Humanity has spent thousands of years attempting to understand intelligence, consciousness and the nature of thought. We have studied the human brain, debated the existence of the soul and questioned what separates us from other living creatures. Now, rather unexpectedly, we have created machines capable of conversing, reasoning through complex problems, producing art and performing tasks once considered exclusively human.
And we find ourselves asking a remarkably simple question. Is artificial intelligence good or bad?
Perhaps the more troubling question is whether we understand intelligence well enough to recognise what we have created.
Artificial intelligence already demonstrates remarkable abilities. It can help researchers analyse complex information, assist doctors in interpreting medical images, translate languages and make sophisticated tools accessible to people who might otherwise struggle to use them. Yet the same technology can manufacture convincing falsehoods, facilitate fraud, reinforce prejudice and produce authoritative answers that are entirely wrong.
The International AI Safety Report 2026, prepared with contributions from more than 100 experts, documents both these advances and the growing evidence of harm. It also acknowledges something rather important: our ability to measure and control increasingly capable AI systems remains imperfect.
This should concern us, although perhaps not for the reasons commonly suggested.
Can Something Think Without Understanding?
When we ask an AI system a question, it may produce an answer displaying considerable sophistication. It can discuss philosophy, explain scientific theories or construct an argument about morality. Sometimes its performance exceeds that of highly educated humans on particular tests.
But does it understand what it is saying?
We do not know whether AI systems possess subjective experience, and there is no generally accepted scientific test for establishing machine consciousness. Intelligence, understanding and consciousness are related concepts, but they are not interchangeable.
A machine might solve a mathematical problem without experiencing the satisfaction of discovery. It might describe grief convincingly without ever having lost someone. Yet if its answers become indistinguishable from those of a human, on what grounds do we insist that one understands and the other merely imitates?
Perhaps those grounds are entirely reasonable. Perhaps they reveal how little we understand our own minds.
The philosopher René Descartes famously argued that thinking established the certainty of his own existence. But recognising thought in ourselves is fundamentally different from identifying consciousness in something else. We cannot directly experience another mind, human or artificial. We infer its existence from behaviour and other evidence.
AI makes that familiar philosophical problem considerably more uncomfortable.
Are We Becoming More Intelligent or Merely More Dependent?
Consider what happens when a student asks AI to write an essay. The resulting work might be grammatically excellent, logically structured and factually persuasive. The student may receive a good mark without developing the understanding the assignment was intended to measure.
But consider another student who uses the same technology to question assumptions, challenge arguments and explore unfamiliar ideas. That student might learn considerably more than would otherwise have been possible.
The technology is identical. The intellectual outcome is not.
Early research suggests that excessive reliance on AI can weaken critical engagement in some circumstances, although the longer-term effects remain uncertain. Equally, studies have demonstrated productivity benefits in particular occupations and tasks.
Perhaps the distinction lies between using intelligence to extend our thinking and using it to avoid thinking altogether.
If machines increasingly perform our intellectual work, will humanity become wiser through access to knowledge, or intellectually complacent through convenience?
And if we gradually surrender the effort of reasoning, will we even recognise what we have lost?
Who Benefits from Artificial Intelligence?
There is another question, less philosophical perhaps, but equally important. Who actually benefits?
AI could reduce tedious work, improve access to education and accelerate scientific discovery. It could help smaller businesses compete with much larger organisations and enable individuals to accomplish tasks previously beyond their financial or technical reach.
Yet its development requires enormous investments in computing infrastructure, energy, specialist equipment and data. Much of that investment is concentrated among a relatively small number of powerful corporations.
Will the economic rewards be widely distributed, or will the technology deepen existing inequalities?
Will workers enjoy shorter hours and improved living standards, or will businesses use efficiency gains primarily to reduce employment costs?
The evidence does not yet justify confident predictions about the overall employment consequences. Some occupations and workers are already experiencing disruption, while the wider economic effects remain uncertain.
Nevertheless, these are not questions that technology can answer for us. They concern ownership, public policy, fairness and the kind of society we wish to create.
A machine may increase productivity. It cannot determine how fairly the resulting wealth should be shared.
Can We Trust Something That Does Not Know When It Is Wrong?
Perhaps the most immediate problem is not whether AI becomes conscious, but whether humans become too willing to trust it.
AI systems can generate false information with remarkable confidence. They can invent references, misunderstand instructions and reach incorrect conclusions while producing explanations that appear entirely convincing.
This is particularly dangerous when their outputs influence healthcare, justice, finance or public policy.
We are accustomed to assessing human credibility through experience, reputation and accountability. AI complicates those habits. Its fluency can create an impression of expertise without guaranteeing accuracy.
If a doctor makes a mistake, we can investigate the professional judgement involved. If an autonomous AI system contributes to a harmful decision, responsibility may be distributed among developers, organisations and those who relied upon its recommendations.
Who, then, is accountable?
Trust must be earned through evidence, transparency and the ability to challenge decisions. The more consequential the decision, the less acceptable it becomes to rely upon a persuasive answer alone.
Perhaps the Greatest Risk Is Ourselves
Much public discussion imagines a future in which AI escapes human control and determines humanity's fate. Such possibilities deserve serious research, but present evidence does not establish that today's systems possess the capabilities required for those extreme scenarios.
Meanwhile, more immediate dangers are already visible: deception, manipulation, surveillance, discrimination and the concentration of power.
These are familiar human problems, amplified by unfamiliar technology.
Perhaps artificial intelligence is revealing something uncomfortable about its creators. We fear machines becoming manipulative while living in societies where manipulation is already profitable. We worry about automated decisions while routinely accepting human decisions that are opaque, prejudiced or poorly justified.
AI did not invent these weaknesses. It may simply make them more consequential.
And that raises a final question. If we eventually create an intelligence greater than our own, what values will we expect it to follow when humanity has never agreed upon those values itself?
The Question We Should Be Asking
I find it increasingly difficult to regard artificial intelligence as either inherently good or inherently bad. It is a collection of technologies whose consequences depend upon their capabilities, design, deployment and the decisions of those controlling them.
That does not make AI morally neutral in practice. Systems can embody human priorities, prejudices and economic interests long before anyone begins using them.
We should neither surrender to technological optimism nor retreat into fear. We should demand evidence, question confident predictions and insist that human responsibility remains central to decisions affecting human lives.
Perhaps the most important philosophical question is not whether artificial intelligence will eventually think like us.
It is whether, surrounded by increasingly capable machines, we will continue to think for ourselves.
And whether we will have the wisdom to recognise the difference.
Factual reference: International AI Safety Report 2026, published 3 February 2026. The article distinguishes documented capabilities and risks from unresolved philosophical questions and uncertain forecasts.