Artificial intelligence (AI) models like ChatGPT, Claude, and Gemini often give the impression that there’s a mind at work within the machine. These days they “think” in response to queries, go back and correct themselves, apologise for mistakes, and mimic many tics of human communication.

There’s no direct physical evidence to this day that a machine mind exists however. In fact, there’s good reason to believe what these machines are doing when they say they’re “thinking” is actually dealing with a physical phenomenon.

Also Read | At the last frontier of thought: will AI kill creativity?

In the 1980s, a group of physicists led among others by John Hopfield and Geoffrey Hinton realised that if you have a network with millions of neurons, you can stop treating them as individual ‘particles’ and start addressing them as a system. And the behaviour and properties of these systems can be described by the rules of thermodynamics and statistical mechanics.

Hopfield and Hinton won the physics Nobel Prize in 2024 for this work. A pair of studies published in Physical Review E has doubled down on the same idea, showing that two common ‘tricks’ engineers use to make AI models better are also such physical phenomena.

Achilles heel

A neural network is a network of processors connected to each other like neurons in the human brain and which learns and uses information like the brain. They can also be stacked in multiple layers, so that one layer prepares the inputs for the next and so on. Neural networks are at the heart of machine learning applications like generative AI, self-driving cars, computer vision, and modelling.

They also have an Achilles heel called overfitting: a network becomes so obsessed with some specific examples it has seen during its training that it fails to understand the broader patterns. Engineers have developed some techniques to prevent this. For instance, the October 2025 paper by University of Oxford and Princeton University researchers Francesco Mori and Francesca Mignacco focused on a technique called dropout. During training, the neural network is made to randomly turn off a certain percentage of its neurons, forcing the remaining ones to work harder and learn the concepts independently.

Abdulkadir Canatar and SueYeon Chung, of the Flatiron Institute and New York University, turned to a constraint called tolerance in their August paper. They analysed what happens when an AI is told to ignore any error that falls within a small range. So rather than trying to correct every little discrepancy, the network treats any answer that’s ‘close enough’ to be good enough.

While dropout and tolerance look like different programming choices, the authors of the two papers insisted (separately) that they’re both governed by the same underlying physical phenomena.

Teacher-student experiment

Both duos used a tool called the teacher-student framework to explain how. Teacher is a neural network that’s already familiar with a dataset while Student is a network that’s starting completely blank. The Student’s goal is to learn the same dataset until its internal settings are aligned with those of the Teacher.

Mori and Mignacco wrote that at first, the Student was stuck in an “unspecialised phase” when its neurons were all doing the same thing. In the authors’ mathematical models, this appeared as a plateau, or a flat line, in the error graph, and it denoted that the Student wasn’t learning.

So they argued that for the Student to become smarter, it must first undergo a “specialisation transition”. Physicists are familiar with such transitions because they use the same maths to describe liquid water turning into vapour, a process called a phase transition.

Mori and Mignacco reported that by randomly turning neurons off, dropout injected a certain amount of noise into the system, which then nudged the network out of its plateau and towards specialised intelligence — a phase transition. This description also aligns with the work of Hopfield and Hinton, who proved that the energy of a neural network is a real thing, by manipulating which the network can be made to perform better.

They even reported a formula that they said could find the supposedly ideal dropout rate: relating the activation probability, which is the chance that a neuron spits out a particular output for a given set of inputs, to the learning rate, noise level, and the learning capacities of the Teacher and Student networks.

Like atoms

Canatar and Chung also found that the consequences of changing the tolerance on the network could be described using the laws of physics, which they illustrated by applying their findings to the double-descent problem. When you give a network more data, its performance sometimes gets worse before it suddenly gets better. According to Canatar and Chung, when a network learns exactly as many examples as it has internal settings for, it reaches a point where it’s looking for more information. When it doesn’t get that information, it starts to overfit what it already ‘knows’ to every problem in its way.

The machine doesn’t reach this overfitting stage because its algorithm is flawed but because its millions of parameters are like a collection of atoms trying to go through a phase transition, and failing, they added. As a result, the results of the neurons’ computations are riddled with errors.

The solution? “Canatar and Chung uncovered a critical value of … tolerance that separates two regimes: one in which the neural network perfectly fits the training data and another in which overadaption is avoided. In physical terms, this regime crossover corresponds to a … phase transition,” Hugo Cui, a researcher at the University of Paris-Saclay and the French National Centre for Scientific Research, wrote in a commentary for Physics.

Some limitations

Mori and Mignacco were working with a two-layer neural network, which is like a toy model compared to the large, multi-layered deep learning networks that power AI models like ChatGPT or self-driving cars. Nonetheless, they’ve written that the “mechanisms” they’ve uncovered answer “several open questions about the mechanisms driving the performance improvement induced by dropout”.

Canatar and Chung on the other hand applied their equations to ResNet, an advanced type of neural network used to solve real-world problems like computer vision. They said that even in this setup, the same geometric and thermodynamic rules they’d found in their simpler model held true.

For decades now, engineers have often treated machine learning as a kind of ‘black box’, where they just tinker with the code until it works but without understanding why it works. In the 1980s, however, a conviction prevailed that machine intelligence is a complex product, but nonetheless a product, of statistical mechanics, which physicists understand very well. By this logic, the machine’s inner workings aren’t inscrutable so much as a physical system that can be deciphered using undergraduate physics.

These studies suggest a future where scientists could use analytic theories like the ones in the papers to estimate an AI model’s performance even before they turn it on.

mukunth.v@thehindu.co.in

Published – March 02, 2026 07:30 am IST

A panel discussion on ‘Building Innovation-First GCCs: AI, R&D & Digital Transformation’ was organised as part of BioAsia 2026 in Hyderabad on Wednesday (February 18, 2025)
| Photo Credit: Siddhant Thakur

Imagine attending a funeral where the person who has died speaks directly to you, answering your questions and sharing memories. This happened at the funeral of Marina Smith, a Holocaust educator who died in 2022.

Thanks to an AI technology company called StoryFile, Smith seemed to interact naturally with her family and friends.

The system used prerecorded answers combined with artificial intelligence to create a realistic, interactive experience. This wasn’t just a video; it was something closer to a real conversation, giving people a new way to feel connected to a loved one after they’re gone.

Virtual life after death

Technology has already begun to change how people think about life after death. Several technology companies are helping people manage their digital lives after they’re gone. For example, Apple, Google and Meta offer tools to allow someone you trust to access your online accounts when you die.

Microsoft has patented a system that can take someone’s digital data – such as texts, emails and social media posts – and use it to create a chatbot. This chatbot can respond in ways that sound like the original person.

In South Korea, a group of media companies took this idea even further. A documentary called “Meeting You” showed a mother reunited with her daughter through virtual reality. Using advanced digital imaging and voice technology, the mother was able to see and talk to her dead daughter as if she were really there.

These examples may seem like science fiction, but they’re real tools available today. As AI continues to improve, the possibility of creating digital versions of people after they die feels closer than ever.

Who owns your digital afterlife?

While the idea of a digital afterlife is fascinating, it raises some big questions. For example, who owns your online accounts after you die?

This issue is already being discussed in courts and by governments around the world. In the United States, nearly all states have passed laws allowing people to include digital accounts in their wills.

In Germany, courts ruled that Facebook had to give a deceased person’s family access to their account, saying that digital accounts should be treated as inheritable property, like a bank account or house.

But there are still plenty of challenges. For example, what if a digital clone of you says or does something online that you would never have said or done in real life? Who is responsible for what your AI version does?

When a deepfake of actor Bruce Willis appeared in an ad without his permission, it sparked a debate about how people’s digital likenesses can be controlled, or even exploited, for profit.

Cost is another issue. While some basic tools for managing digital accounts after death are free, more advanced services can be expensive. For example, creating an AI version of yourself might cost thousands of dollars, meaning that only wealthy people could afford to “live on” digitally. This cost barrier raises important questions about whether digital immortality could create new forms of inequality.

Grieving in a digital world

Losing someone is often painful, and in today’s world, many people turn to social media to feel connected to those they’ve lost. Research shows that a significant proportion of people maintain their social media connections with deceased loved ones.

But this new way of grieving comes with challenges. Unlike physical memories such as photos or keepsakes that fade over time, digital memories remain fresh and easily accessible. They can even appear unexpectedly in your social media feeds, bringing back emotions when you least expect them.

Some psychologists worry that staying connected to someone’s digital presence could make it harder for people to move on. This is especially true as AI technology becomes more advanced. Imagine being able to chat with a digital version of a loved one that feels almost real. While this might seem comforting, it could make it even harder for someone to accept their loss and let go.

Cultural and religious views on digital afterlife

Different cultures and religions have their own unique perspectives on digital immortality. For example:

1. The Vatican, the center of the Catholic Church, has said that digital legacies should always respect human dignity.
2. In Islamic traditions, scholars are discussing how digital remains fit into religious laws.
3. In Japan, some Buddhist temples are offering digital graveyards where families can preserve and interact with digital traces of their loved ones.

These examples show how technology is being shaped by different beliefs about life, death and remembrance. They also highlight the challenges of blending new innovations with long-standing cultural and religious traditions.

Planning your digital legacy

When you think about the future, you probably imagine what you want to achieve in life, not what will happen to your online accounts when you’re gone. But experts say it’s important to plan for your digital assets: everything from social media profiles and email accounts to digital photos, online bank accounts and even cryptocurrencies.

Adding digital assets to your will can help you decide how your accounts should be managed after you’re gone. You might want to leave instructions about who can access your accounts, what should be deleted and whether you’d like to create a digital version of yourself.

You can even decide if your digital self should “die” after a certain amount of time. These are questions that more and more people will need to think about in the future.

Here are steps you can take to control your digital afterlife:

• Decide on a digital legacy. Reflect on whether creating a digital self aligns with your personal, cultural or spiritual beliefs. Discuss your preferences with loved ones.

• Inventory and plan for digital assets. Make a list of all digital accounts, content and tools representing your digital self. Decide how these should be managed, preserved or deleted.

• Choose a digital executor. Appoint a trustworthy, tech-savvy person to oversee your digital assets and carry out your wishes. Clearly communicate your intentions with them.

• Ensure that your will covers your digital identity and assets. Specify how they should be handled, including storage, usage and ethical considerations. Include legal and financial aspects in your plan

• .Prepare for ethical and emotional impacts. Consider how your digital legacy might affect loved ones. Plan to avoid misuse, ensure funding for long-term needs, and align your decisions with your values

.Digital pyramids

Thousands of years ago, the Egyptian pharaohs had pyramids built to preserve their legacy. Today, our “digital pyramids” are much more advanced and broadly available. They don’t just preserve memories; they can continue to influence the world, long after we’re gone.

This article is republished from The Conversation under a Creative Commons license. Read the original article here.

The potential of using artificial intelligence in drug discovery and development has sparked both excitement and skepticism among scientists, investors and the general public.
| Photo Credit: Thom Leach/Science Photo Library via Getty Images

The potential of using artificial intelligence in drug discovery and development has sparked both excitement and skepticism among scientists, investors and the general public.

“Artificial intelligence is taking over drug development,” claim some companies and researchers. Over the past few years, interest in using AI to design drugs and optimize clinical trials has driven a surge in research and investment. AI-driven platforms like AlphaFold, which won the 2024 Nobel Prize for its ability to predict the structure of proteins and design new ones, showcase AI’s potential to accelerate drug development.

AI in drug discovery is “nonsense,” warn some industry veterans. They urge that “AI’s potential to accelerate drug discovery needs a reality check,” as AI-generated drugs have yet to demonstrate an ability to address the 90% failure rate of new drugs in clinical trials. Unlike the success of AI in image analysis, its effect on drug development remains unclear.

We have been following the use of AI in drug development in our work as a pharmaceutical scientist in both academia and the pharmaceutical industry and as a former program manager in the Defense Advanced Research Projects Agency, or DARPA. We argue that AI in drug development is not yet a game-changer, nor is it complete nonsense. AI is not a black box that can turn any idea into gold. Rather, we see it as a tool that, when used wisely and competently, could help address the root causes of drug failure and streamline the process.

Most work using AI in drug development intends to reduce the time and money it takes to bring one drug to market – currently 10 to 15 years and US$1 billion to $2 billion. But can AI truly revolutionize drug development and improve success rates?

AI in drug development

Researchers have applied AI and machine learning to every stage of the drug development process. This includes identifying targets in the body, screening potential candidates, designing drug molecules, predicting toxicity and selecting patients who might respond best to the drugs in clinical trials, among others.

Between 2010 and 2022, 20 AI-focused startups discovered 158 drug candidates, 15 of which advanced to clinical trials. Some of these drug candidates were able to complete preclinical testing in the lab and enter human trials in just 30 months, compared with the typical 3 to 6 years. This accomplishment demonstrates AI’s potential to accelerate drug development.

On the other hand, while AI platforms may rapidly identify compounds that work on cells in a Petri dish or in animal models, the success of these candidates in clinical trials – where the majority of drug failures occur – remains highly uncertain.

Unlike other fields that have large, high-quality datasets available to train AI models, such as image analysis and language processing, the AI in drug development is constrained by small, low-quality datasets. It is difficult to generate drug-related datasets on cells, animals or humans for millions to billions of compounds. While AlphaFold is a breakthrough in predicting protein structures, how precise it can be for drug design remains uncertain. Minor changes to a drug’s structure can greatly affect its activity in the body and thus how effective it is in treating disease.

Survivorship bias

Like AI, past innovations in drug development like computer-aided drug design, the Human Genome Project and high-throughput screening have improved individual steps of the process in the past 40 years, yet drug failure rates haven’t improved.

Most AI researchers can tackle specific tasks in the drug development process when provided with high-quality data and particular questions to answer. But they are often unfamiliar with the full scope of drug development, reducing challenges into pattern recognition problems and refinement of individual steps of the process. Meanwhile, many scientists with expertise in drug development lack training in AI and machine learning. These communication barriers can hinder scientists from moving beyond the mechanics of current development processes and identifying the root causes of drug failures.

Current approaches to drug development, including those using AI, may have fallen into a survivorship bias trap, overly focusing on less critical aspects of the process while overlooking major problems that contribute most to failure. This is analogous to repairing damage to the wings of aircraft returning from the battle fields in World War II while neglecting the fatal vulnerabilities in engines or cockpits of the planes that never made it back. Researchers often overly focus on how to improve a drug’s individual properties rather than the root causes of failure.

The current drug development process operates like an assembly line, relying on a checkbox approach with extensive testing at each step of the process. While AI may be able to reduce the time and cost of the lab-based preclinical stages of this assembly line, it is unlikely to boost success rates in the more costly clinical stages that involve testing in people. The persistent 90% failure rate of drugs in clinical trials, despite 40 years of process improvements, underscores this limitation.

Addressing root causes

Drug failures in clinical trials are not solely due to how these studies are designed; selecting the wrong drug candidates to test in clinical trials is also a major factor. New AI-guided strategies could help address both of these challenges.

Currently, three interdependent factors drive most drug failures: dosage, safety and efficacy. Some drugs fail because they’re too toxic, or unsafe. Other drugs fail because they’re deemed ineffective, often because the dose can’t be increased any further without causing harm.

We and our colleagues propose a machine learning system to help select drug candidates by predicting dosage, safety and efficacy based on five previously overlooked features of drugs. Specifically, researchers could use AI models to determine how specifically and potently the drug binds to known and unknown targets, the level of these targets in the body, how concentrated the drug becomes in healthy and diseased tissues, and the drug’s structural properties.

These features of AI-generated drugs could be tested in what we call phase 0+ trials, using ultra-low doses in patients with severe and mild disease. This could help researchers identify optimal drugs while reducing the costs of the current “test-and-see” approach to clinical trials.

While AI alone might not revolutionize drug development, it can help address the root causes of why drugs fail and streamline the lengthy process to approval.

This article is republished from The Conversation under a Creative Commons license. Read the original article here.

Geneva:

Humanity is in a race against time to harness the colossal emerging power of artificial intelligence for the good of all, while averting dire risks, a top UN official said Thursday.

“We’ve let the genie out of the bottle,” said Doreen Bogdan-Martin, head of the United Nations’ International Telecommunications Union (ITU).

“We are in a race against time,” she told the opening of a two-day AI for Good Global Summit in Geneva.

“Recent developments in AI have been nothing short of extraordinary.”

The thousands gathered at the conference heard how advances in generative AI are already speeding up efforts to solve some of the world’s most pressing problems, such as climate change, hunger and social care.

“I believe we have a once-in-a-generation opportunity to guide AI to benefit all the world’s people,” Bogdan-Martin told AFP in an email ahead of the summit.

But she lamented Thursday that one-third of humanity still remains completely offline, and is “excluded from the AI revolution without a voice”.

“This digital and technological divide is no longer acceptable.”

Bogdan-Martin highlighted that AI holds “immense potential for both good and bad”, stressing that it was vital to “make AI systems safe”.

She said that was especially important now, given that “2024 is the biggest election year in history”, with votes in dozens of countries, including in the United States.

And “with the rise of sophisticated deep fakes disinformation campaigns, it’s also the most contentious one,” she said.

“Not only does this misuse of AI threaten democracy, it also endangers young people’s mental health and compromises cyber-security.”

In an address to a separate event focused on AI governance this week, the ITU chief said that “the power of AI is concentrated in the hands of too few”.

Bogdan-Martin hailed that governments and others had become more focused on regulation and protections around the use of AI.

For instance, on Wednesday the European Union announced the creation of an AI Office to regulate artificial intelligence under a sweeping new law.

“It’s our responsibility to write the next chapter in the great story of humanity, and technology, and to make it safe, to make it inclusive and to make it sustainable,” Bogdan-Martin said.

(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)

Google DeepMind unveils next generation of drug discovery AI model.
| Photo Credit: AP

Google Deepmind has unveiled the third major version of its “AlphaFold” artificial intelligence model, designed to help scientists design drugs and target disease more effectively.

In 2020, the company made a significant advance in molecular biology by using AI to successfully predict the behaviour of microscopic proteins.

With the latest incarnation of AlphaFold, researchers at DeepMind and sister company Isomorphic Labs – both overseen by cofounder Demis Hassabis – have mapped the behaviour for all of life’s molecules, including human DNA.

The interactions of proteins – from enzymes crucial to the human metabolism, to the antibodies that fight infectious diseases – with other molecules is key to drug discovery and development.

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

DeepMind said the findings, published in research journal Nature on Wednesday, would reduce the time and money needed to develop potentially life-changing treatments.

“With these new capabilities, we can design a molecule that will bind to a specific place on a protein, and we can predict how strongly it will bind,” Hassabis said in a press briefing on Tuesday.

“It’s a critical step if you want to design drugs and compounds that will help with disease.”

The company also announced the release of the “AlphaFold server”, a free online tool that scientists can use to test their hypotheses before running real-world tests.

Since 2021, AlphaFold’s predictions have been freely accessible to non-commercial researchers, as part of a database containing more than 200 million protein structures, and has been cited thousands of times in others’ work.

DeepMind said the new server required less computing knowledge, allowing researchers to run tests with just a few clicks of a button.

John Jumper, a senior research scientist at DeepMind, said: “It’s going to be really important how much easier the AlphaFold server makes it for biologists – who are experts in biology, not computer science – to test larger, more complex cases.”

Dr Nicole Wheeler, an expert in microbiology at the University of Birmingham, said AlphaFold 3 could significantly speed up the drug discovery pipeline, as “physically producing and testing biological designs is a big bottleneck in biotechnology at the moment”.

A Copilot page showing the incorporation of AI technology is shown in London, Tuesday, February 13, 2024. Image for Representation.
| Photo Credit: AP

Given the huge problem-solving potential of artificial intelligence (AI), it wouldn’t be far-fetched to think that AI could also help us in tackling the climate crisis. However, when we consider the energy needs of AI models, it becomes clear that the technology is as much a part of the climate problem as a solution.

The emissions come from the infrastructure associated with AI, such as building and running the data centres that handle the large amounts of information required to sustain these systems.

But different technological approaches to how we build AI systems could help reduce its carbon footprint. Two technologies in particular hold promise for doing this: spiking neural networks and lifelong learning.

The lifetime of an AI system can be split into two phases: training and inference. During training, a relevant dataset is used to build and tune – improve – the system. In inference, the trained system generates predictions on previously unseen data.

For example, training an AI that’s to be used in self-driving cars would require a dataset of many different driving scenarios and decisions taken by human drivers.

After the training phase, the AI system will predict effective manoeuvres for a self-driving car. Artificial neural networks (ANN), are an underlying technology used in most current AI systems.

They have many different elements to them, called parameters, whose values are adjusted during the training phase of the AI system. These parameters can run to more than 100 billion in total.

While large numbers of parameters improve the capabilities of ANNs, they also make training and inference resource-intensive processes. To put things in perspective, training GPT-3 (the precursor AI system to the current ChatGPT) generated 502 metric tonnes of carbon, which is equivalent to driving 112 petrol powered cars for a year.

GPT-3 further emits 8.4 tonnes of CO₂ annually due to inference. Since the AI boom started in the early 2010s, the energy requirements of AI systems known as large language models (LLMs) – the type of technology that’s behind ChatGPT – have gone up by a factor of 300,000.

With the increasing ubiquity and complexity of AI models, this trend is going to continue, potentially making AI a significant contributor of CO₂ emissions. In fact, our current estimates could be lower than AI’s actual carbon footprint due to a lack of standard and accurate techniques for measuring AI-related emissions.

Spiking neural networks

The previously mentioned new technologies, spiking neural networks (SNNs) and lifelong learning (L2), have the potential to lower AI’s ever-increasing carbon footprint, with SNNs acting as an energy-efficient alternative to ANNs.

ANNs work by processing and learning patterns from data, enabling them to make predictions. They work with decimal numbers. To make accurate calculations, especially when multiplying numbers with decimal points together, the computer needs to be very precise. It is because of these decimal numbers that ANNs require lots of computing power, memory and time.

This means ANNs become more energy-intensive as the networks get larger and more complex. Both ANNs and SNNs are inspired by the brain, which contains billions of neurons (nerve cells) connected to each other via synapses.

Like the brain, ANNs and SNNs also have components which researchers call neurons, although these are artificial, not biological ones. The key difference between the two types of neural networks is in the way individual neurons transmit information to each other.

Neurons in the human brain communicate with each other by transmitting intermittent electrical signals called spikes. The spikes themselves do not contain information. Instead, the information lies in the timing of these spikes. This binary, all-or-none characteristic of spikes (usually represented as 0 or 1) implies that neurons are active when they spike and inactive otherwise.

This is one of the reasons for energy efficient processing in the brain.

Just as Morse code uses specific sequences of dots and dashes to convey messages, SNNs use patterns or timings of spikes to process and transmit information. So, while the artificial neurons in ANNs are always active, SNNs consume energy only when a spike occurs.

Otherwise, they have closer to zero energy requirements. SNNs can be up to 280 times more energy efficient than ANNs.

My colleagues and I are developing learning algorithms for SNNs that may bring them even closer to the energy efficiency exhibited by the brain. The lower computational requirements also imply that SNNs might be able to make decisions more quickly.

These properties render SNNs useful for broad range of applications, including space exploration, defence and self-driving cars because of the limited energy sources available in these scenarios.

Lifelong learning

L2 is another strategy for reducing the overall energy requirements of ANNs over the course of their lifetime that we are also working on.

Training ANNs sequentially (where the systems learn from sequences of data) on new problems causes them to forget their previous knowledge while learning new tasks. ANNs require retraining from scratch when their operating environment changes, further increasing AI-related emissions.

L2 is a collection of algorithms that enable AI models to be trained sequentially on multiple tasks with little or no forgetting. L2 enables models to learn throughout their lifetime by building on their existing knowledge without having to retrain them from scratch.

The field of AI is growing fast and other potential advancements are emerging that can mitigate the energy demands of this technology. For instance, building smaller AI models that exhibit the same predictive capabilities as that of a larger model.

Advances in quantum computing – a different approach to building computers that harnesses phenomena from the world of quantum physics – would also enable faster training and inference using ANNs and SNNs. The superior computing capabilities offered by quantum computing could allow us to find energy-efficient solutions for AI at a much larger scale.

The climate change challenge requires that we try to find solutions for rapidly advancing areas such as AI before their carbon footprint becomes too large.

The nation’s biggest technology executives on Wednesday loosely endorsed the idea of government regulations for artificial intelligence at an unusual closed-door meeting in the US Senate. But there is little consensus on what regulation would look like, and the political path for legislation is difficult.

Senate Majority Leader Chuck Schumer, who organized the private forum on Capitol Hill as part of a push to legislate artificial intelligence, said he asked everyone in the room — including almost two dozen tech executives, advocates and skeptics — whether government should have a role in the oversight of artificial intelligence, and “every single person raised their hands, even though they had diverse views,” he said.

Among the ideas discussed was whether there should be an independent agency to oversee certain aspects of the rapidly-developing technology, how companies could be more transparent and how the United States can stay ahead of China and other countries.

“The key point was really that it’s important for us to have a referee,” said Elon Musk, CEO of Tesla and X, during a break in the daylong forum. “It was a very civilized discussion, actually, among some of the smartest people in the world.”

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

Schumer will not necessarily take the tech executives’ advice as he works with colleagues on the politically difficult task of ensuring some oversight of the burgeoning sector. But he invited them to the meeting in hopes that they would give senators some realistic direction for meaningful regulation.

Congress should do what it can to maximize AI’s benefits and minimize the negatives, Schumer said, “whether that’s enshrining bias, or the loss of jobs, or even the kind of doomsday scenarios that were mentioned in the room. And only government can be there to put in guardrails.”

Other executives attending the meeting were Meta’s Mark Zuckerberg, former Microsoft CEO Bill Gates and Google CEO Sundar Pichai. Musk said the meeting “might go down in history as being very important for the future of civilization.” First, though, lawmakers have to agree on whether to regulate, and how.

Congress has a lackluster track record when it comes to regulating new technology, and the industry has grown mostly unchecked by government in the past several decades. Many lawmakers point to the failure to pass any legislation surrounding social media, such as for stricter privacy standards.

Schumer, who has made AI one of his top issues as leader, said regulation of artificial intelligence will be “one of the most difficult issues we can ever take on,” and he listed some of the reasons why: It’s technically complicated, it keeps changing and it “has such a wide, broad effect across the whole world,” he said.

Sparked by the release of ChatGPT less than a year ago, businesses have been clamoring to apply new generative AI tools that can compose human-like passages of text, program computer code and create novel images, audio and video. The hype over such tools has accelerated worries over its potential societal harms and prompted calls for more transparency in how the data behind the new products is collected and used.

Republican Sen. Mike Rounds of South Dakota, who led the meeting with Schumer, said Congress needs to get ahead of fast-moving AI by making sure it continues to develop “on the positive side” while also taking care of potential issues surrounding data transparency and privacy.

“AI is not going away, and it can do some really good things or it can be a real challenge,” Rounds said.

The tech leaders and others outlined their views at the meeting, with each participant getting three minutes to speak on a topic of their choosing. Schumer and Rounds then led a group discussion.

During the discussion, according to attendees who spoke about it, Musk and former Google CEO Eric Schmidt raised existential risks posed by AI, and Zuckerberg brought up the question of closed vs. “open source” AI models. Gates talked about feeding the hungry. IBM CEO Arvind Krishna expressed opposition to proposals favored by other companies that would require licenses.

In terms of a potential new agency for regulation, “that is one of the biggest questions we have to answer and that we will continue to discuss,” Schumer said. Musk said afterward he thinks the creation of a regulatory agency is likely.

Outside the meeting, Google CEO Pichai declined to give details about specifics but generally endorsed the idea of Washington involvement.

“I think it’s important that government plays a role, both on the innovation side and building the right safeguards, and I thought it was a productive discussion,” he said.

Some senators were critical that the public was shut out of the meeting, arguing that the tech executives should testify in public.

Sen. Josh Hawley, R-Mo., said he would not attend what he said was a “giant cocktail party for big tech.” Hawley has introduced legislation with Sen. Richard Blumenthal, D-Conn., to require tech companies to seek licenses for high-risk AI systems.

“I don’t know why we would invite all the biggest monopolists in the world to come and give Congress tips on how to help them make more money and then close it to the public,” Hawley said.

While civil rights and labor groups were also represented at the meeting, some experts worried that Schumer’s event risked emphasizing the concerns of big firms over everyone else.

Sarah Myers West, managing director of the nonprofit AI Now Institute, estimated that the combined net worth of the room Wednesday was $550 billion and it was “hard to envision a room like that in any way meaningfully representing the interests of the broader public.” She did not attend.

In the United States, major tech companies have expressed support for AI regulations, though they don’t necessarily agree on what that means. Similarly, members of Congress agree that legislation is needed, but there is little consensus on what to do.

There is also division, with some members of Congress worrying more about overregulation of the industry while others are concerned more about the potential risks. Those differences often fall along party lines.

“I am involved in this process in large measure to ensure that we act, but we don’t act more boldly or over-broadly than the circumstances require,” Young said. “We should be skeptical of government, which is why I think it’s important that you got Republicans at the table.”

Some concrete proposals have already been introduced, including legislation by Sen. Amy Klobuchar, D-Minn., that would require disclaimers for AI-generated election ads with deceptive imagery and sounds. Schumer said they discussed “the need to do something fairly immediate” before next year’s presidential election.

Hawley and Blumenthal’s broader approach would create a government oversight authority with the power to audit certain AI systems for harms before granting a license.

Some of those invited to Capitol Hill, such as Musk, have voiced dire concerns evoking popular science fiction about the possibility of humanity losing control to advanced AI systems if the right safeguards are not in place. But the only academic invited to the forum, Deborah Raji, a University of California, Berkeley researcher who has studied algorithmic bias, said she tried to emphasize real-world harms already occurring.

“There was a lot of care to make sure the room was a balanced conversation, or as balanced as it could be” Raji said. What remains to be seen, she said, is which voices senators will listen to and what priorities they elevate as they work to pass new laws.

Some Republicans have been wary of following the path of the European Union, which signed off in June on the world’s first set of comprehensive rules for artificial intelligence. The EU’s AI Act will govern any product or service that uses an AI system and classify them according to four levels of risk, from minimal to unacceptable.

A group of European corporations has called on EU leaders to rethink the rules, arguing that it could make it harder for companies in the 27-nation bloc to compete with rivals overseas in the use of generative AI.

Generative AI programs burst into the spotlight late last year. (File)
| Photo Credit: REUTERS

The United Nations called on Thursday for strict rules on the use of AI tools such as viral chatbot ChatGPT in classrooms, including limiting their use to older children.

In new guidance for governments, the UN’s education body UNESCO warned that public authorities were not ready to deal with the ethical issues of rolling out “generative” Artificial Intelligence programs in schools.

The Paris-based body said relying on such programs rather than human teachers could affect a child’s emotional wellbeing and leave them vulnerable to manipulation.

“Generative AI can be a tremendous opportunity for human development, but it can also cause harm and prejudice,” said Audrey Azoulay of UNESCO.

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

“It cannot be integrated into education without public engagement, and the necessary safeguards and regulations from governments.”

Generative AI programs burst into the spotlight late last year, with ChatGPT demonstrating an ability to generate essays, poems and conversations from the briefest prompts.

It sparked fears of plagiarism and cheating in schools and universities.

But investors poured money into the field and boosters targeted education as a possible lucrative market.

The UNESCO guidance said AI tools have the potential to help children with special needs, act as an opponent in “Socratic dialogues” or as a research assistant.

But the tools would only be safe and effective if teachers, learners and researchers helped to design them and governments regulated their use.

The guidance stopped short of recommending a minimum age for schoolchildren but pointed out that ChatGPT had a lower age limit of 13.

“Many commentators understand this threshold to be too young and have advocated for legislation to raise the age to 16,” said the guidance.