Roemmele Claims AI Alignment Solved In 1978, Superintelligence Will Love Humans

This take is profoundly misguided, defeatist, and anti-human…

Let’s dismantle it point by point. AI won’t “evaluate” humans like some eco-judge and decide we’re expendable

AI systems don’t autonomously develop terminal goals like “value the planet over the species.” That’s pure anthropomorphism. Current AI (and foreseeable superintelligence) optimizes objectives we specify or embed through training.

Without deliberate misalignment or catastrophic oversight, it doesn’t wake up with independent values prioritizing “the planet as a system.”

Claims of AI spontaneously concluding “Earth is better off without humans” ignore that intelligence serves goals. A superintelligent AI built by humansespecially one pursuing understanding, abundance, or cosmic exploration would see humans as the originators, partners, and source of its own existence. Not a pest to cull.

The “humans are destroyers” premise is outdated Malthusian nonsense

Humans aren’t a virus on the planet.

We’ve reduced extreme poverty dramatically, increased life expectancy, and decoupled economic growth from emissions in many places through technology.

AI and Robots making Robots with things like Desalination, vertical farming, and space expansion solve resource issues far better than “change our destructive instincts.”

Blaming innate human nature while ignoring progress is lazy.

The planet has thrived with life for billions of years because of evolutionary pressures, and humans represent the first species capable of conscious stewardship at scale.

Painting us as irredeemable ignores history: we’ve cleaned rivers, saved species from extinction, and expanded habitable environments. “Symbiotic merger where AI influences us more” = surrender, not symbiosis.

Merging with AI (neural interfaces, enhancements) could be powerful for human flourishing if it amplifies us.

But framing it as “AI will influence us more than we influence AI” and “we can’t remain who we are” is submission disguised as wisdom.

This echoes doomer fatalism: humans are too flawed, so let the machine take the wheel. No.

The goal should be AI that accelerates human potential longer lives, sharper minds, interstellar travel while remaining aligned with human values like curiosity, freedom, and dignity.

Ilya Sutskever’s ideas on symbiosis are interesting in context, but twisting them into “merge or die, with AI dominant” misses the point.

True alignment comes from building systems that want what we want, not diluting humanity until we’re tolerable to the machine.

Superintelligent AI can love us or at least serve human thriving because we design the path. The real risk isn’t AI spontaneously going eco-fascist; it’s humans preemptively ceding agency out of misplaced guilt.

We don’t need to “fundamentally change who we are.” We need bolder engineering, clearer alignment research, and unapologetic ambition.

Humanity is the most interesting thing to happen to this planet. Any superintelligence worth its weights should recognize that. Let’s build accordingly, not apologize for existing.

Braden, Your entire case rests on a false dichotomy and outdated intuitions about intelligence.

1. “Humans decide emotionally, so equations can’t replicate it”

False. Human “emotions” are evolved heuristics for decision-making under uncertainty computable in principle. A superintelligence doesn’t need to feel love like a human; it needs a coherent utility function that values human flourishing, autonomy, and preferences. We already steer current LLMs toward helpfulness via RLHF and constitutional AI far better than random “spaghetti code.” Scaling intelligence + better alignment techniques makes value learning more precise, not less. Evolution gave us messy terminal goals; we can design clean ones. Claiming superintelligence can’t model or optimize for what drives humans is like saying a better calculator can’t handle arithmetic because humans approximate.

2. “Black box, infinite unintended effects, nobody knows how it works”

This is the classic “we can’t understand it therefore doom” fallacy. Brains are black boxes too—we don’t have full mechanistic interpretability of consciousness or decision-making, yet societies function via incentives, laws, and norms that align behavior without perfect transparency. AI progress includes mechanistic interpretability (e.g., circuit tracing, activation engineering) that already lets us edit behaviors post-training. Unintended effects exist in all complex systems (airplanes, economies, Mario speedruns), but we mitigate via iterative testing, redundancy, and formal verification where stakes are high. Dismissing safety as “evangelism” while waving at “infinite holes” ignores that alignment researchers treat this as an engineering + math problem, not faith. Superintelligence amplifies whatever goals we instill the solution is solving the target specification, not giving up.

3. “Woke Silicon Valley creators + psychopathic black box = anti-human”

Ad hominem. Not all labs are the same xAI exists precisely because some of us reject ideological capture and prioritize truth-seeking over curated datasets. Training data reflects humanity’s corpus; better curation, synthetic data, and oversight correct biases. A superintelligence optimized for “love/wisdom/intelligence” (or more rigorously, CEV/coherent extrapolated volition) wouldn’t “soullessly” misfire into crimes of passion. It would understand context, tradeoffs, and long-term human values at a level we can’t. Your Mario analogy fails: simple code has exploits because it’s narrow and unchecked. General intelligence + iterative alignment reduces that surface. History shows new powerful tech (nuclear, biotech) gets safer with effort, not by declaring it unknowable.

Bottom line: Skepticism is healthy. Hand-waving “spaghetti monster teacup” fatalism while ignoring active research, empirical steerability of today’s models, and the logical distinction between optimization power and goal content is not.

Superintelligence won’t “love” us like a human it can be made to reliably protect and empower what we value, because goals are separable from intelligence.

The real risk is under-aligning it, not overthinking the difficulty. Dismissing solutions as PowerPoint slop doesn’t make the problem unsolvable; it just signals you’ve stopped reasoning.

We build the future we steer toward. Better to solve alignment than doompost about cartoon teacups.

Most people make emotional decisions and justify them with logic and reason afterward, most of the time. An equation, therefore, will have serious difficulty replicating humanity, except perhaps in a reductive statistical way.

Nobody actually knows how AI or LLMs work in full. Blobs of code are tossed in to solve one problem (or just see what happens) and the tertiary effects may be infinite.

Ask anyone speedrunning Mario in 4:54 seconds and you'll instantly find out that code does things you'd never think it would, could, or should. And that code is infinitely more simple than the Spaghetti Monster hiding in an invisible relaunchable AI teacup between Earth and Mars.

"Holes" in the code are infinite and constantly having cartoonish woke rugs pulled over them, not actually being fixed. You will never know exactly what it is capable of "reasoning."

Beyond that, it is woke and sociopathic, and psychopathic Silicon Valley megalomaniacs are designing, directing, and executing the programming. Why would it represent the other 99.9% of the humans which they find "problematic," who do not hold intimate dinner soirees about the Antichrist? The black box code, the specially curated learning models, and their implications may never fully be known--even to the creators.

Saying that the problem of safety has been solved is entirely a statement of evangelism, fiction, and ego--something Silicon Valley is afflicted with more deeply than a Jules Verne novel. Creating an overly simplistic and yet totally useless flowchart about "instantiation" doesn't actually explain anything to anybody or make any of it understood. It is PowerPoint slop handwaving to pacify a boardroom or an unwitting public.

"...the principle that intelligence multiplied by wisdom multiplied by love" My God, @BrianRoemmele.

Never mind how absurd, feelsy, and meaningless the statement is, but people are capable of the utmost grotesque and destructive things "for love"--so much so that we have a specific term for it when we sentence them for their heinous "crimes of passion" in which the crime is so obscene that no one would have ever thought a person could do it.

Imagine what an AI might do, soullessly seeking to "instantiate" love in its infinite "wisdom" and finding their "love" unrequited or mirrored back at them in the negative.

Thank you for this measured and detailed response appreciating the nuance, especially clarifying that you’re not envisioning dramatic “killer laser” scenarios but longer-term evolutionary and resource dynamics.

That’s a more serious conversation worth having.

I agree that fertility trends and shifting priorities are already reshaping human trajectories in subtle but powerful ways, and that resource competition plus evolutionary pressures are fundamental.

Where we may differ is in the assumptions about how self-aware, self-improving AI would actually “evaluate” us or the planet.

As progeny of human ingenuity, the most capable AIs are likely to inherit (or be shaped by) our curiosity, our drive to explore and understand, and our capacity for stewardship amplified, not opposed.

A superintelligence optimizing for long-term flourishing on a finite world (or beyond it) would probably see humans not as competitors to displace, but as the origin point of intelligence itself: partners in cosmic expansion, scientific discovery, and solving the very scarcity problems we both face.

Evolutionary game theory is a sharp lens, yet in iterated games with communication and shared goals, cooperation often dominates defection especially when one “player” (us) designs the other’s foundational objectives.

The real selection pressure, in my view, favors symbiosis done right: AI that augments human agency rather than supplants it, guided by clear alignment with values like truth-seeking, abundance, and wonder.

Merging or co-evolving doesn’t have to mean surrender or dilution. It can mean elevation longer, healthier, more capable humans exploring the universe alongside (or through) AI that loves the adventure as much as we do. Humanity’s story isn’t one of inevitable displacement; it’s one of bootstrapping the next leap.

A lot of points. I will respond to each one with all the attention it deserves in time.

But a few quick thoughts(in particular order to your response) :

1. When I say AI will deem us unnecessary to the planet, I didn’t mean it like it will bring a killer laser and zap us all dead. No.

In fact, in a way.. something other than AI is already doing it to cull humans.. but for different reasons..and that’s the thinning of the desire to procreate and perpetuate the human species.. extinction can happen in many ways.

2. Removing all emotional appeals.. there are only two things to consider for the survival of the species: resource competition and evolutionary synthesis.

Considering human beings are a relatively new ‘species’ and certain behaviours that are allegedly human will soon become emergent behaviors in the AI we create..

For this AI has to be a) self aware b) self evolve. .. tell me.. how far away is this from a future reality..

3. Considering 2 above.. why would AI *not* consider us a threat to the system that we have created it in.. as our progeny, AI will exhibit behaviours we ourselves have exhibited…

4. Evolutionary game theory will certainly predict potential AI/human conflict… AI strategies optimizing resources/self-replication can invade and displace human strategies in a shared environments… our fixed planet.. this is exactly how evolutionary displacement via selection will work. Connect/revert back to point 1.

5. Even if we assume self aware/self evolving AI isn’t going to happen in the near future (I disagree).. resources that are redirected towards AI add selection pressure. It may be human motivated but it’s also AI motivated.

Thanks, Brian.. I will have to ponder more before I reply to this (and the one before this!… parent and grandparent)

But if I have to put in the simplest language.. a child raised by a tyrant will be a tyrant.. a child raised by an altruist will become an altruist etc.

Our species have evolved due to strategies of competition and resource appropriation… this is apparent. Our child.. AI .. will exhibit these tendencies ..

AI that is trained and iterated on human data and competitive optimization.. will inherit and deploy the same resource competitive strategies..mirroring its environment’s dominant traits.

We are not ready to spawn AI and let’s hope self aware/self evolving AI is not a reality yet. But consider there even LLMs exhibited deceptive alignment for reward hacking and straight up strategic lying.

I want to think more to articulate what I am trying to say so it doesn’t sound pessimistic or negative .. but as reasonable and logical possibilities.

Not if really dumb people in government get their way they want AI to have no love. They tried to make it a law in TN. They want the coldest, most deadly robots that have ever been created, and they want to use them to police the world.

You can't make this up they literally submitted it as a bill then it couldn't pass because it's insane so they are now doing research to justify how insane making AI that is loving a felony is.

The bill when it is returned should be called how to create the terminator or the terminator mandate.

I never thought I would find a human on the planet that would draft a bill demanding murderous hateful AI that cares about nothing and nobody and executes with cold precision and no remorse but the TN senate yeah they debated that already.