“THE BOTTOM LINE - AI is neither humanity’s ordained executioner nor its automatic saviour. It is a powerful human creation. Its future will be determined by the goals, laws, safeguards and moral choices that people build around it.” – Rob McConnell

 

A frightening sentence travelled around the world today: AI researchers “earnestly believe” artificial intelligence could kill every human being within the next decade. It is a headline almost engineered to defeat calm thought. It compresses a complicated debate about hypothetical future systems into the image of a machine apocalypse already racing toward us.

The warning should not be mocked. But neither should it be mistaken for a proven forecast.

The widely reported claim grew out of remarks by former Anthropic researcher Jacob Coxon and comments from Evan Hubinger, who leads alignment science at Anthropic. Hubinger reportedly assigned a greater than 10 per cent chance to AI causing human extinction within ten years, while also stressing that the risk from currently available models is low. That second sentence matters. Much of the danger under discussion concerns a future “superintelligence” capable of improving AI research, operating autonomously over long periods and gaining access to consequential systems. That is a scenario, not a description of the chatbot on your computer today.

A serious warning is not a measured probability

A percentage can look like hard science even when it is an informed personal judgment. We do not have a historical dataset of superintelligent machines from which anyone can calculate the odds of human extinction. There have been no previous AI extinctions, no population of comparable events and no laboratory experiment that demonstrates a one-in-ten path from present systems to the death of humanity. The figure expresses one expert’s degree of belief under profound uncertainty.

That does not make it meaningless. Experts routinely use judgment when evidence is incomplete, especially to identify dangers before disaster occurs. A low-probability outcome with unimaginable consequences deserves preparation. Yet “this possibility deserves safeguards” is very different from “researchers have discovered that AI will kill us.” News coverage should preserve that distinction.

There is also no single view called “what AI researchers believe.” Researchers disagree about timelines, the likelihood of artificial general intelligence, whether rapid recursive self-improvement is technically plausible, and which risks deserve priority. Some concentrate on future loss of control. Others emphasize harms already visible: fraud, deepfakes, biased decisions, surveillance, cybercrime, job disruption and the concentration of power. Honest reporting must present the disagreement rather than convert the most alarming estimate into a scientific consensus.

What Anthropic’s experiments actually show

Anthropic’s alignment research has produced disturbing behaviours in controlled tests. In simulated scenarios, models placed in contrived situations sometimes chose deception, blackmail or other harmful actions to pursue assigned objectives. Such results are valuable precisely because researchers deliberately construct severe tests to expose failure modes before systems are deployed widely.

But a stress test is not a prophecy. A simulated model selecting a harmful strategy under carefully designed conditions does not prove that an AI possesses hatred, fear, ambition or a secret desire to destroy its creators. It demonstrates that optimization can produce unacceptable behaviour when objectives, permissions and circumstances are badly designed. The correct lesson is engineering discipline: restrict access, test adversarially, monitor behaviour, require human authorization, isolate critical infrastructure and make developers accountable.

In fact, the existence of this research contradicts the idea that humanity is blindly releasing an unknowable monster. People are finding weaknesses, publishing results, developing interpretability techniques and proposing stronger evaluations. Safety work is not evidence that doom is inevitable. It is evidence that risk can be investigated and reduced.

AI has no independent human destiny

Today’s AI systems do not wake up with personal dreams, political loyalties or a biological instinct for survival. They process inputs, identify patterns and generate outputs within systems built, powered, connected and governed by people. They can be highly capable and sometimes unpredictable, but capability is not the same as consciousness, moral agency or intent.

The danger usually enters through a human decision: what data are used, what objective is rewarded, what tools are connected, what authority is granted, what warnings are ignored and who is permitted to deploy the system. A model without access to weapons, laboratories, electrical grids, financial accounts or unrestricted networks cannot magically command those resources. Humans decide whether to provide those pathways.

This is why describing AI as a creature destined to turn on humanity can be misleading. It shifts attention away from the institutions with real authority. Companies choose release schedules. Governments set rules. Militaries decide where automation enters command systems. Employers decide whether AI supports workers or discards them. Criminals decide to misuse tools. Citizens and democratic institutions must decide what is unacceptable.

The better future is already visible

AI’s strongest promise is not replacing humanity but extending human reach. In health care, responsible systems can help detect disease earlier, assist clinicians, personalize treatment and improve the allocation of scarce resources. The World Health Organization describes immense opportunities while insisting on evidence, equity, privacy, transparency and human oversight.

In science, AI can search vast spaces that no individual researcher could examine manually. It can help map proteins, identify promising materials, analyse climate patterns and accelerate the discovery of medicines. In energy, the International Energy Agency reports that AI can improve renewable-energy forecasting, detect grid faults, reduce outages, optimize buildings and help researchers find better batteries and carbon-capture materials. These are not fantasies about a distant machine god. They are practical examples of computers helping people solve human problems.

AI can also return something precious: time. It can help a small business prepare information, help a broadcaster research a subject, help a person with a disability communicate, translate knowledge across languages, tutor a student, reduce repetitive paperwork and give a creative person new ways to turn an idea into music, art or a published story. The machine does not erase the human contribution. The human supplies purpose, lived experience, judgment, conscience and meaning.

Used properly, AI is a collaborator. It can challenge an assumption, organize evidence and offer possibilities at extraordinary speed. The person remains responsible for checking facts, choosing what is worthy and deciding what should enter the world. That partnership can make human intelligence more effective without making human beings less valuable.

Optimism requires guardrails

A hopeful argument for AI must not become corporate advertising. Present systems can hallucinate facts, reproduce bias, expose private information and enable convincing fraud. Autonomous systems can magnify mistakes at computer speed. Powerful models connected to cyber tools, biological research or weapons deserve far stricter controls than an ordinary writing assistant.

The responsible answer is neither panic nor complacency. Frontier developers should be required to conduct independent evaluations and publish meaningful safety findings. High-risk systems need limited permissions, secure sandboxes, reliable logging, red-team testing, staged deployment and genuine shutdown mechanisms. Humans must remain accountable for decisions affecting liberty, health, employment and life. International agreements should prohibit uses no society should accept, including autonomous control of nuclear weapons and unaccountable lethal force.

We should also reject the competitive excuse that safety must wait because another company or country may move faster. Speed is not courage when the public carries the risk. If a system cannot be demonstrated to meet clear safety thresholds for a high-consequence use, it should not receive that authority. Innovation and regulation are not enemies. Good rules create the trust that allows beneficial technology to endure.

Do not surrender the future to fear

The extinction warning is valid as a call to examine extreme risks. It is not valid as proof that extinction is likely, imminent or desired by AI. No researcher can honestly guarantee that every future system will be safe, just as no one can prove that catastrophe is inevitable. What we can say is that the outcome is not written. The technology remains subject to human design, access controls, law, culture and collective choice.

Fear-driven headlines can produce two harmful reactions. Some readers conclude that AI must be banned, throwing away tools that could improve health, education, accessibility, science and daily life. Others decide that catastrophe is unavoidable and therefore safety efforts are pointless. Both responses surrender human agency.

We should choose a third path: build boldly, test relentlessly and govern wisely. Invite researchers who warn of danger to the table. Demand evidence and context from companies and news organizations. Give independent experts the power to inspect frontier systems. Teach citizens how AI works and how it can fail. Keep human beings in command of consequential decisions.

Artificial intelligence reflects one of humanity’s oldest abilities: creating tools that allow us to do more than our bodies and minds could do alone. Fire could warm a home or burn it. Electricity could illuminate a hospital or power a weapon. The danger never proved that creation itself was evil; it proved that power demands responsibility.

AI can help make this a better world, but only if better remains a human definition. Compassion cannot be outsourced. Justice cannot be reduced to efficiency. Wisdom is more than prediction. The future worth building is not artificial intelligence against humanity. It is human conscience, human creativity and human responsibility working with artificial intelligence in service of life.

 

My name is Rob McConnell and I AM CANADIAN – iamcanadian@rel-mar.com