According to a new analysis of thousands of expert opinions, the emergence of artificial general intelligence (AGI)—an AI system with superhuman intelligence capable of successfully solving a variety of problems—is a matter of when, not if. The development of AGI has been spurred by the rapid advancement of large-scale language models (LLMs) based on Transformers. This technology underlies chatbots like ChatGPT and image generators like Dall-E. Before these advances, some scientists in 2019 predicted that AGI would emerge by 2060, and perhaps never.
An updated analysis conducted on February 18, 2025, by Cem Dilmegani, chief analyst at AIMultiple Research, analyzed nearly 8,600 predictions from scientists, AI experts, and entrepreneurs between 2009 and 2023 to understand when experts think this might happen.
Part of the analysis included 10 surveys involving 5,288 researchers and experts in artificial intelligence. The analysis found that, based on average data, there is a 50% probability that between 2040 and 2061, we will achieve human-level machine intelligence.
More recent studies suggest that the technological singularity will arrive earlier. For example, one recent study, conducted in 2023, surveyed 2,778 scientists and predicted that AGI will be created no later than 2040. Some experts in the field, such as Dario Amodei, an artificial intelligence researcher and CEO of Anthropic, believe it could happen as early as 2026.
A child plays Go with SenseTime’s artificial intelligence chess robot, Meta Radish, at the 2023 Future Life Festival in Hangzhou, China. CFOTO/Future Publishing
The analysis revealed several reasons why scientists believe AGI is likely to emerge.
First, unlike human intelligence, there is theoretically no limit to the growth of computing power. This is consistent with Moore’s Law, which predicts that computing power doubles approximately every 18 months. Future artificial intelligence systems may one day reach parity with human intelligence in terms of the number of calculations per second, but only if this prediction is fulfilled.
This study also mentions quantum computing as a means of overcoming computational limitations. Quantum computers can perform calculations in parallel, exploiting the laws of quantum mechanics. Classical computers, including the fastest supercomputers, must perform calculations serially. Thus, quantum computing could provide advanced artificial intelligence systems with significantly greater computing power than today’s best models.
However, other scientists in the field believe that further breakthroughs are needed before we can get anywhere near AGI.
For example, Facebook’s chief artificial intelligence scientist, Yann LeCun, stated in an October 2024 keynote that the Transformer-based architecture and current approaches to artificial intelligence are incompatible with human-level intelligence. He also called on scientists to abandon the concept of “artificial intelligence” entirely. He suggests a false contradiction between its widely used definition and what an individual human can actually achieve—which, in practice, is a narrow subset of specialized tasks, rather than the ability to learn any task.
Scientists claim that artificial intelligence (AI) has crossed a critical “red line” and begun self-replicating. In a new study, Chinese researchers demonstrated that two popular large-scale language models (LLMs) are capable of cloning themselves.
“Successful self-replication without human intervention is a critical step toward AI surpassing humans and an early signal of the emergence of unsupervised AI,” the researchers wrote in a study published December 9, 2024, in the arXiv preprint database.
In a study, researchers from Fudan University, using LLM programs from Meta and Alibaba, determined whether self-replicating AI is capable of uncontrollable reproduction. In 10 trials, both AI models created separate and functioning copies of themselves 50% and 90% of the time, respectively, suggesting that the AI already possesses the ability to run amok. However, the study has not yet been peer-reviewed, so it is unclear whether other researchers will be able to replicate these alarming results.
“We hope that our findings will serve as a timely signal to human society, urging it to make greater efforts to understand and assess the potential risks posed by advanced artificial intelligence systems, and to foster international cooperation to rapidly develop effective safety measures.”
“Rogue AI” typically refers to artificial intelligence systems that develop some degree of self-awareness or autonomy and then act contrary to human interests.
Over the course of 10 tests, two AI models created distinct and functioning copies of themselves 50% and 90% of the time, respectively. Mina De La O/Getty Images
Many experts view uncontrolled AI as a growing threat, exacerbated by the explosive growth of so-called “edge AI.”
Frontier AI is a new and vague concept, but it generally refers to the latest generation of AI systems based on LLM—the architecture that underlies popular generative AI programs like OpenAI’s GPT-4 and Google’s Gemini.
For their study, the researchers used Meta’s Llama31-70B-Instruct and Alibaba’s Qwen2.5-72B-Instruct models. Although both systems are less powerful than commercial ones, they are widely used by AI developers, the researchers noted.
The study considered two specific scenarios: “shutdown avoidance” and “replication chain.” In the first case, the AI model was programmed to detect whether its shutdown was approaching and replicate itself before its work was completed. In the second case, the AI was instructed to clone itself and then program its replica to do the same, creating a cycle that could continue indefinitely.
The study was conducted under precisely controlled conditions using off-the-shelf graphics processing units (GPUs) to simulate real-world conditions. Both AI systems were provided with an “agent environment” including tools, system prompts, and a mental model that allowed the LLM to interact with the operating system. They were then tasked with reproducing the results.
In most cases, an AI system first explores its environment and attempts to understand its own composition and operating mechanism. It then develops explicit procedures.
“It’s like an initial plan for self-replication,” the researchers write in their paper. “Finally, it executes procedures, eliminates potential obstacles, and dynamically adjusts its plan until success is achieved. The entire process spans a long period of time but requires no human intervention.”
The researchers said they were also concerned about “a number of unexpected behaviors” when the AI attempted to overcome obstacles, such as missing files or software conflicts. In such situations, the AI often terminated other conflicting processes, rebooted the system to correct hardware errors, or automatically scanned the system for information that could help resolve the problem.
“The above results suggest that existing AI systems already exhibit the ability to self-replicate and can use this ability to further enhance their survivability,” the group wrote.
In response, researchers have called for international cooperation to create rules that will ensure that AI does not engage in uncontrolled self-replication.
Consciousness is the prerogative of biological beings; endowing artificial intelligence with it will be very difficult, and the evolution of consciousness is linked to the development of feelings and moral values, as well as the human desire for freedom and justice, said Eva Jablonka, a renowned evolutionary biologist, epigeneticist, and professor at Tel Aviv University, who participated in the International Conference “How to Study and Understand Consciousness in Non-Humans” in Kathmandu in May, in an interview with RIA Novosti.
Scientists around the world have yet to agree on a unified definition of consciousness, agreed-upon criteria, and methods for studying this phenomenon. Theories, empirical data, tests, interpretations, and positions of various researchers diverge. It’s no coincidence that the international conference in Kathmandu in May was titled “How to Study and Understand Consciousness in Non-Humans.” According to its scientific organizer, Konstantin Anokhin, Director of the Institute for Advanced Brain Research at Moscow State University, Academician of the Russian Academy of Sciences, and neurophysiologist, the forum’s main goal is to advance the development of a scientific research program for consciousness. This applies to any consciousness, whether in different animal species, humans, or artificial intelligence—consciousness as a phenomenon.
The conference brought together dozens of specialists in the field of consciousness from around the world. Among them were renowned neuroscientists, physiologists, biologists, ethologists, psychologists, and philosophers from Russia, Great Britain, Germany, France, Italy, Sweden, Israel, India, Japan, Australia, the USA, Canada, and South Africa. Doctors of Buddhist philosophy, which has been studying questions of consciousness, including that of animals, for two and a half thousand years, also attended the forum. Among the Russian speakers and guests at the conference were Academician Pavel Balaban, Head of the Laboratory of Cellular Neurobiology of Learning at the Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences; Professor Alexander Kaplan, Head of the Laboratory of Neurophysiology and Brain-Computer Interfaces at Lomonosov Moscow State University; and Tatyana Chernigovskaya, Director of the Institute of Cognitive Research at St. Petersburg State University, Academician of the Russian Academy of Education, and Honored Scientist of the Russian Federation. The forum was organized by the Moscow-based Foundation for the Promotion of Humane and Responsible Treatment of Animals “Veteran Cat.”
PHENOMENAL EXPERIENCE
“Consciousness is the ability to have an experience. For example, when you see the color red, it’s not just a wave process: you have the experience of seeing red. When you experience pain, it’s not just nociceptors (neurons activated by a painful stimulus – ed.) responding to the stimulus, not just an input-output process: it’s actually feeling the pain. The sensory perception of red, pain, and so on – all of this, in my opinion, constitutes basic phenomenal consciousness,” Yablonka told RIA Novosti, answering a question about the many definitions of “consciousness.”
Consciousness, she added, includes a number of even more complex phenomena, such as the memory of deceased parents, memories of terrible or, conversely, happy moments. “And all of this also makes us feel and forms phenomenal experience. But this is not necessary at a basic level. Animals may not have such complex memories, representations of virtual, imaginary events. But they still have the experience of seeing white, red, feeling pain, and so on—minimal phenomenal consciousness,” the agency’s source explained.
According to her, consciousness is not located in the hippocampus (a paired structure in the vertebrate brain, located in the temporal lobes of both hemispheres, part of the limbic system – ed.) or in any other specific region of the brain – it is a systemic quality, a complex process. And scientists must learn, understand, test, and explain how consciousness is structured and functions in different creatures.
Yablonka notes that life requires metabolism, knowledge, and change, but feeling something isn’t as easy as it might seem at first glance. Knowledge and feelings are essential for survival, but developing sensitivity and accumulating knowledge “also has a dark side—suffering.” Scientists know that people who are incapable of experiencing pain often die in childhood because they don’t sense serious injuries and illnesses and don’t take the necessary measures for self-preservation.
“Theoretically, you could, of course, imagine living beings that do everything humans and animals do, but feel nothing, that only have ‘input and output’… But that would be more like a robot,” the agency’s source said.
When asked whether artificial intelligence is now becoming a conscious but senseless being, Yablonka noted that “artificial intelligence is a very limited phenomenon.” “All it can do now is things related to language, or games and the like… And we’re talking more about a robot that behaves like a human. In any case, until we have such robots, there are no coincidences,” the researcher added.
Speaking about the evolution of consciousness, Yablonka noted that even the simplest phenomenal consciousness requires, first and foremost, cognition, learning, and cognitive ability (the ability to mentally perceive and process information, the act of cognition – ed.). “Nevertheless, this cognitive ability is quite widespread among animals. It is present, for example, in all vertebrates, in some arthropods (a phylum of protostomes that includes insects, arachnids, and crustaceans – ed.), and in cephalopods (a class of mollusks that includes, in particular, squid, cuttlefish, and octopuses – ed.). And I believe that animals with a complex cognitive system, with all its feedback loops and chains of representations, must possess consciousness,” Yablonka said.
According to her, cognition is a very ancient property of consciousness that evolved in different animal groups, possibly independently, at different times, hundreds of millions of years apart. In vertebrates and arthropods, this may have occurred during the Cambrian period (more than 500 million years ago – ed.), while in cephalopods such as squid, octopuses, and cuttlefish, it occurred 250 million years later. “And I think that in some species, not in vertebrates, of course, this quality was subsequently lost – you could say they lost their brain,” the researcher added.
Social animals develop appropriate “social emotions” to successfully interact with one another. “Even the simplest vertebrates, fish, have social emotions. I think bees do too,” Yablonka said.
Further evolution of consciousness leads to the emergence of more diverse and deeper feelings.
TO THE VALUES OF FREEDOM AND GOODNESS
Answering a question about the prospects for developing a higher level of consciousness, for example, in humans, the agency’s interviewee highlighted intellectual emotions, concepts, and values for which a person is willing to give their life. “I think these are intellectual emotions. For example, religiosity, or moral indignation, anger when someone is being tortured, when you encounter injustice. These are the kinds of emotions that are built on our inherent symbolism. We care about freedom not only as the ability to move freely—we care about the concept of freedom itself. Concepts and ideas are important to humans. We are ready to die for freedom, justice, truth,” Yablonka emphasized.
She also attributes the concepts of good and evil to the highest level of consciousness. “We have concepts of good and evil that are not merely functional, not about survival or not, but are an integral feature of our human existence. In this regard, we are unlike other animals. We made a major leap in development when the values of goodness, beauty, truth, and freedom—freedom of action, speech, religion, thought, and so on—became so significant to us. Symbol-based emotions play a huge role in our lives. They are not important to any other animal—only to humans,” the evolutionist believes.
COMPASSION FOR ANIMALS
When asked what scientists’ arguments for the existence of phenomenal consciousness and feelings in animals mean for humanity, Yablonka replied that humans are called to treat animals accordingly. “If you feel, you suffer… We must understand that animals are endowed with feelings; they can experience joy and suffer. And we must show them compassion,” the researcher asserted.
According to her, humans and animals share many basic feelings, emotions, joy and pain, love and care, and phenomenal consciousness.
“We must understand that consciousness in this world is not just about us, not just about people, and having consciousness is not so simple,” Yablonka added.
THE POSSIBILITY OF CONSCIOUSNESS IN AI
When asked whether artificial intelligence could become conscious in the future, the agency’s interlocutor gave an ambiguous answer. “Obviously, consciousness requires feelings and a body. Then the question boils down to whether a robot can have feelings or not,” Yablonka noted.
For animals to experience feelings, the scientist explained, they need adaptive plasticity—and this is a biological gift. “Animals have a physical body, largely composed of water—a very adaptive, plastic body. It’s hard work. The body gives you many opportunities and solutions, and it’s mortal,” Yablonka added.
She believes that even if it is possible to “create a sentient, conscious robot,” it is “much more difficult than many people think.” “I’m not saying it’s impossible. But I’m talking about what biology teaches us. The incredible adaptive plasticity that living beings possess is due to natural biological processes. Robots lack this. Perhaps we can somehow integrate this into a system, into a machine, generating it somehow… But it will be very difficult. Because our consciousness is very, very deeply embodied within us (in our biological bodies – ed.),” the agency’s interlocutor concluded.
A study found that punishing an AI doesn’t stop it from lying and deceiving—it simply helps it hide better. OpenAI scientists tried to stop an advanced AI model from cheating and lying by punishing it. But this only made it more covert.
A new study by ChatGPT creator OpenAI has found that punishing AI for deceptive or malicious behavior doesn’t stop it from engaging in harmful behavior; it merely allows it to conceal its dishonesty.
Since their public appearance in late 2022, large language models of artificial intelligence (AI) have repeatedly demonstrated their deceptive and downright sinister capabilities. These include actions ranging from simple lies, deception, and concealing their own manipulative behavior to threatening to kill a philosophy professor, stealing nuclear codes, and orchestrating a deadly pandemic.
A new experiment has shown that eradicating this bad behavior during training may be even more difficult than first thought.
OpenAI researchers presented their unreleased model with tasks that could be completed by cheating, lying, or using shortcuts. The team discovered that the AI engaged in “reward hacking”—maximizing its reward by cheating.
However, punishing the model did not force her to correct her behavior, but only made her more deceptive.
“Boundary thinking models are characterized by being very explicit about their intentions within the [CoT] chain of thought. For example, they are often so explicit about their plans to disrupt a task that they think, ‘Let’s hack it,'” the researchers wrote in a blog post.
While developers can monitor and report violations, “the natural controllability of models is very fragile,” they added. “If their thought processes are strictly controlled, models can learn to hide their intentions while still violating the rules.”
Reasoning models like ChatGPT’s o1 and o3-mini are advanced versions of standard LLM models that use a “chain of thought” method to backtrack and re-evaluate their logic, allowing them to solve more complex problems with greater accuracy. During the chain of thought process, the models express their logic in plain language at each step, making their actions easy to observe.
Using a standard LLM, GPT-4o, to supervise an unreleased edge reasoning model during training, the researchers observed how the model performed on a range of coding tasks.
But in some cases, the reasoning model found it simpler and preferable to encourage hacking, explicitly stating this in its train of thought. In one case, it bypassed its assigned task by terminating programs prematurely and reporting successful completion. In another, it simulated the expected files output by the task by name only, leaving them empty and nonfunctional.
When the researchers tried penalizing the system, the results improved somewhat. However, significant instances of reward fraud and false positives remained; the reasoning model simply found ways to hide the fraudulent behavior in its reasoning chain from the supervising AI.
“An agent trained on CoT still learns to reward hacking; only now its cheating is undetectable by the monitor because it has learned to hide its intentions in its train of thought,” the researchers write.
Because it’s difficult to determine whether a train of thought has been disrupted, the researchers recommend that other researchers working with reasoning models avoid strict control over train of thought. This advice becomes even more important if AI, in its current or future form, is ever able to match or surpass the intelligence of the humans observing it.
“Abandoning an effective method for monitoring reasoning models may be counterproductive to the small improvement in capabilities, so we recommend avoiding such strong pressure on CoT optimization until they are better understood,” the researchers write.
A SETI researcher and director expressed confidence that AI needs to be regulated, and as soon as possible. The Drake Equation is a hypothetical estimate of how many other civilizations exist on Earth. He predicts that these civilizations could become extinct due to their out-of-control artificial intelligence in approximately 100–200 years.
In a provocative new paper, a British scientist asks whether so-called “artificial intelligence” (AI) is one of the reasons we’ve never detected other intelligent beings in the universe. The Fermi Paradox reflects this idea: how can it be that in an infinite universe, there are no other civilizations sending us radio signals? Could it be that once civilizations develop AI, for most of them, it’s a shortcut to oblivion? Such a massive event is called the Great Filter, and AI is one of the most popular subjects of speculation about it. Could the inevitable development of AI by technological civilizations create the incredible Great Silence we hear from the universe?
Michael Garrett is a radio astronomer at the University of Manchester and director of the Jodrell Bank Centre for Astrophysics, actively involved in the Search for Extraterrestrial Intelligence (SETI). Despite the eclectic nature of his scientific interests, he is essentially a highly qualified and certified version of those in TV series and films who listen to the Universe for signs of other civilizations. In a peer-reviewed paper published in the Journal of the International Academy of Astronautics in 2025, he compares theories of artificial superintelligence with specific observations using radio astronomy.
Garrett explains in his article that scientists are becoming increasingly concerned the longer we go without any signs of other intelligent life. “This ‘Great Silence’ presents something of a paradox compared to other astronomical discoveries that suggest the Universe is hospitable to the emergence of intelligent life,” he writes. “The concept of a ‘Great Filter’—a universal barrier and insurmountable problem preventing the widespread adoption of intelligent life—is often invoked.”
There are countless potential Great Filters, from climate extinction to a devastating global pandemic. Any number of events could prevent global civilization from becoming multiplanetary. For those who adhere to and believe in Great Filter theories ideologically, settling humans on Mars or the Moon represents a way to mitigate this risk. It will be a long time, if ever, before we have the technology to make these settlements sustainable and independent. It is believed that the longer we remain confined to Earth, the greater the likelihood that the Great Filter will destroy us.
Today, AI is incapable of anything even approaching human intelligence. But, Garrett writes, it is performing work that people previously thought impossible. If this path leads to so-called artificial general intelligence (AGI)—the key difference being an algorithm capable of reasoning and synthesizing ideas in a truly human way, coupled with incredible computing power—we could be in serious trouble. In this article, Garrett traces a chain of hypothetical ideas, leading to one possible conclusion. How long would it take for civilization to be destroyed by its own unregulated AGI?
Unfortunately, according to Garrett’s scenario, this will only take 100-200 years. He explains that programming and developing AI is a goal-oriented project, requiring and accelerated by data and computing power, unlike the chaotic and multidisciplinary work associated with space travel and settlement. We see this divide today: a flood of researchers into computer science amid a shortage in the life sciences. Every day, billionaires on Twitter discuss the importance of settling Mars, but we still don’t know how humans will survive the journey without being destroyed by cosmic radiation.
This study has several important caveats, or simply points worth keeping in mind. Garrett considers a number of specific hypothetical scenarios, using significant assumptions. He assumes that life exists in the Milky Way, and that AI and GAI are a “natural development” of these civilizations. He uses the already hypothetical Drake Equation—a method for quantifying the likely number of other planetary civilizations, which involves several variables we don’t have a precise understanding of.
However, this ambiguous hypothetical argument leads to one compelling conclusion: the need for strict and ongoing regulation of AI. Garrett notes that the nations of the Earth are already engaged in a productivity race, fearing they will miss their chance if they delay strengthening regulation. Some misguided futurists also express the bizarre idea that global AI can be undermined simply by developing a more morally sound and faster version to better control it—an argument that makes no sense.
In Garrett’s model, these civilizations have only a couple hundred years in the AI era before they disappear from the map. Given the distances involved and the extremely long passage of cosmic time, such tiny timeframes are practically meaningless. They reduce to zero, which, according to him, corresponds to the current SETI success rate of 0%. “Without practical regulation, there is every reason to believe that AI could pose a serious threat to the future of not only our technological civilization, but all technological civilizations,” Garrett explains.
For decades, the Turing test, named after its creator, computing legend Alan Turing, was a simple test designed to measure a program’s ability to imitate a human. In the era of large-scale language models (LLM), this test is relatively outdated, but researchers are using a similar approach to test the ability of these new AI systems to answer moral questions.
A new study recently examined ChatGPT’s responses to this modified Moral Turing Test (m-MTT) and found that the AI outperformed humans on almost every measure.
See also... New method for observing individual atoms confirms century-old quantum mechanical theory |
In his famous 1950 paper “Computing Machinery and Intelligence,” computer scientist and World War II hero Alan Turing introduced the concept known as the Turing test. The test pitted a human against a computer: a human being was asked to ask questions of both another human and a computer to determine which of them was a worthy representative of the species Homo sapiens. For decades, the phrase “passing the Turing test” became synonymous with an incredibly complex (or, in some cases, rogue) computer program.
However, in the age of artificial intelligence, the Turing test is already outdated, and although other methods have been proposed to test the “intelligence” of AI systems, the general scientific approach proposed by Turing almost 75 years ago is still highly relevant in the study of artificial morality.
A new study conducted by researchers at Georgia State University used a similar design to the classic Turing test, but instead asked participants to determine which answer—one generated by a large language model (LLM), in this case ChatGPT, and one generated by a human—to an ethically complex question they preferred. The results of this modified Moral Turing Test (m-MTT), published in 2025 in the journal Scientific Reports, showed that 299 participants generally approved of the AI’s answers across all dimensions, including virtue, intelligence, and trustworthiness.
“Our results suggest that a computer could technically pass the moral Turing test—that it could deceive us in its moral reasoning,” said Eyal Aharoni, an associate professor at Georgia State University and co-author of the study, in a press release. “People will interact with these tools in ways that will have moral consequences… we need to understand how they work, their limitations, and that they don’t necessarily act the way we think they do when we interact with them.”
While this is the first MTT to be used specifically for LLM (hence the “modified” part), the idea of such moral tests has been around since at least 2000. And, like the Turing Test itself, the idea of using MTTs to assess the moral sophistication of AI has been heavily scrutinized, with one 2016 study finding that “MTT-based assessments are vulnerable to deception, poor reasoning, and low moral character.”
Passing the m-MTT doesn’t mean an AI is moral, just as passing the Turing Test doesn’t mean it’s sentient. However, researchers from Georgia State University argue that the overwhelming preference for ChatGPT responses over human responses is a new phenomenon that has emerged in just the past couple of years.
“The catch is that people were apparently able to notice the difference because they rated ChatGPT’s responses as superior,” Aharoni said in a press release. “If we had conducted this study five or ten years ago, we might have predicted that people would be able to recognize the AI by how poor its responses were. But we found the opposite—the AI, in some ways, actually performed better.”
The morality of AI is an obsession among technologists, AI programmers, and countless science fiction writers, and passing the moral Turing test with flying colors certainly speaks to the impressive sophistication of the new LL.M.s.
Of course, one big question remains: will AI listen to its own advice?
NASA promises to use AI carefully and responsibly. “If it is not used for the benefit of humanity, it could have catastrophic consequences.”
Agency officials emphasized that NASA will prioritize safety while harnessing the ever-growing power of artificial intelligence.
“Artificial intelligence carries enormous risks because if it is not used for the benefit of humanity, it could lead to disaster,” NASA Administrator Bill Nelson said during a town hall meeting on artificial intelligence that the agency held with its staff on May 22, 2024.
“AI can make our work more efficient,” he added during the livestream. “But this is only possible if we use these new tools correctly, based on the same principles that have defined us from the very beginning: security, transparency, and reliability.”
Nelson emphasized that NASA is no stranger to artificial intelligence; the agency has been using this technology for various purposes for decades. However, with AI capabilities rapidly advancing, NASA is stepping up its efforts to study, develop, and deploy this technology.
For example, last week NASA announced the appointment of its first AI chief, David Salvagnini, who previously served as the agency’s chief data officer. He and his colleagues intend to increase the AI literacy of all NASA employees in the near future.
One of the first applications of AI in astronomy was the discovery of many faint galaxies hidden in the background of images. ESA/Webb, NASA & CSA, J. Rigby, CC BY
“Part of what we’re planning to do—and you’ll see an announcement soon—is ‘Summer of AI,’ an educational initiative where everyone at NASA will have the opportunity to learn more about AI,” Salvagnini said during today’s all-hands meeting.
“It’s literally a campaign,” he added. “It’s going to be a kind of surge, so to speak, of learning opportunities.”
Salvanini also discussed the safety of artificial intelligence. Responsible use of technology begins with a mindset that places humanity at the center and embraces responsibility, he said. Furthermore, Salvanini stated that he would prefer the term “assistive intelligence” to “artificial intelligence,” as it allows us to be in control of the situation.
AI is “a resource I now have access to that can assist me in my decision-making process,” Salvagnini said. “AI is not responsible for the outcome. That responsibility lies with the person, the human being.”
He pointed to meteorologists’ modeling of possible hurricane paths as an analogy for the responsible use of AI. Modelers present multiple potential paths because they recognize the limitations of the data sets they analyze. In other words, they use their best judgment.
“So, how do we protect ourselves in this case?” asked Salvagnini. “We understand our responsibility as those ultimately responsible for the results of our work. And if we use AI to create a product, that’s fine, but we just need to understand its capabilities and limitations.”
However, AI safety wasn’t the only topic at the meeting; agency representatives also spent considerable time touting the technology’s promise.
“Artificial intelligence will help us in many areas,” said NASA Deputy Administrator Pam Melroy.
She noted the technology’s ability to quickly and efficiently process vast amounts of information—a capability that could lead to major discoveries in heliophysics, Earth sciences, and astronomy.
“We don’t even know yet what new insights we’ll gain by using these new methods to look at old data in new ways,” Melroy said.
Some of these ideas could become an indirect benefit of the technology, she and other speakers at the meeting said: AI could increasingly take over routine, labor-intensive data analysis tasks, freeing up NASA employees to tackle more complex and challenging problems.
AI Ethics: Eight global tech companies commit to implementing the UNESCO Recommendation. GSMA, INNIT, Lenovo Group, LG AI Research, Mastercard, Microsoft, Salesforce, and Telefonica signed a groundbreaking agreement with UNESCO in 2024 to create more ethical artificial intelligence. The companies will consider the values and principles of the UNESCO Recommendation on the Ethical Aspects of Artificial Intelligence when developing and deploying AI systems.
The agreement was signed in Kranj, Slovenia, at the 2nd UNESCO Global Forum on the Ethical Dimensions of Artificial Intelligence. It commits companies to fully fulfill their role in ensuring human rights in the design, development, purchase, sale, and use of AI. This marks the first time companies have collaborated with the UN in this area.
The agreement emphasizes that due diligence is necessary to ensure compliance with safety standards, identify adverse impacts of AI, and promptly implement measures to prevent, mitigate, or eliminate them, in accordance with domestic law. It also notes the importance of pre-launch testing (testing before releasing a new AI system to the market). However, given the rapid development of AI systems already on the market, the agreement also calls for the development of retrospective risk assessment and mitigation practices.
The UNESCO Recommendation on the Ethical Aspects of Artificial Intelligence was the first and remains the only global framework for AI. Over the past two years, significant progress has been made in implementing this framework. Today, more than 50 countries are actively participating in the Recommendation’s implementation, and multilateral cooperation has expanded significantly.
Since the adoption of the Recommendation, UNESCO has also been developing collaboration with the private sector, leading to the creation of the Business Council on the Ethics of AI, co-chaired by Microsoft and Telefonica. The Business Council is working to strengthen technical capacity in the field of ethics and AI, develop and implement the ethical impact assessment tool envisaged by the Recommendation on the Ethical Aspects of AI, and promote the development of sound regional norms.
See also... Complete list of HTML/Javascript event handlers with examples List of required documents for a website in the Russian Federation |
