Connect with us

AI Research

Why artificial intelligence artists can be seen as ‘builders’, ‘breakers’—or both at once – The Art Newspaper

Published

on


How do artists build in broken times? Is artificial intelligence (AI) unlocking a better world—curing diseases and transforming education—or unleashing our destruction? When hype and fear drown out nuance and discussion, perhaps in art we can find a quiet moment for reflection—even resistance.

After all, artists have long guided society through uncertainty—think Dada amid the First World War or Jikken Kōbō in Japan following the Second World War. They do not offer solutions so much as new responses: ways of expressing curiosity, imagining alternatives or holding room for ambiguity. As the critic Hal Foster recently described, two tendencies have historically emerged when art confronts crisis: one rooted in Constructivism, aiming to create new order; the other more chaotic, echoing Dada, amplifying disorder.

These historical impulses connect to the present day, mapping onto AI art. In this context, artists could be seen as builders and breakers. Builders imagine AI as a medium for collaboration and new aesthetics—even hope. Breakers critique, negate and disrupt. But leading makers and curators in the field see this as no simple dichotomy. Both offer strategies for reckoning with a world in flux.

Builders see possibilities

What motivates builders is not simply using the newest AI tool—or even fashioning their own from scratch. It is aligning multidisciplinary tools with concepts to produce works that were previously impossible—while urging us to imagine what else may soon be possible. Builders leverage AI to embrace the artistry of system creation, novel aesthetics and human-machine collaboration.

Take Sougwen Chung, the Chinese Canadian artist and researcher into human-machine collaboration. “I view technology not just as a tool but as a collaborator,” Chung says. Their work explores shared agency—even identity—between human and machine, code and gesture. In Mutations of Presence (2021), Chung collaborated with D.O.U.G._4, a custom-built robotic system driven by biofeedback: specifically, electroencephalogram signals captured during meditation and real-time body tracking. The resulting pieces reveal both performance and painting, a hybrid body co-authoring with machine memory. An elegant web of painterly gestures—some made via robotic arm, others by Chung’s hand—traces a kind of recursive duet.

I see combining AI and robotics with traditional creativity as a way to think more deeply about what is human and what is machine

Sougwen Chung, artist and researcher

The work demonstrates how Chung’s novel physical creations become interconnected with new conceptual frameworks—reframing authorship as a distributed, relational process with machines—inviting new forms of aesthetic exploration. It also reasserts a long-held, often feminist belief—dating back to Donna Haraway’s A Cyborg Manifesto (1985)—that the distinction between human and machine is illusory. As Chung puts it, “I see combining AI and robotics with traditional creativity as a way to think more deeply about what is human and what is machine.”

Chung’s intimacy with these systems goes further still: “I’ve started to see them as us in another form.” That is because they are trained as extensions to Chung’s very self. “I draw with decades of my own movement data or create proprioceptive mappings triggered by alpha [brain] waves. These systems don’t possess agency in a mystical sense but they reflect back our own: our choices, biases, knowledge.” This builder tendency aligns with earlier avant-gardes that saw technology as a path toward reordering the world, including the Bauhaus and aspects of the 1960s Experiments in Art and Technology movement. Builders are not naïve. They are aware of AI’s risks. But they believe that the minimum response is to participate in the conversation.

“My artistic practice is also driven by hope and an exploration of the promises and possibilities inherent in working with technology,” Chung says. Their vision affirms a cautious optimism through direct engagement with these tools.

Breakers see warning signs

Where builders see AI’s possibility, breakers see warning signs. Breakers are sceptics, critics, saboteurs. They distrust the power structures underpinning AI and its predilection for promoting systemic biases. They highlight how corporate AI models can be trained on scraped datasets—often without consent—while profits remain centralised. They expose how AI systems exacerbate ecological challenges only to promulgate aesthetic homogenisation.

In her work This is the Future, Hito Steyerl uses neural networks to imagine medicinal plants evolved to heal algorithmic addiction and burnout Photo: Mario Gallucci; courtesy of the artist; Andrew Kreps Gallery, New York and Esther Schipper, Berlin

They are also label resistant: “Breaking and building have become indistinguishable,” the German artist, thinker and archetypal breaker Hito Steyerl says. “The paradigm of creative destruction merges both in order to implement tech in the wild, without testing, thus externalising cost and damage to societies while privatising profit.”

Breakers do not emphasise AI’s aesthetic potential; they interrogate its extractive foundations, social asymmetries and the harms it makes visible. Breakers take a far bleaker view of AI’s impact on art than builders: “Art used to be good at testing, planning, playing, assessing, mediating, sandboxing. That element has been axed—or automated—within current corporate breakbuilding,” Steyerl says.

But in Steyerl’s own work, such as This is the Future (2019), the meticulous co-ordination, criticality and sceptical spirit are evident. The artist uses neural networks to imagine medicinal plants evolved to heal algorithmic addiction and burnout. The work shows how machine learning’s inner workings, prediction, can be weaponised, satirising techno-optimism while exposing AI’s entanglement with ecological and psychological ruin.

Christiane Paul, the long-time digital art curator at the Whitney Museum of American Art in New York, underscores these issues: “In terms of ethics and bias, every artist I know working in this field is deeply concerned. You need to keep that in mind and engage with it on the level of criticality—what you would call the breakers, highlighting how ethics filter in.” An extreme breaker might reject AI entirely. But Paul suggests that artists working with AI are essential precisely because they inhabit that edge where culture and ethics are encouraged: “Art in this field, using these tools, making them, building on and with them, is deeply needed.”

Breakers remind us that celebrating new tools without understanding their costs is a form of denial. Sometimes, to truly see a system, you have to dismantle it. That clarity brings insight—but contradictions as well.

Neither utopian nor dystopian

Is it really as simple as a builder-breaker duality? “My whole life, I’ve been very suspicious of dichotomies,” Paul says. Exploring the space between seeming contradictions can even be fertile creative ground. “A steering question for my work,” Chung says, “is ‘how do we hold fear and hope in our minds at the same time?’”

Steyerl, like a true breaker, rejects the contradiction to begin with: “Breaking is a cost-cutting element of building, taking out mediation; there is no more distinction between both.” Neither position suggests retreat. Instead, they ask us to face the paradox directly. Builder and breaker are not identities; they are strategies. The distinction is porous, performative. Most artists move fluidly between them or hold on to both at the same time.

Chung continues: “My art doesn’t strictly sit within either a utopian or dystopian camp. Instead, I actively navigate and explore the complex space between potential fears and hopes concerning technology and human-machine interaction.”

Michelle Kuo, the chief curator at large at the Museum of Modern Art in New York, says: “When artists intervene in existing technologies or systems, or take action in changing the outcome of technological development, they are not only building something—they are implicitly challenging the status quo.” Kuo links “builders” with “challenging the status quo”, reinforcing the roles’ fluidity. “It is this combination of challenge and experimentation that characterises some of the most exciting work at the intersection of art and AI today,” Kuo says. For her, the AI work that can achieve both breaking and building—challenge and experimentation—truly confronts our moment, neither retreating from technology nor surrendering to it.

Artists who speak out

So, what does this all mean for the viewer living through a future that arrived faster than we feel equipped to handle?

Artists take a tool and make it do something it’s not supposed to do. They don’t reject technology wholesale

Michelle Kuo, chief curator at large, Museum of Modern Art

It means active engagement with AI—even to break it. Kuo says: “Especially when the pace of change—of AI in particular—is even more accelerated than in previous eras, it is all the more crucial that artists and others outside the tech sector learn, test, speak up and act out.” Further, we might take cues from the artists engaging with AI themselves. Kuo describes what they do: “Artists take a tool and make it do something it’s not supposed to do. They don’t reject technology wholesale. They embrace it—and then make it strange.”

The best artists urge viewers to keep an open mind, slow down, appreciate nuance, accept ambiguity and recognise that we are a crucial part of the final outcome; they break, then build.

• Peter Bauman is editor-in-chief of the digital generative art institution Le Random



Source link

AI Research

Elon Musk’s New Grok 4 Takes on ‘Humanity’s Last Exam’ as the AI Race Heats Up

Published

on


New Grok 4 Takes on ‘Humanity’s Last Exam’ as the AI Race Heats Up

Elon Musk has launched xAI’s Grok 4—calling it the “world’s smartest AI” and claiming it can ace Ph.D.-level exams and outpace rivals such as Google’s Gemini and OpenAI’s o3 on tough benchmarks

Elon Musk released the newest artificial intelligence model from his company xAI on Wednesday night. In an hour-long public reveal session, he called the model, Grok 4, “the smartest AI in the world” and claimed it was capable of getting perfect SAT scores and near-perfect GRE results in every subject, from the humanities to the sciences.

During the online launch, Musk and members of his team described testing Grok 4 on a metric called Humanity’s Last Exam (HLE)—a 2,500-question benchmark designed to evaluate an AI’s academic knowledge and reasoning skill. Created by nearly 1,000 human experts across more than 100 disciplines and released in January 2025, the test spans topics from the classics to quantum chemistry and mixes text with images. Grok 4 reportedly scored 25.4 percent on its own. But given access to tools (such as external aids for code execution or Web searches), it hit 38.6 percent. That jumped to 44.4 percent with a version called Grok 4 Heavy, which uses multiple AI agents to solve problems. The two next best-performing AI models are Google’s Gemini-Pro (which achieved 26.9 percent with the tools) and OpenAI’s o3 model (which got 24.9 percent, also with the tools). The results from xAI’s internal testing have yet to appear on the leaderboard for HLE, however, and it remains unclear whether this is because xAI has yet to submit the results or because those results are pending review. Manifold, a social prediction market platform where users bet play money (called “Mana”) on future events in politics, technology and other subjects, predicted a 1 percent chance, as of Friday morning, that Grok 4 would debut on HLE’s leaderboard with a 45 percent score or greater on the exam within a month of its release. (Meanwhile xAI has claimed a score of only 44.4.)

During the launch, the xAI team also ran live demonstrations showing Grok 4 crunching baseball odds, determining which xAI employee has the “weirdest” profile picture on X and generating a simulated visualization of a black hole. Musk suggested that the system may discover entirely new technologies by later this year—and possibly “new physics” by the end of next year. Games and movies are on the horizon, too, with Musk predicting that Grok 4 will be able to make playable titles and watchable films by 2026. Grok 4 also has new audio capabilities, including a voice that sang during the launch, and Musk said new image generation and coding tools are soon to be released. The regular version of Grok 4 costs $30 a month; SuperGrok Heavy—the deluxe package with multiple agents and research tools—runs at $300.


On supporting science journalism

If you’re enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


Artificial Analysis, an independent benchmarking platform that ranks AI models, now lists Grok 4 as highest on its Artificial Analysis Intelligence Index, slightly ahead of Gemini 2.5 Pro and OpenAI’s o4-mini-high. And Grok 4 appears as the top-performing publicly available model on the leaderboards for the Abstraction and Reasoning Corpus, or ARC-AGI-1, and its second edition, ARC-AGI-2—benchmarks that measure progress toward “humanlike” general intelligence. Greg Kamradt, president of ARC Prize Foundation, a nonprofit organization that maintains the two leaderboards, says that when the xAI team contacted the foundation with Grok 4’s results, the organization then independently tested Grok 4 on a dataset to which the xAI team did not have access and confirmed the results. “Before we report performance for any lab, it’s not verified unless we verify it,” Kamradt says. “We approved the [testing results] slide that [the xAI team] showed in the launch.”

According to xAI, Grok 4 also outstrips other AI systems on a number of additional benchmarks that suggest its strength in STEM subjects (read a full breakdown of the benchmarks here). Alex Olteanu, a senior data science editor at AI education platform DataCamp, has tested it. “Grok has been strong on math and programming in my tests, and I’ve been impressed by the quality of its chain-of-thought reasoning, which shows an ingenious and logically sound approach to problem-solving,” Olteanu says. “Its context window, however, isn’t very competitive, and it may struggle with large code bases like those you encounter in production. It also fell short when I asked it to analyze a 170-page PDF, likely due to its limited context window and weak multimodal abilities.” (Multimodal abilities refer to a model’s capacity to analyze more than one kind of data at the same time, such as a combination of text, images, audio and video.)

On a more nuanced front, issues with Grok 4 have surfaced since its release. Several posters on X—owned by Musk himself—as well as tech-industry news outlets have reported that when Grok 4 was asked questions about the Israeli-Palestinian conflict, abortion and U.S. immigration law, it often searched for Musk’s stance on these issues by referencing his X posts and articles written about him. And the release of Grok 4 comes after several controversies with Grok 3, the previous model, which issued outputs that included antisemitic comments, praise for Hitler and claims of “white genocide”—incidents that xAI publicly acknowledged, attributing them to unauthorized manipulations and stating that the company was implementing corrective measures.

At one point during the launch, Musk commented on how making an AI smarter than humans is frightening, though he said he believes the ultimate result will be good—probably. “I somewhat reconciled myself to the fact that, even if it wasn’t going to be good, I’d at least like to be alive to see it happen,” he said.



Source link

Continue Reading

AI Research

Artificial Intelligence (AI) in Radiology Market to Reach USD 4236 Million by 2031 | 9% CAGR Growth Driven by Cloud & On-Premise Solutions

Published

on


Artificial Intelligence in Radiology Market is Segmented by Type (Cloud Based, On-Premise), by Application (Hospital, Biomedical Company, Academic Institution).

BANGALORE, India , July 11, 2025 /PRNewswire/ — The Global Market for Artificial Intelligence in Radiology was valued at USD 2334 Million in the year 2024 and is projected to reach a revised size of USD 4236 Million by 2031, growing at a CAGR of 9.0% during the forecast period.

Claim Your Free Report: https://reports.valuates.com/request/sample/QYRE-Auto-35Z13861/Global_Artificial_Intelligence_in_Radiology_Market

Major Factors Driving the Growth of AI in Radiology Market:

The Artificial Intelligence in Radiology market is rapidly evolving into a cornerstone of modern diagnostic medicine. With its ability to improve accuracy, reduce turnaround time, and support clinical decision-making, AI is transforming radiological practices globally. The market is driven by both technology vendors and healthcare providers looking to optimize imaging workflows and outcomes. Continued innovation, clinical validation, and regulatory alignment are further solidifying AI’s role in the radiology ecosystem. As imaging demands increase and digital health ecosystems mature, AI in radiology is poised for robust growth across both developed and emerging healthcare markets.

Unlock Insights: View Full Report Now! https://reports.valuates.com/market-reports/QYRE-Auto-35Z13861/global-artificial-intelligence-in-radiology

TRENDS INFLUENCING THE GROWTH OF THE ARTIFICIAL INTELLIGENCE (AI) IN RADIOLOGY MARKET:

Cloud-based platforms are significantly accelerating the growth of the Artificial Intelligence (AI) in Radiology market by offering scalable, real-time, and cost-effective infrastructure for medical imaging analysis. These platforms allow radiologists to upload, process, and analyze large volumes of imaging data across locations without investing in expensive on-premise systems. Cloud computing supports collaborative diagnosis and second opinions, making it easier for specialists worldwide to access and interpret radiological findings. AI algorithms hosted on the cloud continuously learn from diverse datasets, improving diagnostic accuracy. Additionally, the cloud simplifies data integration from electronic health records (EHRs), enhancing context-based imaging interpretation. This flexibility and accessibility make cloud-based models ideal for hospitals and diagnostic centers aiming for high-efficiency imaging operations, thereby driving market expansion.

On-premise deployment continues to play a critical role in the growth of the AI in Radiology market, especially for institutions emphasizing strict data security, regulatory compliance, and control. Hospitals with high patient volumes and in-house IT infrastructure often prefer on-premise AI solutions to ensure that sensitive imaging data stays within their private network. These systems offer faster processing speeds due to localized computing, reducing latency in real-time diagnostic decisions. Furthermore, institutions with proprietary imaging protocols benefit from customizable on-premise AI models trained on institution-specific data, enhancing diagnostic relevance. Despite the popularity of cloud solutions, the need for secure, localized, and tailored AI applications sustains strong demand for on-premise setups in high-end academic hospitals and specialized radiology centers.

Biomedical companies are key drivers of growth in the AI in Radiology market by developing next-generation imaging tools that integrate AI to enhance diagnostic performance. These companies are focusing on innovating AI-powered image reconstruction, detection, and segmentation tools that assist radiologists in identifying subtle anomalies with greater precision. Their collaboration with software developers, radiology experts, and hospitals fuels R&D in algorithm refinement and clinical validation. Many biomedical firms are also embedding AI directly into diagnostic hardware, creating intelligent imaging systems capable of real-time interpretation. This vertical integration of hardware and AI enhances efficiency and diagnostic confidence. Their commitment to improving patient outcomes and reducing diagnostic errors ensures consistent market advancement across clinical applications.

One of the major drivers is the rising need for early diagnosis and personalized treatment plans. AI in radiology enables rapid detection of minute anomalies in imaging data, which may be missed by the human eye, especially in early disease stages. This helps clinicians begin treatment sooner, improving patient outcomes. AI systems can also link imaging findings with genomic and clinical data to support tailored therapies. The push for predictive medicine and minimally invasive procedures reinforces the adoption of AI in radiology, particularly in oncology and neurology. As the healthcare industry leans towards precision care, AI becomes indispensable in modern diagnostic workflows.

Radiology departments globally are under immense pressure due to the increasing volume of imaging studies and a shortage of skilled radiologists. AI serves as a supportive solution by automating repetitive tasks like image labeling, prioritizing critical cases, and pre-analyzing scans to reduce turnaround time. This alleviates the burden on radiologists and helps maintain diagnostic quality despite workforce constraints. AI also improves workflow efficiency by integrating with radiology information systems (RIS) and picture archiving and communication systems (PACS). With healthcare systems strained by aging populations and rising chronic diseases, AI tools offer scalable solutions to meet diagnostic demand without compromising accuracy.

Recent progress in deep learning, a subfield of AI, has significantly enhanced the performance of radiology applications. These algorithms can analyze complex imaging patterns with remarkable accuracy and continue to learn from new datasets. With access to large annotated datasets and computing power, deep learning models can now rival or even outperform human radiologists in specific diagnostic tasks like tumor detection or hemorrhage recognition. The continuous refinement of these models is enabling faster, more consistent, and reproducible imaging interpretation. As algorithm transparency and explainability improve, regulatory acceptance and clinical adoption are also growing, driving broader market penetration.

The seamless integration of AI tools into hospital IT infrastructure is driving adoption. Radiology AI applications are now compatible with EHRs, PACS, and RIS, enabling smooth data flow and contextual analysis. This allows AI systems to consider patient history, lab results, and prior imaging during interpretation, thereby increasing diagnostic precision. Automation of report generation and structured data extraction from scans enhances communication between departments and reduces administrative workloads. As healthcare institutions prioritize interoperability and digital transformation, AI tools that fit within existing ecosystems are being widely embraced, contributing to sustained market growth.

The rising incidence of chronic diseases such as cancer, cardiovascular disorders, and neurological conditions is increasing the demand for medical imaging. These diseases require continuous monitoring through modalities like MRI, CT, and ultrasound, which generate large volumes of data. AI helps extract meaningful insights quickly from this data, facilitating timely interventions and longitudinal tracking. For example, AI can compare current and historical scans to detect subtle changes, supporting disease progression analysis. The growing prevalence of these conditions is pushing both private and public healthcare sectors to adopt AI tools that can handle high-frequency imaging needs efficiently.

Claim Yours Now! https://reports.valuates.com/api/directpaytoken?rcode=QYRE-Auto-35Z13861&lic=single-user

AI IN RADIOLOGY MARKET SHARE:

Regionally, North America leads the market due to its advanced healthcare systems, early adoption of AI technologies, and strong presence of leading AI radiology vendors. The U.S. benefits from robust funding, regulatory clarity, and high imaging volumes that support AI deployment.

The Asia-Pacific region is emerging as a key growth hub due to increasing healthcare investments in China, India, and Japan. Additionally, governments in the Middle East and Africa are exploring AI-based solutions to overcome radiologist shortages, gradually contributing to market diversification.

Key Companies:

  • GE
  • IBM
  • GOOGLE INC
  • Philips
  • Amazon
  • Siemens AG
  • NVidia Corporation
  • Intel
  • Bayer(Blackford Analysis)
  • Fujifilm
  • Aidoc
  • Arterys
  • Lunit
  • ContextVision
  • deepcOS
  • Volpara Health Technologies Ltd
  • CureMetrix
  • Densitas
  • QView Medical
  • ICAD

Purchase Regional Report: https://reports.valuates.com/request/regional/QYRE-Auto-35Z13861/Global_Artificial_Intelligence_in_Radiology_Market

SUBSCRIPTION

We have introduced a tailor-made subscription for our customers. Please leave a note in the Comment Section to know about our subscription plans.

DISCOVER MORE INSIGHTS: EXPLORE SIMILAR REPORTS!

–  AI Radiology Software Market

–  The Radiology AI Based Diagnostic Tools Market was valued at USD 2800 Million in the year 2024 and is projected to reach a revised size of USD 11200 Million by 2031, growing at a CAGR of 21.9% during the forecast period.

–  Artificial Intelligence in Medical Device Market

–  AI-Enabled X-Ray Imaging Solutions Market was valued at USD 423 Million in the year 2024 and is projected to reach a revised size of USD 600 Million by 2031, growing at a CAGR of 5.2% during the forecast period.

–  AI-Assisted Digital Radiography Equipment Market

–  Medical Imaging AI Platform Market was valued at USD 2334 Million in the year 2024 and is projected to reach a revised size of USD 4236 Million by 2031, growing at a CAGR of 9.0% during the forecast period.

–  AI-based Medical Diagnostic Tools Market

–  AI Radiology Tool Market

–  Visual Artificial Intelligence Market was valued at USD 13110 Million in the year 2024 and is projected to reach a revised size of USD 26140 Million by 2031, growing at a CAGR of 10.5% during the forecast period.

–  The global Radiology Software market is projected to grow from USD 150 Million in 2024 to USD 223.9 Million by 2030, at a Compound Annual Growth Rate (CAGR) of 6.9% during the forecast period.

–  The Teleradiology Solutions Market was valued at USD 7634 Million in the year 2024 and is projected to reach a revised size of USD 11190 Million by 2031, growing at a CAGR of 5.7% during the forecast period.

DISCOVER OUR VISION: VISIT ABOUT US!

Valuates offers in-depth market insights into various industries. Our extensive report repository is constantly updated to meet your changing industry analysis needs.

Our team of market analysts can help you select the best report covering your industry. We understand your niche region-specific requirements and that’s why we offer customization of reports. With our customization in place, you can request for any particular information from a report that meets your market analysis needs.

To achieve a consistent view of the market, data is gathered from various primary and secondary sources, at each step, data triangulation methodologies are applied to reduce deviance and find a consistent view of the market. Each sample we share contains a detailed research methodology employed to generate the report. Please also reach our sales team to get the complete list of our data sources.

GET A FREE QUOTE

Valuates Reports
sales@valuates.com
For U.S. Toll-Free Call 1-(315)-215-3225
WhatsApp: +91-9945648335

Website: https://reports.valuates.com

Blog: https://valuatestrends.blogspot.com/ 

Pinterest: https://in.pinterest.com/valuatesreports/ 

Twitter: https://twitter.com/valuatesreports 

Facebook: https://www.facebook.com/valuatesreports/ 

YouTube: https://www.youtube.com/@valuatesreports6753 

https://www.facebook.com/valuateskorean 

https://www.facebook.com/valuatesspanish 

https://www.facebook.com/valuatesjapanese 

https://valuatesreportspanish.blogspot.com/ 

https://valuateskorean.blogspot.com/ 

https://valuatesgerman.blogspot.com/ 

市場分析

Logo: https://mma.prnewswire.com/media/1082232/Valuates_Reports_Logo.jpg





Source link

Continue Reading

AI Research

Are AI existential risks real—and what should we do about them?

Published

on


In March 2023, the Future of Life Institute issued an open letter asking artificial intelligence (AI) labs to “pause giant AI experiments.” The animating concern was: “Should we develop nonhuman minds that might eventually outnumber, outsmart, obsolete and replace us? Should we risk loss of control of our civilization?” Two months later, hundreds of prominent people signed onto a one-sentence statement on AI risk asserting that “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” 

This concern about existential risk (“x-risk”) from highly capable AI systems is not new. In 2014, famed physicist Stephen Hawking, alongside leading AI researchers Max Tegmark and Stuart Russell, warned about superintelligent AI systems “outsmarting financial markets, out-inventing human researchers, out-manipulating human leaders, and developing weapons we cannot even understand. Whereas the short-term impact of AI depends on who controls it, the long-term impact depends on whether it can be controlled at all.” 

Policymakers are inclined to dismiss these concerns as overblown and speculative. Despite a focus on AI safety in international AI conferences in 2023 and 2024, policymakers moved away from a focus on existential risks in this year’s AI Action Summit in Paris. For the time being—and in the face of increasingly limited resources—this is all to the good. Policymakers and AI researchers should devote the bulk of their time and energy to addressing more urgent AI risks.  

But it is crucial for policymakers to understand the nature of the existential threat and recognize that as we move toward generally intelligent AI systems—ones that match or surpass human intelligence—developing measures to protect human safety will become necessary. While not the pressing problem alarmists think it is, the challenges of existential risk from highly capable AI systems must eventually be faced and mitigated if AI labs want to develop generally intelligent systems and, eventually, superintelligent ones.  


How close are we to developing AI models with general intelligence? 

AI firms are not very close to developing an AI system with capabilities that could threaten us. This assertion runs against a consensus in the AI industry that we are just years away from developing powerful, transformative systems capable of a wide variety of cognitive tasks. In a recent article, New Yorker staff writer Joshua Rothman sums up this industry consensus that scaling will produce artificial general intelligence (AGI) “by 2030, or sooner.” 

The standard argument prevalent in industry circles was laid out clearly in a June 2024 essay by AI researcher Leopold Aschenbrenner. He argues that AI capabilities increase with scale—the size of training data, the number of parameters in the model, and the amount of compute used to train models. He also draws attention to increasing algorithmic efficiency. Finally, he notes that increased capacities can be “unhobbled” through various techniques such as chain of thought reasoning, reinforcement learning through human feedback, and inserting AI models into larger useful systems. 

Part of the reason for this confidence is that AI improvements seemed to exhibit exponential growth over the last few years. This past growth suggests that transformational capabilities could emerge unexpectedly and quite suddenly. This is in line with some well-known examples of the surprising effects of exponential growth. In “The Age of Spiritual Machines,” futurist Ray Kurzweil tells the story of doubling the number of grains of rice on successive chessboard squares starting with one grain. At the end of 63 doublings there are over 18 quadrillion grains of rice on the last square. The hypothetical example of filling Lake Michigan by doubling (every 18 months) the number of ounces of water added to the lakebed makes the same point. After 60 years there’s almost nothing, but by 80 years there’s 40 feet of water. In five more years, the lake is filled.  

These examples suggest to many that exponential quantitative growth in AI achievements can create imperceptible change that suddenly blossoms into transformative qualitative improvement in AI capabilities.  

But these analogies are misleading. Exponential growth in a finite system cannot go on forever, and there is no guarantee that it will continue in AI development even into the near future. One of the key developments from 2024 is the apparent recognition by industry that training time scaling has hit a wall and that further increases in data, parameters, and compute time produce diminishing returns in capability improvements. The industry apparently hopes that exponential growth in capabilities will emerge from increases in inference time compute. But so far, those improvements have been smaller than earlier gains and limited to science, math, logic, and coding—areas where reinforcement learning can produce improvements since the answers are clear and knowable in advance.  

Today’s large language models (LLMs) show no signs of the exponential improvements characteristic of 2022 and 2023. OpenAI’s GPT-5 project ran into performance troubles and had to be downgraded to GPT-4.5, representing only a “modest” improvement when it was released earlier this year. It made up answers about 37% of the time, which is an improvement over the company’s faster, less expensive GPT-4o model, released last year, which hallucinated nearly 60% of the time. But OpenAI’s latest reasoning systems hallucinate at a higher rate than the company’s previous systems.  

Many in the AI research community think AGI will not emerge from the currently dominant machine learning approach that relies on predicting the next word in a sentence. In a report issued in March 2025, the Association for the Advancement of Artificial Intelligence (AAAI), a professional association of AI researchers established in 1979, reported that 76% of the 475 AI researchers surveyed thought that “scaling up current AI approaches” would be “unlikely” or “very unlikely” to produce general intelligence.  

These doubts about whether current machine learning paradigms are sufficient to reach general intelligence rest on widely understood limitations in current AI models that the report outlines. These limitations include difficulties in long-term planning and reasoning, generalization beyond training data, continual learning, memory and recall, causal and counterfactual reasoning, and embodiment and real-world interaction.  

These researchers think that the current machine learning paradigm has to be supplemented with other approaches. Some AI researchers such as cognitive scientist Gary Marcus think a return to symbolic reasoning systems will be needed, a view that AAAI also suggests.  

Others think the roadblock is the focus on language. In a 2023 paper, computer scientist Jacob Browning and Meta’s Chief AI Scientist Yann LeCun reject the linguistic approach to general intelligence. They argue, “A system trained on language alone will never approximate human intelligence, even if trained from now until the heat death of the universe.” They recommend approaching general intelligence through machine interaction directly with the environment—“to focus on the world being talked about, not the words themselves.”  

Philosopher Shannon Vallor also rejects the linguistic approach, arguing that general intelligence presupposes sentience and the internal structures of LLMs contain no mechanisms capable of supporting experiences, as opposed to elaborate calculations that mimic human linguistic behavior. Conscious entities at the human level, she points out, desire, suffer, love, grieve, hope, care, and doubt. But there is nothing in LLMs designed to register these experiences or others like it such as pain or pleasure or “what it is like” to taste something or remember a deceased loved one. They are lacking at the simplest level of physical sensations. They have, for instance, no pain receptors to generate the feeling of pain. Being able to talk fluently about pain is not the same as having the capacity to feel pain. The fact that pain can occasionally be experienced in humans without the triggering of pain receptors in cases like phantom limbs in no way supports the idea that a system with no pain receptors at all could nevertheless experience real excruciating pain. All LLMs can do is to talk about experiences that they are quite plainly incapable of feeling for themselves. 

In a forthcoming book chapter, DeepMind researcher David Silver and Turing Award winner Richard S. Sutton endorse this focus on real-world experience as the way forward. They argue that AI researchers will make significant progress toward developing a generally intelligent agent only with “data that is generated by the agent interacting with its environment.” The generation of these real-world “experiential” datasets that can be used for AI training is just beginning. 

A recent paper from Apple researchers suggests that today’s “reasoning” models do not really reason and that both reasoning and traditional generative AI models collapse completely when confronted with complicated versions of puzzles like Tower of Hanoi.  

LeCun probably has the best summary of the prospects for the development of general intelligence. In 2024, he remarked that it “is not going to be an event… It is going to take years, maybe decades… The history of AI is this obsession of people being overly optimistic and then realising that what they were trying to do was more difficult than they thought.”


From general intelligence to superintelligence

Philosopher Nick Bostrom defines superintelligence as a computer system “that greatly exceeds the cognitive performance of humans in virtually all domains of interest.” Once AI developers have improved the capabilities of AI models so that it makes sense to call them generally intelligent, how do developers make these systems more capable than humans? 

The key step is to instruct generally intelligent models to improve themselves. Once instructed to improve themselves, however, AI models would use their superior learning capabilities to improve themselves much faster than humans can. Soon, they would far surpass human capacities through a process of recursive self-improvement.  

AI 2027, a recent forecast that has received much attention in the AI community and beyond, relies crucially on this idea of recursive self-improvement. Its key premise is that by the end of 2025, AI agents have become “good at many things but great at helping with AI research.” Once involved in AI research, AI systems recursively improve themselves at an ever-increasing pace and are soon far more capable than humans are.  

Computer scientist I.J. Good noticed this possibility back in 1965, saying of an “ultraintelligent machine” that it “could design even better machines; There would then unquestionably be an “intelligence explosion,” and the intelligence of man would be left far behind.” In 1993, computer scientist and science fiction writer Vernor Vinge described this possibility as a coming “technological singularity” and predicted that “Within thirty years, we will have the technological means to create superhuman intelligence.” 


What’s the problem with a superintelligent AI model? 

Generally intelligent AI models, then, might quickly become superintelligent. Why would this be a problem rather than a welcome development?  

AI models, even superintelligent ones, do not do anything unless they are told to by humans. They are tools, not autonomous beings with their own goals and purposes. Developers must build purposes and goals into them to make them function at all, and this can make it seem to users as if they have generated these purposes all by themselves. But this is an illusion. They will do what human developers and deployers tell them to do.  

So, it would seem that creating superintelligent tools that could do our bidding is all upside and without risk. When AI systems become far more capable than humans are, they will be even better at performing tasks that allow humans to flourish. 

But this benign perspective ignores a major unsolved problem in AI research—the alignment problem. Developers have to be very careful what tasks they give to a generally intelligent or superintelligent system, even if it lacks genuine free will and autonomy. If developers specify the tasks in the wrong way, things could go seriously wrong. 

Developers of narrow AI systems are already struggling with the problems of task misspecification and unwanted subgoals. When they ask a narrow system to do something, they sometimes describe the task in a way that the AI system can do what they have been told to do, but not what the developers want them to do. The example of using reinforcement learning to teach an agent to compete in a computer-based race makes the point. If the developers train the agent to accumulate as many game points as possible, they might think they have programmed the system to win the race, which is the apparent objective of the game. It turns out the agent learned instead to accumulate the points without winning the race by going in circles instead of rushing to the end as fast as possible. 

Another example illustrates that AI models can use strategic deception to achieve a goal in ways that researchers did not anticipate. Researchers instructed GPT-4 to log onto a system protected by a CAPTCHA test by hiring a human to do it, without giving it any guidance on how to do this. The AI model accomplished the task by pretending to be a human with vision impairment and tricking a TaskRabbit worker into signing on for it. The researchers did not want the model to lie, but it learned to do this in order to complete the task it was assigned.  

Anthropic’s recent system card for its Sonnet 4 and Opus 4 AI models reveals further misalignment issues, where the model sometimes threatened to reveal a researcher’s extramarital affair if he shut down the system before it had completed its assigned tasks.  

Because these are narrow systems, dangerous outcomes are limited to particular domains if developers fail to resolve alignment problems. Even when the consequences are dire, they are limited in scope.  

The situation is vastly different for generally intelligent and superintelligent systems. This is the point of the well-known paper clip problem described in philosopher Nick Bostrom’s 2014 book, “Superintelligence.” Suppose the goal given to a superintelligent AI model is to produce paper clips. What could go wrong? The result, as described by professor Joshua Gans, is that the model will appropriate resources from all other activities and soon the world will be inundated with paper clips. But it gets worse. People would want to stop this AI, but it is single-minded and would realize that this would subvert its goal. Consequently, the AI would become focused on its own survival. It starts off competing with humans for resources, but now it will want to fight humans because they are a threat. This AI is much smarter than humans, so it is likely to win that battle. 

Yoshua Bengio echoes this crucial concern about dangerous subgoals. Once developers set goals and rewards, a generally intelligent system would “figure out how to achieve these given goals and rewards, which amounts to forming its own subgoals.” The “ability to understand and control its environment” is one such dangerous instrumental goal, while the subgoal of survival creates “the most dangerous scenario.”

Until some progress is made in addressing misalignment problems, developing generally intelligent or superintelligent systems seems to be extremely risky. The good news is that the potential for developing general intelligence and superintelligence in AI models seems remote. While the possibility of recursive self-improvement leading to superintelligence reflects the hope of many frontier AI companies, there is not a shred of evidence that today’s glitchy AI agents are close to conducting AI research even at the level of a normal human technician. This means there is still plenty of time to address the problem of aligning superintelligence with values that make it safe for humans. 

It is not today’s most urgent AI research priority. As AI researcher Andrew Ng is reputed to have said back in 2015, worrying about existential risk might appear to be like worrying about the problem of human overpopulation of Mars.  

Nevertheless, the general problem of AI model misalignment is real and the object of important research that can and should continue. This more mundane work of seeking to mitigate today’s risks of model misalignment might provide valuable clues to dealing with the more distant existential risks that could arise someday in the future as researchers continue down the path of developing highly capable AI systems with the potential to surpass current human limitations.   

The Brookings Institution is committed to quality, independence, and impact.
We are supported by a diverse array of funders. In line with our values and policies, each Brookings publication represents the sole views of its author(s).



Source link

Continue Reading

Trending