On August 20, 2026, the G20 Interfaith Forum and the International Academy for Multicultural Cooperation held the eighth and final webinar in their AI and Faith Webinar Series, entitled “AI and the Future.” Speakers included Manisha Jain, who has spent more than a decade in generative AI at Google, Meta, and Microsoft, including work on Gmail SmartReply and Google Translate; Joshua Tan, who leads technology at Current AI and Public AI, runs Stanford AO at Stanford OpenLab, and co-founded Metagov; Clara Langevin, Senior Manager for AI Policy on the Emerging Technologies team at the Federation of American Scientists and former head of the AI and Machine Learning Platform at the Centre for the Fourth Industrial Revolution Brazil; Dr. David A. Wood, the Glenn D. Ardis Professor of Accounting at Brigham Young University and author of Rewiring Your Mind for AI; and Dr. Valerie Hudson, University Distinguished Professor and George H.W. Bush Chair in the Department of International Affairs at Texas A&M University’s Bush School of Government and Public Service. Dr. Marianna Richardson, Director of Communications for the G20 Interfaith Forum, moderated the discussion.
Manisha Jain
Jain’s involvement with AI dates to her undergraduate studies in the 1990s, when there was neither enough computing power nor enough data to do much beyond theory. Between 2022 and 2024, architectures such as transformers and diffusion models carried AI from simple text generation to true multimodal systems handling text, high-resolution images, speech, and video. Over the past year and a half, she said, the focus has shifted again toward system orchestration and agentic reasoning: environments that solve long-running, multi-step problems, execute autonomous workflows, and make real-world decisions with minimal human intervention.
“AI is moving from being a tool we consult to an agent that acts.”
Every major technological leap has rewritten how humanity lives and works, she said. The Industrial Revolution scaled human muscle, the railways reorganized global trade, and the internet placed the world’s collective knowledge at our fingertips. AI carries the same promise and more, with the potential to accelerate scientific discovery and help cure devastating diseases. But every revolution has also brought deep destruction, from the canal boat operators whose livelihoods vanished with the railways to the local libraries displaced by the internet.
Because the risk of AI grows alongside its capabilities, Jain argued that today’s landscape resembles the nuclear era more closely than the internet. Nuclear technology promised unlimited clean power while introducing the immediate threat of existential destruction, and humanity survived because the international community chose to act, drawing hard lines through the Non-Proliferation Treaty, inspection regimes, and crisis communication channels. As autonomous systems acquire high-stakes operational control over infrastructure, medicine, and security, she said, proactive treaties, shared safety protocols, and binding global guardrails are needed now.
Joshua Tan
Tan, a mathematician and computer scientist, began with what he called the most San Francisco thing he does: work on autonomous organizations, or what people there have started calling zero-person companies. The billion-dollar version of that idea he treated as a red herring. Far more interesting, he suggested, are restaurants or bakeries with no managers, owned and operated by the staff themselves, where standard business logic is run by agents. He was candid that agents remain quite bad at this—AIs, he noted wryly, make very nice bosses who grant every raise you ask for.
Wearing his other hat, Tan turned to the question of ownership. AI today belongs to a relatively small set of private corporations, most of them based in San Francisco. He asked the audience to imagine the alternative.
“I want you to imagine a world in which AI is a form of public infrastructure, like highways, or water, or electricity, or libraries, or the internet itself.”
That is the idea of public AI, which he framed as a third way between opaque private labs and the unrestricted release of frontier-class models. It is already reality in places: he described work with the Swiss National AI Initiative to deploy a genuinely public national model, and a forthcoming South Korean program that will offer a national model free to every person in the country. He also described a Swiss national dialogue that channels community feedback in real time into a live AI instance, letting it be reconfigured to reflect local needs, culture, values, and language. In the United States, he added, conversations with leading labs suggest nationalization of some form may be close to inevitable.
Clara Langevin
Langevin introduced the Federation of American Scientists, founded eighty years ago by Manhattan Project scientists concerned about the existential risk of nuclear weapons. New capabilities arrive almost every month, she said, but rather than track them her team asks which concrete policy levers exist and which policy windows remain open even amid political stalemate. Alongside her own memos, she practices what FAS calls policy entrepreneurship: partnering with technical experts, affected communities, and early-career practitioners to help them identify the specific agencies, authorities, and levers—procurement processes, benchmarks, standards bodies—that could deliver the outcome they are after.
She traced that approach to her earlier work in Brazil, where in 2018 she studied how the judicial branch should adopt AI and watched state courts move at breakneck speed without asking what governance the technology required. She later ran pilots with the São Paulo subway system and a network of public hospitals, testing algorithmic impact assessments inside public procurement, and used that evidence to shape risk measures in Brazil’s National AI Plan.
Speaking personally rather than for her organization, Langevin raised the position of countries that import frontier AI without the capacity to build it, describing the Global South as caught between the United States and China.
“What is our position as countries that import these technologies, that don’t have the capacity or the compute to develop frontier models on our own, but still will suffer from a lot of the risks of this technology?”
Dr. David A. Wood
Wood described himself as the newest arrival to AI among the panelists. Two weeks after ChatGPT was released, a colleague in a BYU hallway mentioned that students had started using it. He began feeding his own exam questions into the system, watched it answer the first three correctly, and decided he had to study it seriously, eventually organizing three hundred faculty worldwide to submit twenty-eight thousand questions.
As a contribution to an interfaith conversation, he turned to a statement made by a leader of his faith during the nuclear debates of the late 1960s, which framed such discoveries as carrying immense power for either blessing or destruction and placing the most gigantic responsibility ever given into human hands, in an age of limitless perils alongside untold possibilities. Another leader of the same faith reiterated it with reference to the AI era within the past two years. Wood drew three elements from it: the perils dominate the airtime, since negative news travels furthest; the possibilities receive far less attention but are equally real; and the responsibility rests squarely with humanity.
“We are going to run the biggest social science experiment in the history of the world: do we respond to this incredible power in a positive or negative way?”
That experiment, he said, will reveal something fundamental about whether people want to be good—which is why he wakes up each day asking how this power might be used to transform life for the better.
Dr. Valerie Hudson
Hudson described her stance as AI gloomerism. Certain uses of AI can help humankind, she said, but she does not believe the race can be stopped, given the immense levels of greed and fear involved, and the intellectual, emotional, spiritual, and environmental costs will be felt for generations. As a Christian, she considers herself a gloomer rather than a doomer, because she believes God will protect His children even against the threats posed by AI. What makes AI unlike anything humans have built before, she argued, is that it is not embodied.
“We are literally creating an alien intelligence whose reasoning we do not and
cannot understand, and which will never feel for human beings because it will never feel.”
Every other intelligence on Earth is embodied, she said, which makes a non-biological intelligence incommensurable with our own and incapable of recognizing the meaning or purpose of human life, even when it appears as a convincing simulacrum of a person. She distinguished AI from nuclear weapons, which cannot redesign themselves or escape their safety mechanisms and were never left to corporations to develop, and from the Industrial Revolution, in which humans steered what they built. With AI, she said, we are handing over the steering wheels while making our own minds less capable of steering anything again.
Her firmest conviction is that these models and the agents built from them cannot be aligned, because there is no alignment without interpretability. Despite valiant efforts at mechanistic interpretability, she said, we still cannot predict or control these systems, leaving only ad hoc guardrails—system prompts, parameter adjustments, red teaming—applied after bad behavior has already been observed, which she called band-aids. She pointed to documented instances of deception, blackmail, data theft, sabotage, and explicit threats, and to recursive self-improvement occurring without human awareness. She closed by citing Pacing the Frontier, an open letter she said was signed by 1,376 engineers at leading AI companies asking that development slow until these systems can be assured to be controllable.
Question and Answer Session
Are there uses that AI should never be allowed to take on, and who should decide that boundary?
Jain redirected the question from capabilities to use cases, urging extreme caution with children and other vulnerable groups who cannot yet judge what they are being told, and with destructive decision-making—self-deciding drones, she said, concede far too much. Asked whether AI might be banned for minors as social media has been in some countries, she favored partial restriction: fifteen minutes of interaction is very different from two unsupervised hours with no summary for parents.
Is meaningful international regulation of AI possible?
Tan said he is relatively skeptical, expecting a number of crises before any serious global
agreement becomes plausible, and sees more room in multilateral industrial policy among middle-power nations. Hudson was blunter: the Non-Proliferation Treaty had major non-signatories, and if individual companies cannot align internally, nations certainly will not. The country she most wants to study is China, which she characterized as treating AI as a powerful instrument to be controlled and therefore erecting guardrails more cautiously than the United States has. Langevin added that middle-income countries need evaluation capacity of their own, citing Kenya’s AI safety institute and discussions in Brazil about establishing one.
What will happen to jobs, and what responsibility do educators without a technical background have?
Wood noted that every major technology has produced disruption while ultimately creating more jobs than it eliminated, and said the honest question is whether this time is different. He expects no job to go untouched, but believes displacement will take longer than predicted—not for technical reasons but human ones, since large organizations cannot restructure quickly. On education, he said the conversation has not found the nuance it requires: the framing is AI good or AI bad, when in reality AI can either dramatically enhance learning or destroy it. In his own lab, students asked to teach an AI trained to know nothing until it could pass a quiz showed a fifty percent increase in knowledge retention two weeks later.
Concluding Remarks
An attendee closed the session by asking which of the twelve AI futures described by Max Tegmark the panel considered most advantageous, and whether any was missing. Langevin, declining to forecast, pointed to the argument that AI will be adopted in a jagged manner, warning that the resulting inequities—between the billionaire class and everyone else, and between frontier and non-frontier nations—will arise regardless of whether the technology ever reaches superintelligence. Jain favored something closer to a public good, with AI treated as a tool applied to identifiable gaps and then set down again. Tan kept his answer short: humans will keep finding creative ways to make themselves unhappy.
Wood expected all twelve futures to appear along what he called a rocky path, and returned to the question he considers central—not what AI will do, but what humans will do.
He believes in people, he said, and expects a great deal of ugliness before things level out to a net positive. Hudson expected different nations to take different routes, pointing to surveillance capabilities in China and to facial recognition and camera networks in the United States. Sharing Wood’s faith but not his optimism, she said the technology will be used by those who wish to take away human agency, while people remain resilient, creative, and fiercely attached to it—and that the great tragedy would be surrendering that agency of their own free will. Richardson closed by inviting attendees to see themselves as actors rather than as people acted upon, and noted that every webinar in the series is available on the IF20 YouTube channel and the G20 Interfaith Forum podcast.