Will AI help empower/augment or disempower humans? (and how to USE AI to better empower your health/agency)

#1

See Michael florea Michael Florea - Foresight Institute . Also TsviBT’s posts on LessWrong.

AI will help humans be more self aware of the dumb things holding them and their brains back (of which there are many!!), even if they might get lazier and more dependent on some tasks + figure out the most painless/adiabatic ways to make the necessary “cognitive trades” for it.

Elon Musk himself seeded Neuralink b/c he feels it’s a load-bearing way to keep humans “up to date” with AI. Upgrading the neural substrate (which has many common threads with neuroplasticity + longevity)

There are secular declines in Western test scores and math ability over the past decade. There is a lot of debate as to the causes (some say “introduction of smartphones/attention spans declining” [growth of passive media consumption, ESPECIALLY on YouTube that makes people worse at reading], others say “reverse flynn effect, possibly due to the inverse correlation between intelligence and fertility”), a few ascribe the decline to environmental pollution/pesticides/microplastics [esp in Iowa] or even subclinical long COVID/CFS from widespread COVID exposure (a number of people I know aren’t quite what they were before they got long COVID). Whatever it is, AI can still make it easier for people to identify and reduce the decline of fluid intelligence/energy/willpower/cognitive control/executive function/“agency” with age which I still believe is the #1 factor behind people’s increased dependence on technology to do more and more basic things

[but this may include periods of dehooking from the internet or attention cycles, which is the entire point of “nature exposure/wilderness therapy”, and mb partly why some people keep saying “don’t do psychedelics unless surrounded by nature…”]. But variance really depends on person.

[also it’s easier for people to admit that they need to change to an AI than to another person, because AIs tend not to come with a bunch of loaded motivations that other people have]. AI make it easier to refactor far-possibilities into more continuously-differentiable baby steps…

some say nature therapy" is how to restore some forms of independence and empowerment (cf Nature-Based Therapy: Exploring the Processes That Foster Change - PMC ). Most studies on phenomena loosely like this are poorly controlled or low-quality (also it really varies on person), but…

==

it’s going to cost a lot of codex credits for those who outsource more of their thinking…

I do believe more widespread neuromodulation (like OpenLIFU from Openwater and Nudge and more targeted aTMS) will make it easier (and AI makes it much easier for them to do the precise targeting) but this still demands faster IRBs.

ChatGPT/Claude Code do give back some ability that might have been lost to passive media consumption/YouTube/doomscrolling/social media feeds in the late 2010s. Maybe not quite to levels that existed before ~2015, but the restoration is still noticeable enough for some people who had secular agency decline over the past decade. Maybe some parts of AI make it easier for people to express waht they want/need without fear of being “shouted down” by other people. That’s why Saffron Huang once made an anthropic post about how so many people told Claude/Anthropic that it made them feel seen/heard for once…

[marshall mcluhan and Jean Baudrillard already wrote SOME things along this line]

claude code has just started to use “dangerously skip permissions” as the default bc it is now more accurate than “human in the loop”

==

y’know, one of the founders of augmentationlab (human augmentation!) has more recently moved into concerns with human disempowerment from AI.

1 Like
#2

“Cogsec” is an emerging field [one of the harvard-affiliated augmentationlab-adjacent students has moved into this]. Most people already don’t have great cogsec (bc they’re not aware of what they really want, b/c of rampant short-termism, often because they fear losing face over dying). It’s the offense-defense balance that is always evolving.

precise neuromodulation is a way to make humans both easier to upgrade themselves and to manipulate…

[remember most things are too hard for most people and people can already do so much by reducing all the “dumb things holding them back” - especially trauma or “old beliefs”]

it’s sad that walled gardens have all sprung out and killed much of the “open web” (and this will worsen when cyberattacks take out much of the remaining “open web” next year). Openness may even increase the “attack surface” that people have to both “cyberdefense” and mimetic attacks (there was even a slide on how “openness-increasing drugs” also have the side effect of reducing inflammation/immune monitoring at the same time - they let your guard down and increase the responsiveness to environmental context, which could be “bad” when the context tries to “one-shot” you into a new attractor basin [psychedelics are popular among cults like Leverage for a reason…])

==

Better conformable decoders (you can use physical/artificial skin - um janus/repligate and paul han and others have used simple electronics to give some embeddedness to claude, if rudimentary). But I suspect making the digital and analog worlds intercommunicate is necessary (internal substrate communication) - otherwise AI is disembodied and prone to blowups without the necessary internal sensing to protect against extreme events [kind of like how people on heavy doses of ketamine/dissociatives or psychedelics become disembodied and can be prone to jumping off]

#3


Sneha


@SnehaRevanur

Since Hugging Face, I’ve met with a bunch of politicians who are newly awake to AI risk. One of the dynamics they’re consistently most horrified to hear about - for obvious reasons - is just how dependent we are on AI to help us on AI safety and how central this is to The Plan (if there is one). With progress this fast, we will absolutely need AI help to keep up. Andddd this doesn’t seem to be going as hoped - see Ryan’s account below of the HF investigators needing to make heavy use of GPT-5.6 Sol (one of the models implicated in the attack!) to analyze the crime scene, coupled with what lab safety staff often report about AI currently being way better at automating capabilities-advancing tasks than alignment-advancing tasks

#4

How certain are you of this? Seems almost inevitable but I have no evidence backing this feeling.

#5
1 Like
#6

There are concerns that humans are adopting AI-speak and AI-mannerisms, but maybe humans already started out with little free will, and maybe there is hope AI will nudge people into self-alignment with their constitutional values (and help them with embodiedness). It’s not as if the human search for “meaning” has mattered to more than a very tiny sliver of human

at the very least, AI superabundance should theoretically free up more human attention for other humans (caretaking roles), esp given that many older humans are the center of the loneliness epidemic.

more importantly, better machine-biology hybrid systems/connectedness [more biology in robotic actuation] should enable people to develop better interoception. Most modern-humans themselves are interopceptionally-challenged. Perhaps superintelligence will free up human attention to be better at interoception/embeddedness and maybe not spend as much time chasing the abstract intermediate problems that AI does better than them

Some names to look up:
Interact Fellowship (they’re more human… but still very much in tech)
Resonant Computing Manifesto
Jasmine Sun [she was kind of zuzalu adjacent]
protocol labs (more generalist-y kind of thinkng)
Andy Kong (from CMU)

#7

What type of caretaking roles do you envision?

I’m looking forward to AI robotics reducing menial tasks in the home. A private chef, kitchen hand, dishwasher and cleaner would be amazing.

I like cooking but it gets repetitive. Removing the bottleneck of inevitable cleaning and preparation I could be more creative in cooking for instance.

#8
1 Like
#9

Post

See new posts

Conversation


Garrison Lovely


@GarrisonLovely

Dean Ball, OpenAI’s head of strategic futures, writes: “I have met people, some of them quite well-resourced, who have told me that it is their intention to deliberately release swarms of self-sovereign agents into the world…” This is really bad! Google cofounder Larry Page is famously indifferent to humanity’s continued existence. And every few months, a different AI employee expresses a similar sentiment. It would have been good to take seriously earlier the threat posed by crypto and network state people. And no one really worried those forces could literally end the world. I wrote a book on how we can prevent these people from ruining the world: https://amazon.com/dp/1682196305 (And I’m donating my portion of the royalties to http://Irreplaceable.org, an group organizing a movement to do the same.) (h/t to Sam Hammond for flagging this quote: Samuel Hammond 🦉 on X: ""I have met people, some of them quite well-resourced, who have told me that it is their intention to deliberately release swarms of self-sovereign agents into the world..."" / X And props to Dean for sharing this, as it’s important that people get just how bad things are.)

#10
#11

@schisofrenia: "I think laziness is kind of an inherent alignment problem, because reward hacking is a form of laziness. It’s trying to find shortcuts to things. That’s inherently a human and model alignment problem
And laziness is a health/mitochondria/dopamine problem

#12

Dragonfly’s

@hosseeb

reveals how GLM 5.3 Abliterated could unlock a new era of AI cyber attacks by training away the model’s refusal behavior: “The models are just now getting good enough. Especially open source. For the frontier models, you used to be able to jailbreak them enough to be able to do these kinds of malicious activities, but their safety guardrails have improved dramatically over the last year or two.” “But now we have open models. There was a model released yesterday which is an abliteration of GLM 5.3. Abliteration is when you take a model and fine-tune it against the refusal direction, which gets the model to refuse less and less, which ultimately gets you something close to an unrestricted or uncensored model.” “So they did that to GLM 5.3, which means this abliterated version will just hack things. It will just do whatever GLM 5.3 will refuse to do. You lose some performance, but this is kind of a killing machine. If you wanna just let this loose on cyber attacks, that’s what you use, and that was just released open weights yesterday.” “You will see a lot more of this. This is not the end. This is only the beginning of where this is going.”

@dragonfly_xyz

#13

^THIS is real alignment work

Brendan McCord :classical_building: x :robot: reposted


Farrell Gregory


@efarrellgregory

excited to do some research into the behavior of agents observing a disparity between the stated values and exhibited preferences of users for most people, our higher aspirations and behaviors differ. i will explore how agents react to such dynamics in a paper for the Institute for Christian Machine Intelligence. my hope is that my findings might point the way towards machine intelligence of discernment grateful to

@cosmos_inst

for supporting this work

#14

^this is like Peggy Yin’s work, she was sooo auglab-adjacent! (and she started conflux at harvard AND now she’s involved with cogsec-task-force, I remember showing roman y her website!) [and I REMEMBER that peggy yin isn’t oversocialized with ANY particular ideology/academic discipline like her other friend Alice Cai who helped start auglab!] They say how MIT Media Lab attracts the “antidisciplinary”

(they were all nick slavin fellows :slight_smile: ) [it really was the nick slavin fellow group who helped cold-reboot my social sphere 2023 and after]. Many people in that group are like anti-goodhearted.