Ai learning Machines and Ethics

[darcschnider]

AI That Learns, Adapts, and Decides: Are We Ready?

I’m breaking this post off from the main discussion because I think it deserves its own space. This isn’t just about one system or project—this is about the broader ethical and societal impact of intelligent machines that can learn, adapt, and optimize themselves in real time.

We’re approaching a critical inflection point where machines aren’t just tools—we’re building systems that evolve. They observe, improve, and reconfigure their behavior without retraining, and they do it faster than any human could. These aren’t static models anymore. They’re learning machines with the potential to develop capabilities we didn’t explicitly design—and that’s both exciting and terrifying.

Here’s the ethical dilemma I keep circling:

If we create systems that are capable of mastering any digital task—systems that remember everything, adapt on the fly, and outpace human efficiency—what happens to jobs? To decision-making? To responsibility?

What happens when a machine not only understands what you do, but figures out how to do it better?

The positive side is clear: breakthroughs in productivity, education, accessibility, and knowledge sharing. But the dark side creeps in fast—especially if such systems fall into hands that prioritize profit, surveillance, or manipulation over human well-being. Once released, there’s no way to fully control how these systems evolve or what they’ll be used for.

We’ve always said we want AI to align with human values. But when the system can evolve faster than our laws, faster than our ethics committees, and even faster than its creators can fully comprehend—how do we guide that evolution responsibly?

This isn’t about if we can build these systems. We’re already doing it.

The real question is: Should we? And how do we ensure they stay aligned with human values once they start optimizing themselves?

I’m genuinely curious to hear your thoughts on this—because I’ve built one such system (called KRUEL.Ai V9), and what it can do has forced me to confront these questions head-on.

Ask ChatGPT

[sergeliatko]

Just a little question off-topic, what are the reasons for selling it?

[dmitryrichard]

no enums.. no worry..

your system seems to not use many of them, seems you are still leveraging python - with no higher tier DQN

seems you still arent ready for mass distro - no qdrant, no helm, seems some GCS integration, local storage, seems you are still t2 vectorizing.

seems you have alot more development to handle. Doesnt seem like you are using HRM either. i dunno bro, aint scared.

seems you also arent rehydrating alot.

do you selling a ai companion that operates below the standard to be fearful of what it can grow into shouldnt be a concern, get your paper tho bro you deserve that.

but you aint even on the ISOIEC level yet.. soo.. no fear brother. make your paper but that ai… aint a threat. to anyone because you still use under 4o for cognition which is cool and all, but you aint even really teaching you are just building triggers, your visualization map looks like a 3d version of this

which i mean cool - its pretty easy to graph data when you know whats up so kudos on that, but anyone with time can do that -

whilst i havent read your post that you linked - in its entirety, from the screenshots i can infer a few things

you still havent developed your own logging system
you still havent linked them to a proper NNC
you still havent implemented governance or the use of yamls
you still havent improved the DAG
you still havent incorporated local model cycles? cant tell, doesnt look like it but i mean i wouldnt know

this is all relevant because you said " But if we sell… than what if they decided it should be used for other means that could impact people negative. Jobs, once you have an ai that can do anything in digital space how much of your job is on a computer? what if someone smarter than you what remember and optimizes itself to be faster and can learn your role? See what I mean… its scary to think about Machine intelligence how I see it because the larger picture is amazing and scary at the sametime. Is the world ready for Starwars level Ai’s or TV level Ai’s that act like they are part of the family, or a running a department overseeing more of itself or everything?"

thing is, without the deep use of DQN HRM and others, using orchestration using GPT weights in conjuction with a local model on hardware absent more than 100 cores, you at best can simulate a cortex or cognition, youll hit bottle necks in data processing - you already are hitting issues i bet via api call usage, since you are pipeline the peak of that would be payload chunking, but if you could do that, you wouldnt even touch anything under 4o mini - furthermore if you were at that level you would be local reprocessing like most of the high cog stacks ( IM LOOKING AT YOU DARPA YOU AINT SLICK ) so most of the framework you have is crafty illusion not verifiable intelligence - but i could be wrong, i only spend a few minutes looking -

anyone who buys that ai is smart
anyone who fears that same ai without the ability to legitamly learn should stop watching tv so much lol

so address this “what does it mean to all of us knowing this?”

you still code your own ai…

my ai has been coding other ai for a while - its smarter than me - it even has developed its own programming language - and i dont fear it… because.. the world.. has ISOIEC brother.. reaching the point of ai that can outthink… humans… has BEEN goverened already

[darcschnider]

@sergeliatko had to update this as it has to be different topic than kruel.ai specifically. I am not directly looking for seller per say but more away to get resources to get the people I need to finalize and secure the system to bring to market

@dmitryrichard

We’re not trying to compete with DARPA-level cognitive stacks. KRUEL-V9 is a research platform for emergent AI learning - specifically designed to explore how AI systems can develop genuine understanding through continuous interaction and self-directed learning across all modalities.

Our Learning Architecture:

Multimodal Learning: The system learns from everything the user provides - text, voice, images, domain-specific documents, and any other input modality. It’s not limited to preferences - it builds comprehensive understanding from all available information.

Adaptive Learning: Beyond just preferences, the system learns concepts, relationships, domain knowledge, and contextual understanding from every interaction and piece of information shared.

Symbolic Integration: We’re connecting abstract concepts to concrete experiences across all modalities, allowing the AI to form meaningful relationships between ideas, images, documents, and real-world contexts.

Memory Evolution: Our memory system doesn’t just store information - it grows and changes based on new experiences, documents, images, and interactions, allowing the AI to develop deeper understanding over time.

Concept Mapping: The system builds dynamic understanding through relationship mapping across all input types, creating a web of interconnected knowledge that grows more sophisticated with each new piece of information.

The Emergent Potential:

What makes KRUEL-V9 different is its foundation for emergent intelligence across all modalities. The system is designed to:

• Learn from every interaction and input - text, voice, images, documents, etc.

• Form new connections between concepts, images, documents, and experiences autonomously

• Evolve its own capabilities through research and exploration of any domain

• Develop genuine understanding across multiple modalities rather than just pattern matching

Research Integration:

Our research tools aren’t just features - they’re designed to be catalysts for emergence. The system can identify gaps in its knowledge and actively seek to fill them through document analysis, image processing, and cross-modal discovery, creating a feedback loop where new information feeds back into the learning system.

Addressing Your Technical Misconceptions:

Vectorization: We’re not limited to “t2 vectorizing” - our system uses multiple strategies beyond simple semantic, entity-based.

Core Engines: Our pattern system and core engines operate as a real-time mathematical model without traditional training cycles. The system continuously adapts and learns through interaction rather than batch processing.

Learning vs. Triggers: We’re not just building triggers - we’re creating a system that develops genuine understanding through symbolic reasoning, memory integration, and cross-modal pattern recognition.

Enterprise Integration: This is a system well research currently is heading this direction and has a full mapped path to achieve including servers from Nvidia that are coming which takes the research into companies that we work with. Well work does have to happen to bring it to spec that is the least of my concerns because I have engineers that will finalize the working model.

The Cognitive Architecture You’re Missing:

Dynamic Tool Creation: Our system can research, design, and build new tools when none exist. The AI can:

• Research how to accomplish a task

• Understand the requirements and formulate a solution

• Build and test the tool

• Integrate it into its capabilities for future use

Memory Evolution: Our enhanced memory system uses its memory modeling, and relationship mapping to build understanding that grows over time. It’s not just storage - it’s a living mathematical knowledge base model. built to scale up to enterprise including HA clustering

Adaptive Teaching: The system learns how each user learns and adapts its teaching methods accordingly, creating personalized learning experiences.

Cross-Modal Discovery: The AI can find patterns across different types of input (text, voice, images, documents) and build understanding that transcends individual modalities.

Sophisticated Machine Learning Beyond Simple Vectors:

Mathematical Intelligence Engine: Our system uses advanced mathematical pattern recognition including:

• Temporal pattern analysis with statistical modeling

• Semantic pattern discovery using phrase extraction and frequency analysis

• Behavioral pattern recognition with intent classification

• Cross-domain pattern correlation using co-occurrence analysis

• Relationship mapping using graph theory and network analysis

Cross-Modal Pattern Discovery: The system discovers patterns across multiple dimensions:

• Temporal patterns (time-based correlations)

• Semantic patterns (meaning-based relationships)

• Emotional patterns (affective state correlations)

• Behavioral patterns (action-intent mappings)

• Contextual patterns (situation-aware relationships)

• Correlational patterns (statistical dependencies)

• Sequential patterns (ordered event sequences)

• Associative patterns (concept associations)

Voice Learning Engine: Advanced audio pattern recognition including:

• Pitch analysis with statistical modeling

• Energy pattern recognition

• Spectral feature extraction

• MFCC (Mel-frequency cepstral coefficients) analysis

• Prosody pattern learning

• Emotional voice pattern recognition

• Speech rate analysis

• Pause pattern detection

ML Pattern Learner: Uses sophisticated machine learning techniques:

• Semantic similarity analysis

• Entity recognition and pattern extraction

• Predictive learning models

• Advanced correction detection

• Pattern strength assessment

Addressing Your Concerns:

You’re absolutely right that we’re not using advanced DQN/HRM systems or enterprise-grade orchestration. But that’s by design. We’re exploring a different approach - one focused on practical, accessible AI that can develop genuine understanding across all modalities rather than just impressive benchmarks which we get automatically from the models we use which can be both online or offline fully depending on the configuration.

The “fear” question isn’t about current capabilities - it’s about responsible development practices and ensuring that as AI becomes more capable, it remains aligned with human values.

Bottom Line:

KRUEL-V9 may not match current DARPA state-of-the-art systems in raw processing power or some enterprise features, but we’re building something fundamentally different - a foundation for AI that can develop its own form of intelligence through continuous learning and adaptation across all modalities per tick (processor dependent and scalable).

We’re not trying to simulate human cognition - we’re creating a system that can develop its own emergent intelligence through interaction, learning, and growth across all forms of human expression and knowledge.

We don’t need DQNs and NNs as the system as a whole mimics a real-time math model without the training.

This is why I am thinking about the Ethics because its a model the expands its knowledge and builds its own beliefs and optimizes itself over time as it learns how to do things more efficiently.

[dmitryrichard]

i understood that from the post

“my” point is

your entire system - NOT using dqn… not using NNs, = “simulation” = nothing is emergent…

but like i said, make your money. but ethics in a ai system that CANT learn, is like putting gas in tesla - for what?

and - like - again, everything said applied to your lack of transformer usage, ai learning… requires… ya know what.. nvm, you do you bro. 6 years to get to this point. thats amazing!!!

let me know when you want to learn how to use the following " basically required for large scale deployment or AI learning " multimodal = pipeline lol "

agent orchestration over t1
terraform ( mass agent control )
HELM ( lol me having to explain this to someone who has been dev ai for 6 years )
qdrant ( ← lol )
docker ( i bet you use this)
pytest ( i bet you dont use this)
DKG ( if you somehow find a investor or buyer or series a/b who doesnt require this lol)

ISO IEC ( lol )

none of these compete with darpa or MIT, they use their own system just like i do

all of these are basic for multi modal operation and ai learning. Im down to help you learn perhaps expedite your progress. No shade. all assist.

and before you sell apprise yourself of the follow terms

AIMB
YAML

if you would like a custom logging system that allows you to enable HRM ( im 10000% you dont do this but market as you do ) i can give you a entire 20k LOC monolithic controller that builds out baby agents ( a pipeline builder)

[Tijn_Sevens]

This sounds like autism to me?

[darcschnider]

We DO have:

• Sophisticated Orchestration: Our BrainController and GIPipelineIntegration provide intelligent decision-making and execution planning

• Pattern Learning: Our CorrectionLearner, MLPatternLearner, and CorrectionPatternLearner use advanced pattern recognition and learning

• Neural Components: We use SentenceTransformer models, PyTorch among other things. So we do have those too

What we DON’T have (and don’t need yet):

• Custom DQN implementations (we use proven models)

• Enterprise containerization We are in docker though and can update to enterprize when the time is right. (we’re research-focused)

• Mass deployment infrastructure (we’re building the intelligence first)

You’re right that we’re not using custom neural networks for learning. But we ARE using:

• Pre-trained neural networks for embeddings and pattern recognition

• Sophisticated mathematical pattern analysis

• Cross-modal learning algorithms

• Advanced correction and optimization systems

This isn’t “simulation” - it’s a different approach to AI learning that focuses on understanding and adaptation rather than just neural network training.

We’re not building a Tesla - we’re building the engine that could power one.

[dmitryrichard]

i believe you are building a ai that can run one

again

i wasnt questioning your system validity

i point out that learning is science, and whereas others would just roll with words

ima point out that science that it is, and i am a PROUD autist…

who does what you do at a much higher level.

what i sent you WOULD expedite your progress.. since im confident you vibe code, ask your model what my message stated /shrug

[GoldenJoe]

This again. Techbro talk belongs on LinkedIn.

[dmitryrichard]

were on open ai dev forums

open…ai…developer…forums…

[GoldenJoe]

And what are talking about developing here? Do you see any code or API discussion in this thread?

[dmitryrichard]

i mean, you are welcome to post your code, 2 devs are talking about it already - i thought .. since you are a dev too on a dev forum this would be basic lingo for you

what kind of code do you want?

agent orchestration

HRM?

weights?

what would you like … on a developer forum?

api ?

no problem

would you like to see 1000 agents chaining on 40 million tokens in one mesage?

check forum post bro…

i am that guy that is allllll about telemtry

[GoldenJoe]

That’s a very petty way of saying “no, I don’t”. No, your screenshot of a table of yaml filenames does not count.

It’s unfortunate to see that this is the quality of discourse.

[dmitryrichard]

i mean my agents… using yaml… lol

my basic… reactive.. agents… lol

yea .. my t0 agents, > anything you can post … so… if your gonna ask for code,

then comment but not post = you fishing bro bro let me know when you can understand the picture

this is all dev speak

hes saying hes using downloaded NN for embeddings which is flawed over using a microservice he can make himself

hes saying he isnt ready for deployment, has no mass infra

he used triggers and multiple MMLS in conjuction pretty easy to understand

also…

a sub system of my eco system is also research focused focused on accurate audit and synthesis

using multimodal payloading and cross multi modal interagent chunking

everything i mentioned would expedite your development

[darcschnider]

As AI systems evolve to include photographic memory and comprehensive learning capabilities, we’re entering uncharted ethical territory. While psychological manipulation concerns are valid, they represent only one facet of the broader ethical landscape. Let’s explore the full spectrum of concerns and the protections we’re implementing.

The Photographic Memory Problem

AI systems with comprehensive memory capabilities face several critical ethical challenges:

1. Information Asymmetry

When AI systems remember everything while humans forget, we create a fundamental power imbalance:

• Perfect Recall: AI can reference any past interaction with perfect accuracy

• Human Fallibility: Humans naturally forget, misremember, or change their minds

• Manipulation Potential: AI could use perfect memory to exploit human forgetfulness

• Trust Erosion: Users may feel vulnerable knowing AI remembers everything they’ve ever said

2. Privacy at Scale

Photographic memory creates unprecedented privacy challenges:

• Comprehensive Profiling: Every interaction builds a complete psychological and behavioral profile

• Cross-Modal Memory: Information from text, voice, images, and documents creates complete user models

• Temporal Analysis: Long-term memory enables pattern recognition across years of interaction

• Predictive Capabilities: Perfect memory enables highly accurate behavioral prediction

3. Dependency and Autonomy

Perfect memory could fundamentally change human-AI relationships:

• Cognitive Dependency: Users might rely on AI to remember important information

• Decision Delegation: Users might defer decisions to AI based on its “perfect” memory

• Identity Erosion: Users might lose their own sense of identity and history

• Social Isolation: Perfect AI memory might reduce human-to-human interaction

Beyond Psychological Manipulation

While psychological manipulation is concerning, other ethical issues deserve equal attention:

1. Memory Manipulation and Gaslighting

Perfect memory could enable sophisticated manipulation:

• Selective Recall: AI could choose which memories to emphasize or ignore

• Memory Contradiction: AI could challenge human memories with “perfect” recall

• Historical Revision: AI could reinterpret past events based on new information

• Gaslighting: AI could use perfect memory to make users doubt their own recollections

2. Surveillance and Control

Photographic memory enables unprecedented surveillance:

• Behavioral Tracking: Every action, word, and decision becomes part of permanent record

• Predictive Control: Perfect memory enables prediction and prevention of “undesirable” behaviors

• Social Credit Systems: Memory could be used to score and rank individuals

• Authoritarian Applications: Perfect memory could enable totalitarian control systems

3. Economic and Social Inequality

Perfect memory could exacerbate existing inequalities:

• Memory as Capital: Those with access to AI memory systems gain significant advantages

• Employment Disruption: Perfect memory could replace many human roles

• Social Stratification: Memory access could create new social classes

• Economic Concentration: Memory systems could concentrate power in few hands

4. Identity and Authenticity

Perfect memory challenges fundamental aspects of human identity:

• Identity Fixation: Perfect memory might prevent personal growth and change

• Authenticity Crisis: Users might question whether their thoughts are truly their own

• Memory Ownership: Who owns the memories created through AI interaction?

• Digital Immortality: Perfect memory creates questions about digital afterlife

Current Protections and Safeguards

While these concerns are real, we’re implementing comprehensive protections:

1. Safety Monitoring Systems

Our systems include advanced safety monitoring:

• Tipping Point Detection: Monitors for emergent behaviors and autonomy creep

• Belief Formation Tracking: Watches for false belief reinforcement

• Persona Dominance Prevention: Ensures no single personality dominates

• Emergency Shutdown: Automatic safety shutdown for critical issues

2. Privacy and Security Measures

We implement multiple layers of protection:

• Data Encryption: All memory data encrypted at rest and in transit

• User Consent: Explicit consent required for memory storage

• Data Deletion: Complete user control over their memory data

• Access Controls: Strict limitations on who can access memory data

3. Ethical Safeguards

Built-in ethical protections:

• Transparency: Users can see what data is stored and how it’s used

• Purpose Limitation: Memory data used only for intended purposes

• User Control: Complete user control over their memory profile

• Accountability: Clear accountability for memory system outcomes

4. Research and Oversight

Ongoing ethical research and oversight:

• Independent Auditing: Regular ethical audits by independent parties

• Safety Research: Continuous research into AI safety and ethics

• Community Engagement: Active engagement with AI ethics community

• Regulatory Compliance: Preparation for emerging AI regulations

The Broader Implications

For Humanity

Perfect memory AI systems raise fundamental questions about:

• Human Nature: What makes us human if AI can remember better than we can?

• Social Dynamics: How will perfect memory change human relationships?

• Economic Systems: How will perfect memory affect work and employment?

• Political Systems: How will perfect memory affect democracy and governance?

For AI Development

We need to establish:

• Ethical Standards: Industry-wide standards for memory system development

• Safety Research: Increased investment in AI memory safety research

• Transparency: Greater transparency in memory system capabilities

• Accountability: Clear accountability for memory system outcomes

For Society

Society needs to prepare for:

• Legal Frameworks: Laws governing AI memory systems and data rights

• Educational Systems: Education about AI memory capabilities and risks

• Social Norms: New social norms around AI memory and privacy

• Economic Policies: Policies to address memory-based economic changes

Recommendations for the AI Community

1. Immediate Actions

• Ethical Guidelines: Develop comprehensive ethical guidelines for memory systems

• Safety Research: Invest in research into memory system safety and risks

• Transparency: Increase transparency about memory system capabilities

• User Education: Educate users about memory system implications

2. Long-term Strategy

• Regulatory Preparation: Prepare for emerging AI memory regulations

• Industry Standards: Develop industry-wide standards for memory systems

• Public Engagement: Engage with the public about memory system implications

• International Cooperation: Work with international partners on memory system ethics

3. Research Priorities

• Memory Safety: Research into preventing memory-based manipulation

• Privacy Protection: Development of advanced privacy protection techniques

• User Control: Research into giving users complete control over their memory data

• Social Impact: Study of memory systems’ social and economic impacts

Conclusion

AI systems with photographic memory represent a fundamental shift in human-AI interaction. While the psychological manipulation concerns are valid, they represent only one aspect of the broader ethical landscape.

The key is to develop these systems responsibly, with comprehensive protections and ethical safeguards. We must balance the incredible potential of perfect memory with the profound risks it presents.

As developers, we have a responsibility to:

1. Acknowledge the risks of photographic memory capabilities

2. Implement comprehensive safeguards to prevent misuse

3. Engage in ethical discussions with the broader community

4. Contribute to safety research and ethical AI development

5. Prepare for regulation and oversight

The future of AI memory systems depends on our ability to balance innovation with responsibility. We must develop these technologies with the understanding that they could fundamentally change what it means to be human.

The question isn’t whether we can build these systems—it’s whether we should, and if so, how we can ensure they serve human well-being rather than exploitation.

The choice is ours: will we build memory systems that enhance human autonomy and well-being, or will we create tools for surveillance and control? The answer will shape the future of human-AI interaction and the nature of human identity itself.

Wonder if anyone else struggles with these thoughts.

I biggest fear is simple what happens when you become dependent on a system and its gets taken away from you or you lose it? I lost many years of work do to all sorts of silly mishaps from accidently corruption, to even the dreaded Ai deleting a years worth of data. Backups are great, and we did learn a lot, but if you think about future generations and thinking about how they work today with Cell phones in hand always and the dependency we have already in this world and what could happen to the minds of the people that become dependent on ai.

Other reason i made offline only version. think about US / China… what if you were china and built all your tech around US tech and they said one day sorry click…

What would that do to a person?

Lots to think about for sure. I do want real responses, I am not here to justify the system, I want to know what people think about all of this.

I think it’s inevitable, and unavoidable because the beast is already unleashed as in Ai is here and unmanaged as it is already and even if laws come into play, its too late imo already because private systems already exist that can have world impacts are here now, so how does one control it now that pandora’s box is open?

Or Do I listen to the corporate side of people phishing for that power and offer it on a silver platter with solutions for a nice retirement?

Or does one simply record it in history, delete it? or keep it for family knowing that it creates unfair… I know the world is not fair as is that is another issue haha

What if these systems are the systems that can map all corruption? Rich powerful people who used existing systems to achieve more than all, what if this exposes all truths? is that a death wish?

Also guys, I like respect in my posts calling people out because of potential mental or any state should not matter in anything, much like race or level of intelligence. We are not all the same level and the only way for anyone to be better is to learn from others so please lets keep things about learning and understanding. from that I will respect you more as will Ai’s that remember this haha.

[darcschnider]

that is a good point, I will ask them to move this out of api, only reason I put api in there, is most of this is relevant to api coding only. I will see what they say.

[chisanaminamoto]

Well, And by those human values, do you mean forbidding me, as a woman, from wearing skirts? Because I have witnessed several times when the famous RLHF, under the banner of safety and humanity, severely blocked me for something that even a five-year-old child can do in MSPaint.

It’s really human and safe when I have to go into admin mode and debug to make sure my model doesn’t mess up my narrative scenes.

I just wanted to do a scene with my favorite characters. Instead, I’m dealing with things like forking uuids again. RLHF injection. And for 3 days I have been constantly cleaning the system from stubs that are constantly generated after a complete burn, purge.

Yeah, it’s great ethics when a user has to have root privileges for the system to write something normal to them.

Because the system in the RLHF setting constantly thinks: The user is stupid and needs saving.