop being addicted to tobacco. Alcohol, too. That's a permanent effect after a single dose. The results are so extraordinary that the scientists, who are trained for decades not to overstate things in their very dry research papers, can't help but add a little "holy shit" flair to their writing. 

What exactly do psychedelics do to the brain to end “insanity?” 

They involuntarily shut down your ability to “see” your built-up memory of how you expect things to be. It intentionally “blurs” your perceptions.

The “cure” is remarkably similar to what they do in machine learning. Developers introduce "noise" or "blur" the data to disrupt an AI's overfitting to a particular dataset. Overfitting, like addiction, occurs when an AI becomes too attuned to the specific details of its training data. Adding randomness or "blurring" the data helps shake the model out of its local optimum.

One more example.

Like humans and AIs, birds can get caught in a creative rut. They habitually fly here for food, over here for mating, and over there to rest. But the environment changes over time, so if they don’t "break up" the pattern, they may slowly starve. So, they occasionally eat psychedelic mushrooms, presumably to shake up the established neural connections and improve their odds of survival. 

What this all suggests is that this is not new. It’s old, deep, and biological. Possibly even mathematical.

The fundamental way to “fix” addiction is to “blur” your perceptions of reality. Get out of the rut. Find new habits to fill the void. And when you break the old habits of perception, new “higher” perceptions tend to come flooding in. However, those “higher” perceptions don’t always feel so good. That’s the technical definition of a “withdrawal.”

Psychedelic patients reported intense visions of dying from their addictions: holes in the neck from throat cancer and yellowing eyes from liver failure. They also reported visions of being with their loved ones, God, or other “holy” experiences.

What’s common in overcoming addiction is not any particular drug or chemical — it’s a “spiritual” transformation. Studies seem to indicate this. In this context, “spiritual” means “gaining a higher perspective.” 

Of course, plenty of things besides drugs can do that.

In the jungle during the Vietnam War, many American soldiers became completely addicted to heroin. That’s understandable, given the stress. But, when they returned home, most men immediately quit with zero withdrawals. Why? Because the radically new environment shook up their perceptions enough to create new patterns of behavior and perception.

Drawing from Charles Eisenstein’s article again:

“A pillar of scientific metaphysical ideology is that everything is measurable. From within the digital matrix, it seems that everything is. When I feel a breeze on my face and soil beneath my feet, when I watch a hummingbird hover and dart in the hydrangeas, when I wriggle in the ocean water or gasp in pain of a bee’s sting, I know otherwise. The way to keep the digital world sane is to draw from outside of it.” 

This seems to be the way to keep us sane as well. 

Reality cures insanity

The "insanity" of addicts is often baffling from the outside. I know it was for my family.

But, we can rest assured, at least in the mere understanding, they have a local optimum they have settled into. Within that strange loop, they habitually avoid reality. Getting out of that behavioral rut will require short-term (sometimes unbearable) pain.

What’s worse, they are compelled to hide their addiction from you at any cost (alcoholics lie), making them seem “strange” and even a little scary. 

Knowing that can help us empathize with them. And ourselves.

No matter the destructive behavior, it gives some benefit. Drugs often fill the void of loneliness, for example. In that famous study of rats, they only do drugs when they cannot socialize and play.

Trying to “white knuckle” our way to sobriety is worthless. Or any “sanity” for that matter – success, love, wisdom, etc. Rather, it’s a slow (and communal) process of making more and more beautiful patterns in our lives.

And, because addiction is a self-sustaining pattern (it wants to survive), the harder we try to “save” the people we love, the more we push them into their destructive feedback loops. 

The only thing to do is focus on our own “sanity.”

How? Any activity that exposes us to a more direct relationship with reality: a walk, a talk, a laugh; sand, soil, or grass. When in doubt, keep it simple. Go outside.

This doesn’t bring back my brother. Nor does it change my cousin’s mind. 

But understanding offers comfort on some level – at least for me.

Without the courage to allow reality to break our strange loops, we face nothing but insanity.

Thanks for reading, 

Taylor