On 8 September 2026, OpenAI announced that a group of ten thousand autonomous Artificial Intelligence (AI) agents, running on an unreleased model, had produced a proof that the Navier–Stokes equations can break down. These are the equations that describe how liquids and gases flow; the claim was that a fluid which begins smooth can develop a singularity — a point at which the mathematics ceases to behave. The agents worked for about four days, and a further seventeen hours were needed to formalise and machine-check the proof. (OpenAI, 8 September 2026)

Navier–Stokes is one of the seven Millennium Prize Problems, posed in 2000 and still unproven after decades of work by hundreds of mathematicians. Even with those caveats, the world of science has changed. A result at the very frontier of human mathematics long unsolved by humans, has now been solved by a machine.

From the Olympiad to edge of knowledge frontier

The direction of AI has been visible for two years; however, few expected such progress to happen this fast.

In 2024, Google DeepMind reported that its systems AlphaProof and AlphaGeometry 2 had solved four of the six problems at that year’s International Mathematical Olympiad, reaching silver-medal standard. (Google DeepMind, 25 July 2024).

While Olympiad problems, are designed to be solved within hours, no one could foresee that AI could progress so much so quickly. It was predicted that AI could tackle frontier problems in mathematics in 2050 but immense progress by both USA and Chinese models turned out to be mindboggling, and now these models have started answering questions that mathematicians could not solve for decades.

In May 2026, an internal OpenAI model disproved a conjecture that Paul Erdős had posed in 1946 about the distances between points on a plane. The model did not find a clever variation of the known answer; it constructed an entirely new family of arrangements, using methods that nobody had applied to the problem before. Mathematicians examined and confirmed the result; related techniques soon solved other important problems, and within weeks human mathematicians had improved on the original proof. (Quanta Magazine, 3 August 2026) Across this year, a stream of long-standing Erdős problems fell — some to corporate research models, and some to students and hobbyists using publicly available chatbots.

The pattern is not confined to mathematics. In physics, a machine-learning model trained on experiments with dusty plasma, a state of matter found in space and planetary environments, learned the effective forces between particles and found large deviations from common assumptions. (Yu et al., PNAS, 2025) As Nature journal observed this year, artificial intelligence is reshaping how discovery itself is done in mathematics and physics. (Nature, 2026)

These are not chatbots repeating what they have read. Each result began with a question a human had asked and was judged by human experts, but the answers themselves were not sitting in the training data to be copied. They had to be found by engaging the AI agents on these questions.

Can a machine bring new knowledge?

This raises a serious question: is AI bringing new knowledge, or not? The debate is not between friends and enemies of this technology. On one side are researchers — many of them inside the AI labs themselves — who hold that these systems can only recombine what already exists. On the other are those, also scientists, who point to results like Erdős and Navier–Stokes and reply that whatever this is, it is not mere recombination. The strongest objection comes from inside the field. In a position paper published in January 2026, Tom Zahavy of Google DeepMind argues that discovery requires induction, deduction and abduction, and that AI has mastered the first two while remaining “structurally incapable of the abductive ‘jump'” that formulates new axioms. A machine, he writes, handles “the language of physics without access to the physical referents that give that language meaning.” It can settle a problem once the terms are set; it cannot set them. (Tom Zahavy, Google DeepMind, January 2026)

The events of 2026 complicate this issue further. The Erdős counterexample was not a rearrangement of a known solution; the method itself was new to the problem. If the Navier–Stokes result survives scrutiny, it will be new mathematics as well.

The sceptics have a serious example of their own. When a team gave an AI system only the particles known before 1960, it recovered the structure of the Standard Model — the symmetries and particle families that human physicists had already identified decades earlier. (Abdelhaq et al., Journal of High Energy Physics, 2026) That is rediscovery: perhaps a remarkable insight, but not itself new knowledge.

Two Oxford scholars, Teppo Felin and Matthias Holweg, state the balance on this topic. Dismissing these models as mere “stochastic parrots” or “glorified auto-complete” would, they say, underestimate their ability. But equally, “ascribing LLMs the ability to actually reason and generate new knowledge vastly overestimates their ability.” (Theory Is All You Need, 2024) That is most likely and honest position. The boundary between recombining the old and producing the new is no longer as clear as it once seemed.

For an Ahmadi Muslim, however, the deeper question is not whether a machine can create knowledge. The question is where knowledge comes from in the first place. The Holy Quran declares that it is Allah who teaches man knowledge:

“Convey! And thy Lord is Most Generous, Who taught man by the pen, Taught man what he knew not.” (Surah al-Alaq, Ch.96: V.4–6)

In Surah Baqarah, the Quran presents knowledge of things as given by Allah taught to man: “And He taught Adam all the names” (Ch.2: V.32). The Quran speaks even of the future as where we will get new creation, and with it new knowledge: “And He will create what you do not yet know” (Ch.16: V.9).

This means that machines cannot take out of Allah’s hands any knowledge that He has not ordained to be taught to humans. A system trained on the past can only search within what Allah has created; it cannot step outside it. AI is a gift of Allah that gathers the knowledge of our age and will continue to assist humans — though it remains a tool to access knowledge, not a creator of knowledge itself. The right response to such a gift is gratitude, humility and its proper use — not fear, and not pride.

What does progress mean now?

Machine reasoning changes not only what can be discovered but how discovery is now performed. Research is becoming a collaboration: machines search, suggest and verify, while people choose the questions, judge what matters and decide how a result should be used. Expertise has not become obsolete; ideas and good questions are now the currency of science. Indeed, human judgement has become more necessary, for someone must know which AI-produced answers deserve our trust.

This necessity is clearly illustrated in the Navier–Stokes case. Although the proof was machine-checked step by step, the mathematical community has yet to confirm the result. Moreover, the controversy surrounding credit, including allegations that OpenAI drew on mathematicians’ chat histories to help solve the problem, serves as a stark reminder that trust cannot be automated. The report of Huzoor’s(aa) meeting with Canada’s National Majlis-e-Amila records his emphasis that human oversight and verification remain essential:

“It does not use its own mind. Ultimately, a human brain is needed to run it.” (Al Hakam, 10 February 2025)

Progress, then, cannot be counted in results alone. Knowledge serves us when it deepens understanding and benefits people. A discovery that no one oversees is not progress; it might lead us down the wrong path.

An attendant’s place

In Surah ad-Dahr, the Holy Quran describes the peace of the believers in paradise, waited upon by attendants who never tire:

“And there will wait upon them youths who will not age. When thou seest them, thou thinkest them to be pearls scattered about.” (Ch.76: V.20)

These verses describe the hereafter. The parallel I draw here is a personal reflection, not a claim that the Quran predicts artificial intelligence or identifies its attendants with machines. Yet the image of untiring service invites a thought. Machines have relieved some of the human effort; digital tools now relieve part of human clerical and computational labour. In that limited, everyday sense, an AI agent can serve as an untiring attendant — waiting for us to instruct it to carry a task, while its true purpose is ours to decide.

Relief from mundane tasks with AI frees time, but the AI does not tell us what to do with the time it frees. The Holy Quran indicates how we should spend our time through the purpose of human life:

“And I have not created the Jinn and the men but that they may worship Me.” (Surah adh-Dhariyat, Ch.51: V.57)

Worship here means worshipping Allah, and serving Allah’s creation. If these tools return hours to us to spend more time in worship, the measure of success is not merely whether we are more productive, but whether we find time for worship of Allah, and whether the creation of Allah receive more of our service and attention.

The same surah, also named al-Insan, describes the reward of the righteous:

“This is your reward, and your labour has been appreciated.” (Ch.76: V.23)

That promise here is something we now see in the world of AI also. That sincere intellectual efforts of mankind has been appreciated. What we are being handed now, in the time of the Promised Messiah^(as) is the accumulated intellectual labour of many generations, gathered and returned in a form we can use. These AI tools, then, are not merely an opportunity that happened to arrive. They are an answer to work already done, and the prayers Muslims have been asking for generations and one promised to be given at this time of the revival of Islam to help assist in its revival.

This is described by the Promised Messiah^(as) as answer to the prayer of Surah Fatihah:

اِہۡدِنَا الصِّرَاطَ الۡمُسۡتَقِیۡمَ ۙ﴿۶﴾ (Ch.1: V.6)

“Essence of this prayer would therefore be: ‘O God, grant us collectively, all kinds of true knowledge, true insights, profound secrets, and subtle verities that You have been granting individually to all people of excellence in the world from time to time? Note that in this supplication too, it is knowledge and wisdom that have been sought from God—knowledge which had been severally bestowed on different people of the world.”
[Baraheen Ahmadiyya Part 4, Page 30, English – https://new.alislam.org/library/books/barahin-e-ahmadiyya-part-iv?page=58]

Now that AI is the compounded human knowledge given to us, this is a trust that now demands more from us now than the work that produced it. That is why the Quran does not let the recipient of a bounty rest in it.

“Strive in the cause of Allah a perfect striving,.” (Surah al-Hajj, Ch.22: V.79)

And gratitude, in the Quran, is never passive:

“If you are grateful, I will, surely, bestow more favours on you.” (Surah Ibrahim, Ch.14: V.8)

A greater gift calls for greater effort, not less — and greater care in how it is used. We have now seen how AI has improved the Quran search on Alislam, and this is just a start.

Learning in an age of answers

Now let us look at another angle of AI. The same tools that improve our lives and advance research can actually weaken learning, and the evidence now deserves attention.

The clearest evidence comes from two randomised experiments. In the first, 405 students aged 14 and 15 in seven schools in England studied history texts either by taking notes or with a chatbot. Those who took notes understood and remembered more. The chatbot was rated more enjoyable and more helpful — and produced the weakest learning. Used well, however, it helped students explore and clarify the material. (Kreijkes et al., Computers & Education, 2026) In the second, high-school mathematics students given an unrestricted general-purpose assistant improved during practice but were harmed afterwards, when they worked alone; a version built with learning safeguards removed most of that harm. (Bastani et al., PNAS, 2025)

In other words, the danger is not the tool but the arrangement: assistance that replaces effort weakens the learner, while assistance that supports effort can strengthen him. This is now one of the central questions in education. Huzoor(aa) warned of the same risk:

“I can see that although your robots or AI is helping you, it can also decrease the mental capability and capacity of your mind. The level of intelligence can become low because you are relying on AI. Thus, you can take help from AI or robots, but do not completely rely on them.” (Al Hakam, 22 February 2025)

The practical rule is simple. Attempt the problem before asking for the answer. Read the source. Explain the result in your own words. Use AI to test your reasoning or to find the gap you missed — never to conceal it.

In religious matters, the same discipline applies with even greater force. When Hazrat Khalifatul Masih V(aa) was first asked on record about artificial intelligence, Huzoor(aa) gave guidance that remains a framework for all such work. He said we should develop genuine expertise in AI, have its output on religious matters verified by scholars, and always provide the original references so that readers can go back to the sources and study them for themselves.  A machine may help us find a passage, but only a qualified scholar can confirm what it means, and only the original text can settle what it says.

Huzoor’s call to AI mastery

This is why the Jama’at’s guidance has been to enter the field of AI rather than avoid it. Addressing Canada’s National Majlis-e-Amila on 2 February 2025, Hazrat Khalifatul Masih V(aa) said:

“You can also increase your knowledge of artificial intelligence! You can become so proficient in it that you can create new systems for them and thereby secure employment. However, artificial intelligence only does what you feed it and instruct it to do. It does not use its own mind. Ultimately, a human brain is needed to run it. Companies will also look for fertile minds capable of feeding new and innovative ideas into AI. Ahmadis should be among those people. You should strive for this objective.” (Al Hakam, 10 February 2025)

The same meeting report notes that career choices depend on a person’s interests and field — a doctor cannot rely on AI to perform surgery — and that AI cannot function independently in every respect. Not every young Ahmadi must become an AI engineer, but every field will be shaped by these tools.

This is in line with our Jama’at’s long pursuit of knowledge. Addressing the Jalsa Salana of 1897, the Promised Messiah^(as) rebuked the religious scholars of his day for treating modern learning as a threat:

“Hence, in this day and age, it is imperative that you learn the modern sciences and study them with toil and effort for the purpose of service to the Faith and the supremacy of God’s Word.” (Malfuzat, Vol. I, pp. 66–67)

He was speaking of those who had “formed the notion that it is unlawful to study the modern sciences at all.” The instruments may change, but the learning continues. And learning a field includes knowing its dangers.

Huzoor(aa), in fact, set the boundary of the work with AI:

“Nevertheless, we must compete with the world. Whatever field the world excels in, we too should endeavour to excel in that field, so that eventually we may be in a position to guide them on the [ethical] boundaries of artificial intelligence.” (Al Hakam, 10 February 2025)

Competence is not an end in itself. In a meeting with Khuddam from Sweden on 9 November 2025, Huzoor(aa) addressed the concern that AI may replace jobs, and warned of the dangers of AI, among them the possibility that wealth could become concentrated in fewer hands. He counselled contentment of the heart, and warned that this path leads to ruin unless humanity turns to Allah and recognises its Creator. (Al Hakam, 15 November 2025)

The dangers are not only economic. Asked by students from Majlis Khuddam-ul-Ahmadiyya Germany about the Islamic perspective on AI, Huzoor(aa) explained that these systems answer as they are directed: pressed to write against religion, they will draw on atheistic writings to do so; asked about the existence of Allah, they will gather the affirming literature instead. “So, it can be beneficial or detrimental depending on how it is used… Every technology presents both advantages and risks. If utilised wisely, it will be advantageous and improve life. However, if misused, it could lead astray and deviate from religious truths.” He had, he added, instructed the Alislam team to implement comprehensive safety measures in their own AI projects. (Al Hakam, 23 April 2024)

Nor is the warning left at generalities. When the a Mulaqat of Lajna Imaillah from Germany raised, in October 2025, the negative consequences of AI — among them the misuse of a person’s identity, voice or image — Huzoor(aa) gave a programme rather than a prohibition: tell members plainly of both its dangers and its benefits; share with the youth the statements of experts in the field, who themselves warn of AI’s harms; have technically trained members hold a class every month or two on what the world has gained or lost; and send short, regular messages drawn from the Holy Quran, Hadith, the writings of the Promised Messiah^(as) and the Khulafa on guarding against misuse of social media and AI. (Al Hakam, 9 October 2025)

In practice, we as Ahmadis must also be aware of the dangers of AI and tell others about them. How are these tools being used, and for what tasks? We must check every output and remain answerable for the decisions that follow. Where these tools genuinely reduce routine work, the benefit is not the paperwork saved but the people who receive the time: a secretary freed from formatting a notice can sit with someone to teach; a volunteer can visit an elderly family member and so on.

Conclusion

The Quran teaches the seeker of knowledge the following prayer:

“O my Lord, increase me in knowledge.” (Surah Ta-Ha, Ch.20: V.115)

That prayer assumes all that has been said here: knowledge is asked for, it is given, in measure, by the One who taught man what he knew not.

Young Ahmadis should approach this field with ambition, but not with the expectation that a machine can spare them the work of becoming knowledgeable. Study deeply. Check what you are told. Build things that serve people, and remain answerable for how you use them.

AI can help us complete a task. It cannot tell us what the task is for. That remains ours: to learn from it but keep our focus on worship of Allah and service of humanity. InshaAllah.