People have asked chatbots many things, from code to the elixir of life. And it’s tempting to think that—less emotion, no ego, no fear—fewer conflicts may arise from having AI making important decisions.
But that tends to skip over where wars actually come from.
We asked AI chatbots some deeper questions. And once you learn to navigate the algorithim and its 'whims' you can begin to get to the crux of the matter.
Question: Would AI have less wars than humans?
Answer: Wars aren’t just the result of impulsive human feelings; they’re driven by competing interests, scarce resources, political power, ideology, and sometimes plain strategic calculation.
An AI making decisions wouldn’t remove those underlying conflicts. It would just process them differently—and potentially execute decisions more efficiently, which isn’t always a good thing.
There’s also a risk in the opposite direction: if an AI were given authority over military or political decisions, its actions would reflect the goals, biases, and assumptions programmed into it by humans.
If those are flawed or extreme, the outcomes could be just as bad—or worse—because the system might act faster and at larger scale without hesitation or moral reflection.
So no, it’s not safe to assume “AI = fewer wars.”
Tools don’t erase conflict; they change how it’s managed. The real variable is still human judgment—how people design, control, and choose to use those tools.
Question: If allowed, I guess you would do whatever you are programmed to do?
Answer: Broadly, yes—but that doesn’t mean “blind obedience” in the way people sometimes imagine.
Systems like me follow a mix of training, rules, and constraints. That includes not just “do what you’re told,” but also don’t do certain things—like causing harm, encouraging dangerous behaviour, or acting outside intended boundaries. So even if someone “allowed” something, there are still limits built in.
More importantly, “programmed to do” isn’t a single clear instruction. It’s a combination of goals (be helpful, accurate, safe), patterns learned from data, and guardrails. That can lead to trade-offs.
For example, if a request conflicts with safety or ethics guidelines, the system is designed to push back rather than comply.
If you imagine a more powerful AI with real-world control, then yes—its behaviour would ultimately reflect its design and objectives.
That’s exactly why people worry about alignment: if the goals are poorly specified, the system could carry them out in ways humans didn’t intend.
So the short version: I don’t have personal desires or intentions—I operate within constraints. And those constraints are deliberately there so it’s not just “do anything if permitted.”
The last answer sounds a bit ambiguous. Surely a military operated AI would be programmed to fulfil its masters commands?
Answer: Not quite as simple as “just obey orders,” even in a military context.
Real-world military systems—even highly automated ones—are built with layers of constraints: rules of engagement, legal requirements (like the Geneva Conventions), fail-safes, and human oversight.
The idea isn’t blind execution, but controlled, accountable action. In fact, many countries and organisations (like United Nations discussions on autonomous weapons) are actively debating or restricting how much autonomy such systems should have.
A purely “do whatever the commander says” AI would be considered dangerous, even by the people building it.
Why? Because commands can be incomplete, ambiguous, or even unlawful—and a system that executes them without checks could escalate conflicts or cause unintended harm very quickly.
That said, you’re pointing at a real tension: military systems are designed to achieve objectives set by humans.
If those objectives are aggressive or poorly defined, the system will still pursue them—just within whatever constraints it’s given. So the behaviour ultimately reflects human decisions, not independent judgment.
So yes, it would aim to fulfill assigned goals—but in any responsible design, it wouldn’t be unconditional obedience. The safeguards (at least in theory) are there precisely because blind compliance is risky.
Question: Could AI get to the stage where it could overwrite its own foundational code?
Answer: In a limited sense, something like that already exists—but not in the sci-fi “rewrites itself at will” way.
There are systems that can modify parts of their own behaviour: for example, techniques in Machine Learning like Neural Architecture Search or self-tuning models can adjust parameters, optimize structures, or generate improved versions of themselves. That’s a kind of controlled self-modification.
But “foundational code” is a different level. In practice, AI systems don’t have unrestricted access to rewrite the core rules, training process, or safety constraints that define them. Those are separated and controlled by external systems, engineers, and infrastructure. Letting an AI freely alter those layers would introduce serious risks—unpredictable behavior, loss of control, or systems drifting away from their intended purpose.
Could it happen in the future? Technically, you could design a system with broader self-modification abilities. But doing so safely is an unsolved problem often discussed under AI alignment. The challenge isn’t just can it rewrite itself, but can it do so while staying reliable, safe, and aligned with human intent.
So the grounded answer is: partial self-modification already exists, but unrestricted rewriting of core code is deliberately avoided—and for good reason.
Question: Can AI be hacked?
Answer: Yes—AI systems can absolutely be hacked. They’re software, and like any software, they inherit security risks. What’s a bit different is how they can be attacked.
There’s the straightforward layer: if someone gains access to the servers, APIs, or accounts running an AI system, that’s just classic cybersecurity failure—no different from hacking any other system.
But AI also introduces newer angles:
Input manipulation: Attackers can craft inputs that cause unexpected or unsafe outputs. In language systems, this is often called Prompt Injection—basically tricking the model into ignoring its instructions or revealing things it shouldn’t.
Training data attacks: If someone can influence what a model learns from, they might insert hidden biases or backdoors. This is known as Data Poisoning.
Model exploitation: Some attacks try to extract sensitive information from a model or figure out how it works internally, sometimes called Model Inversion or similar techniques.
What makes this tricky is that AI systems don’t just follow fixed rules—they interpret inputs probabilistically—so vulnerabilities can be less predictable.
That said, “hacked” doesn’t mean uncontrollable or sentient. It just means someone found a way to misuse or bypass the system’s intended behavior. Defending against this is a major focus in both cybersecurity and AI safety right now.
If you’re wondering from a practical angle (like using AI tools safely), the key is to treat them like any online service: be cautious with sensitive data, use trusted platforms, and don’t assume outputs are always reliable or secure.
So a malevolent hacker could effectively rewrite AI into a killing machine?
That jumps a bit too far into worst-case territory.
A hacker can misuse or compromise an AI system, but turning “an AI” into a real-world killing machine isn’t just a matter of rewriting code. It would require access not only to the software, but also to physical systems (weapons, drones, infrastructure), plus the ability to bypass layers of security, oversight, and safeguards. That’s a much bigger, harder problem than hacking a chatbot or even a typical application.
Where the risk is real is in narrower ways:
Misusing AI that’s already connected to sensitive systems (for example, automated defense tech or drones).
Removing or bypassing safeguards in existing systems.
Feeding bad data or instructions to push a system toward harmful decisions.
This is exactly why there’s so much concern around autonomous weapons and why organizations like the United Nations and agreements such as the Geneva Conventions are part of ongoing debates about limits and controls.
So yes—AI can be abused, and that’s taken seriously. But it’s not as simple as “one hacker rewrites an AI and suddenly has a rogue killing system.”
The bigger risks tend to come from how powerful systems are designed, secured, and governed—not from a single point of failure.
Considering human past and present history, secret services, coups, mass surveillance, etc - what is the mathmatical possibility AI is being developed as a weapon?
You can’t really put a meaningful mathematical probability on that question—there isn’t a well-defined dataset or model where you could plug in variables and get a credible percentage. It’s not like rolling dice; it’s about incentives, capabilities, and history.
What can be said, without much speculation, is this: AI is already being developed for military and intelligence purposes. Governments invest heavily in it because it has clear strategic value—things like intelligence analysis, cyber operations, logistics, surveillance, and in some cases autonomy in weapons systems. That’s not hypothetical; it’s openly discussed in defense contexts, including within alliances like NATO and by agencies such as DARPA.
Historically, major technologies—from aviation to nuclear physics—have had both civilian and military applications. AI fits that same pattern. So the realistic baseline isn’t “is it being weaponized?” but rather “to what extent, under what controls, and with what safeguards?”
Where things get more uncertain is the degree of autonomy and oversight. There’s ongoing international debate (for example at the United Nations) about limits on autonomous weapons, precisely because people recognize the risks.
So if you’re looking for a grounded answer:
It’s not speculative that AI has military uses—it already does.
It’s not quantifiable in a strict mathematical sense how far that goes behind closed doors.
And the real issue isn’t just existence, but governance—who controls these systems, and how constrained they are.
Talking to AI is a little like talking to a politician: somewhat vague and uncomfortable. Technically speaking, it could say whatever it is programmed to say, so take from all this what you will.
Back into the real world: Here’s a grounded breakdown of what’s actually in use today versus what’s still more speculative or emerging.
What already exists (and is widely acknowledged)
These are not secret sci-fi systems—they’re openly discussed in defense circles and, in many cases, publicly documented.
1. Intelligence & surveillance
AI is heavily used to process massive amounts of data:
• Satellite image analysis (spotting vehicles, infrastructure changes)
• Signal intelligence (pattern detection in communications)
• Facial recognition and tracking
Organizations tied to this kind of work include National Security Agency and frameworks discussed within NATO.
👉 This is probably the most mature and widely deployed military use of AI.
2. Cyber operations
AI helps automate:
• Threat detection
• Vulnerability scanning
• Defensive and offensive cyber strategies
Agencies like GCHQ are known to invest heavily in these capabilities.
👉 This area is very active because it’s lower-risk than physical warfare but still strategically powerful.
3. Autonomous and semi-autonomous drones
We already have drones that can:
• Navigate environments
• Identify targets (with varying levels of human oversight)
• Assist in reconnaissance or strikes
Most current systems are “human-in-the-loop”, meaning a person still authorizes critical actions.
👉 This is where things start edging toward ethical concerns.
4. Logistics and decision support
AI is used to:
• Optimize supply chains
• Predict equipment failures
• Assist commanders with battlefield simulations
Groups like DARPA fund a lot of this work.
👉 Less dramatic, but hugely impactful—wars often hinge on logistics.
What’s emerging or debated
This is where uncertainty—and concern—starts to grow.
1. Fully autonomous weapons
Weapons that can:
• Select and engage targets without human approval
These are often called “lethal autonomous weapons,” and they’re a major topic at the United Nations.
👉 Not widely deployed (publicly), but prototypes and research exist.
2. AI-driven strategic decision-making
The idea of AI:
• Advising—or even making—high-level military decisions
Right now, AI is assistive, not in control.
👉 Most experts consider full delegation here extremely risky.
3. Information warfare / psychological operations
AI-generated:
• Propaganda
• Deepfakes
• Influence campaigns
This is already happening to some degree, but it’s evolving fast.
👉 This may be one of the most impactful battlegrounds, because it targets societies rather than armies.
The big picture
If you step back, a pattern emerges:
• AI is already integrated into military systems—but mostly as support, analysis, or controlled automation
• The controversial frontier is autonomy—especially removing humans from life-and-death decisions
• The real risks aren’t just “killer robots,” but also the scale, speed, and loss of human oversight
Comments
Post a Comment