InvestMacro

By Michael Timothy Bennett, Australian National University and Elija Perrier, Stanford University 

A new artificial intelligence (AI) model has just achieved human-level results on a test designed to measure “general intelligence”.

On December 20, OpenAI’s o3 system scored 85% on the ARC-AGI benchmark, well above the previous AI best score of 55% and on par with the average human score. It also scored well on a very difficult mathematics test.

Creating artificial general intelligence, or AGI, is the stated goal of all the major AI research labs. At first glance, OpenAI appears to have at least made a significant step towards this goal.

While scepticism remains, many AI researchers and developers feel something just changed. For many, the prospect of AGI now seems more real, urgent and closer than anticipated. Are they right?

Generalisation and intelligence

To understand what the o3 result means, you need to understand what the ARC-AGI test is all about. In technical terms, it’s a test of an AI system’s “sample efficiency” in adapting to something new – how many examples of a novel situation the system needs to see to figure out how it works.

An AI system like ChatGPT (GPT-4) is not very sample efficient. It was “trained” on millions of examples of human text, constructing probabilistic “rules” about which combinations of words are most likely.

The result is pretty good at common tasks. It is bad at uncommon tasks, because it has less data (fewer samples) about those tasks.

Until AI systems can learn from small numbers of examples and adapt with more sample efficiency, they will only be used for very repetitive jobs and ones where the occasional failure is tolerable.

The ability to accurately solve previously unknown or novel problems from limited samples of data is known as the capacity to generalise. It is widely considered a necessary, even fundamental, element of intelligence.

Grids and patterns

The ARC-AGI benchmark tests for sample efficient adaptation using little grid square problems like the one below. The AI needs to figure out the pattern that turns the grid on the left into the grid on the right.

Each question gives three examples to learn from. The AI system then needs to figure out the rules that “generalise” from the three examples to the fourth.

These are a lot like the IQ tests sometimes you might remember from school.

Weak rules and adaptation

We don’t know exactly how OpenAI has done it, but the results suggest the o3 model is highly adaptable. From just a few examples, it finds rules that can be generalised.

To figure out a pattern, we shouldn’t make any unnecessary assumptions, or be more specific than we really have to be. In theory, if you can identify the “weakest” rules that do what you want, then you have maximised your ability to adapt to new situations.

What do we mean by the weakest rules? The technical definition is complicated, but weaker rules are usually ones that can be described in simpler statements.

In the example above, a plain English expression of the rule might be something like: “Any shape with a protruding line will move to the end of that line and ‘cover up’ any other shapes it overlaps with.”

Searching chains of thought?

While we don’t know how OpenAI achieved this result just yet, it seems unlikely they deliberately optimised the o3 system to find weak rules. However, to succeed at the ARC-AGI tasks it must be finding them.

We do know that OpenAI started with a general-purpose version of the o3 model (which differs from most other models, because it can spend more time “thinking” about difficult questions) and then trained it specifically for the ARC-AGI test.

French AI researcher Francois Chollet, who designed the benchmark, believes o3 searches through different “chains of thought” describing steps to solve the task. It would then choose the “best” according to some loosely defined rule, or “heuristic”.

This would be “not dissimilar” to how Google’s AlphaGo system searched through different possible sequences of moves to beat the world Go champion.

You can think of these chains of thought like programs that fit the examples. Of course, if it is like the Go-playing AI, then it needs a heuristic, or loose rule, to decide which program is best.

There could be thousands of different seemingly equally valid programs generated. That heuristic could be “choose the weakest” or “choose the simplest”.

However, if it is like AlphaGo then they simply had an AI create a heuristic. This was the process for AlphaGo. Google trained a model to rate different sequences of moves as better or worse than others.

What we still don’t know

The question then is, is this really closer to AGI? If that is how o3 works, then the underlying model might not be much better than previous models.

The concepts the model learns from language might not be any more suitable for generalisation than before. Instead, we may just be seeing a more generalisable “chain of thought” found through the extra steps of training a heuristic specialised to this test. The proof, as always, will be in the pudding.

Almost everything about o3 remains unknown. OpenAI has limited disclosure to a few media presentations and early testing to a handful of researchers, laboratories and AI safety institutions.

Truly understanding the potential of o3 will require extensive work, including evaluations, an understanding of the distribution of its capacities, how often it fails and how often it succeeds.

When o3 is finally released, we’ll have a much better idea of whether it is approximately as adaptable as an average human.

If so, it could have a huge, revolutionary, economic impact, ushering in a new era of self-improving accelerated intelligence. We will require new benchmarks for AGI itself and serious consideration of how it ought to be governed.

If not, then this will still be an impressive result. However, everyday life will remain much the same.

About the Author:

Michael Timothy Bennett, PhD Student, School of Computing, Australian National University and Elija Perrier, Research Fellow, Stanford Center for Responsible Quantum Technology, Stanford University

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

I research the intersection of artificial intelligence, natural language processing and human reasoning as the director of the Advancing Human and Machine Reasoning lab at the University of South Florida. I am also commercializing this research in an AI startup that provides a vulnerability scanner for language models.

From my vantage point, I observed significant developments in the field of AI language models in 2024, both in research and the industry.

Perhaps the most exciting of these are the capabilities of smaller language models, support for addressing AI hallucination, and frameworks for developing AI agents.

Small AIs make a splash

At the heart of commercially available generative AI products like ChatGPT are large language models, or LLMs, which are trained on vast amounts of text and produce convincing humanlike language. Their size is generally measured in parameters, which are the numerical values a model derives from its training data. The larger models like those from the major AI companies have hundreds of billions of parameters.

There is an iterative interaction between large language models and smaller language models, which seems to have accelerated in 2024.

First, organizations with the most computational resources experiment with and train increasingly larger and more powerful language models. Those yield new large language model capabilities, benchmarks, training sets and training or prompting tricks. In turn, those are used to make smaller language models – in the range of 3 billion parameters or less – which can be run on more affordable computer setups, require less energy and memory to train, and can be fine-tuned with less data.

No surprise, then, that developers have released a host of powerful smaller language models – although the definition of small keeps changing: Phi-3 and Phi-4 from Microsoft, Llama-3.2 1B and 3B, and Qwen2-VL-2B are just a few examples.

These smaller language models can be specialized for more specific tasks, such as rapidly summarizing a set of comments or fact-checking text against a specific reference. They can work with their larger cousins to produce increasingly powerful hybrid systems.

Wider access

Increased access to highly capable language models large and small can be a mixed blessing. As there were many consequential elections around the world in 2024, the temptation for the misuse of language models was high.

Language models can give malicious users the ability to generate social media posts and deceptively influence public opinion. There was a great deal of concern about this threat in 2024, given that it was an election year in many countries.

And indeed, a robocall faking President Joe Biden’s voice asked New Hampshire Democratic primary voters to stay home. OpenAI had to intervene to disrupt over 20 operations and deceptive networks that tried to use its models for deceptive campaigns. Fake videos and memes were created and shared with the help of AI tools.

Despite the anxiety surrounding AI disinformation, it is not yet clear what effect these efforts actually had on public opinion and the U.S. election. Nevertheless, U.S. states passed a large amount of legislation in 2024 governing the use of AI in elections and campaigns.

Misbehaving bots

Google started including AI overviews in its search results, yielding some results that were hilariously and obviously wrong – unless you enjoy glue in your pizza. However, other results may have been dangerously wrong, such as when it suggested mixing bleach and vinegar to clean your clothes.

Large language models, as they are most commonly implemented, are prone to hallucinations. This means that they can state things that are false or misleading, often with confident language. Even though I and others continually beat the drum about this, 2024 still saw many organizations learning about the dangers of AI hallucination the hard way.

Despite significant testing, a chatbot playing the role of a Catholic priest advocated for baptism via Gatorade. A chatbot advising on New York City laws and regulations incorrectly said it was “legal for an employer to fire a worker who complains about sexual harassment, doesn’t disclose a pregnancy or refuses to cut their dreadlocks.” And OpenAI’s speech-capable model forgot whose turn it was to speak and responded to a human in her own voice.

Fortunately, 2024 also saw new ways to mitigate and live with AI hallucinations. Companies and researchers are developing tools for making sure AI systems follow given rules pre-deployment, as well as environments to evaluate them. So-called guardrail frameworks inspect large language model inputs and outputs in real time, albeit often by using another layer of large language models.

And the conversation on AI regulation accelerated, causing the big players in the large language model space to update their policies on responsibly scaling and harnessing AI.

But although researchers are continually finding ways to reduce hallucinations, in 2024, research convincingly showed that AI hallucinations are always going to exist in some form. It may be a fundamental feature of what happens when an entity has finite computational and information resources. After all, even human beings are known to confidently misremember and state falsehoods from time to time.

The rise of agents

Large language models, particularly those powered by variants of the transformer architecture, are still driving the most significant advances in AI. For example, developers are using large language models to not only create chatbots, but to serve as the basis of AI agents. The term “agentic AI” shot to prominence in 2024, with some pundits even calling it the third wave of AI.

To understand what an AI agent is, think of a chatbot expanded in two ways: First, give it access to tools that provide the ability to take actions. This might be the ability to query an external search engine, book a flight or use a calculator. Second, give it increased autonomy, or the ability to make more decisions on its own.

For example, a travel AI chatbot might be able to perform a search of flights based on what information you give it, but a tool-equipped travel agent might plan out an entire trip itinerary, including finding events, booking reservations and adding them to your calendar.

In 2024, new frameworks for developing AI agents emerged. Just to name a few, LangGraph, CrewAI, PhiData and AutoGen/Magentic-One were released or improved in 2024.

Companies are just beginning to adopt AI agents. Frameworks for developing AI agents are new and rapidly evolving. Furthermore, security, privacy and hallucination risks are still a concern.

But global market analysts forecast this to change: 82% of organizations surveyed plan to use agents within 1-3 years, and 25% of all companies currently using generative AI are likely to adopt AI agents in 2025.

About the Author:

John Licato, Associate Professor of Computer Science, Director of AMHR Lab, University of South Florida

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

By Tamilla Triantoro, Quinnipiac University 

Online shopping often involves endless options and fleeting discounts. A single search for running shoes can yield hundreds of results across multiple platforms, each promising the “best deal.” The holiday season brings excitement, but it also brings a blend of decision fatigue and logistical nightmares.

What if there were a tool capable of hunting for the best prices, navigating endless sales and making sure your purchases arrive on time?

The next evolution in artificial intelligence is AI agents that are capable of autonomous reasoning and multistep problem-solving. AI shopping agents not only suggest what you might like, but they can also act on your behalf. Major retailers and AI companies are developing AI shopping assistants, and the AI company Perplexity released Buy with Pro on Nov. 18, 2024.

Picture this: You prompt AI to find a winter coat under $200 that’s highly rated and will arrive by Sunday. In seconds, it scans websites, compares prices, checks reviews, confirms availability and places the order, all while you go about your day.

Unlike traditional recommendation engines, AI agents learn your preferences and handle tasks autonomously. The agents are built with machine learning and natural language processing. They learn from their interactions with the people using them and become smarter and more efficient over time from their collective interactions.

Looking ahead, AI agents are likely to not only master personal shopping needs but also negotiate directly with corporate AI systems. They will not only learn your preferences but will likely be able to book tailored experiences, handle payments across platforms and coordinate schedules.

As a researcher who studies human-AI collaboration, I see how AI agents could make the future of shopping virtually effortless and more personalized than ever.

How AI agents help shoppers

Marketplaces such as Amazon and Walmart have been using AI to automate shopping. Google Lens offers a visual search tool for finding products.

Perplexity’s Buy with Pro is a more powerful AI shopping agent. By providing your shipping and billing information, you can place orders directly on the Perplexity app with free shipping on every order. The shopping assistant is part of the company’s Perplexity Pro service, which has free and paid tiers.

For those looking to build custom AI shopping agents, AutoGPT and AgentGPT are open-source tools for configuring and deploying AI agents.

Consumers today are focused on value, looking for deals and comparing prices across platforms. Having an assistant perform these tasks could be a tremendous time saver. But can AI truly learn your preferences?

A recent study using the GPT-4o model achieved 85% accuracy in imitating the thoughts and behaviors of over 1,000 people after they interacted with the AI for just two hours. This breakthrough finding suggests that digital personas can understand and act on people’s preferences in ways that will transform the shopping experience.

How AI shopping reshapes business

AI agents are moving beyond recommendations to autonomously executing complex tasks such as automating refunds, managing inventory and approving pricing decisions. This evolution has already begun to reshape how businesses operate and how consumers interact with them.

Retailers using AI agents are seeing measurable benefits. Since October 2024, data from the Salesforce shopping index reveals that digital retailers using generative AI achieved a 7% increase in average order revenue and attributed 17% of global orders to AI-driven personalized recommendations, targeted promotions and improved customer service.

Meanwhile, the nature of search and advertising is undergoing a major shift. Amazon is capturing billions of dollars in ad revenue as shoppers bypass Google to search directly on its platform. Simultaneously, AI-powered search tools such as Perplexity and OpenAI’s web-enabled chat deliver instant, context-aware responses, challenging traditional search engines and forcing advertisers to rethink their strategies.

The outcome of the battle between Big Tech and open-source initiatives to shape the AI ecosystem is also likely to affect how the shopping experience changes.

The risks: Privacy, manipulation and dependency

While AI agents offer significant benefits, they also raise critical privacy concerns. AI systems require extensive access to personal data, shopping history and financial information. This level of access increases the risk of misuse and unauthorized sharing.

Manipulation is another issue. AI can be highly persuasive and may be optimized to serve corporate interests over consumer welfare. Such technology can prioritize upselling or nudging shoppers toward higher-margin products under the guise of personalization.

There’s also the risk of dependency. Automating many aspects of shopping could diminish the satisfaction of making choices. Research in human-AI interaction indicates that while AI tools can reduce cognitive load, increased reliance on AI could impair people’s ability to critically evaluate their options.

What’s next?

AI-based shopping is still in its infancy, so how much trust should you place in it?

In our book “Converging Minds,” AI researcher Aleksandra Przegalinska and I argue for a balanced and critical approach to AI adoption, recognizing both its potential and its pitfalls.

As cognitive scientist Gary Marcus points out, AI’s moral limitations stem from technical constraints: Despite efforts to prevent errors, these systems remain imperfect.

This cautious perspective is reflected in the responses from my MBA class. When I asked students whether they were ready to outsource their holiday shopping to AI, the answer was an overwhelming no. Ethan Mollick, a professor at the Wharton School at the University of Pennsylvania, has argued that the adoption of AI in everyday life will be gradual, as societal change typically lags behind technological advancement.

Before people are willing to hand over their credit cards and let AI take the reins, businesses will have to ensure that AI systems align with human values and priorities. The promise of AI is vast, but to fulfill that promise I believe that AI will need to be an extension of human intention – not a replacement for it.

About the Author:

Tamilla Triantoro, Associate Professor of Business Analytics and Information Systems, Quinnipiac University

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

By Lauren Labrecque, University of Rhode Island 

Artificial intelligence is revolutionizing the way companies market their products, enabling them to target consumers in personalized and interactive ways that not long ago seemed like the realm of science fiction.

Marketers use AI-powered algorithms to scour vast amounts of data that reveals individual preferences with unrivaled accuracy. This allows companies to precisely target content – ads, emails, social media posts – that feels tailor-made and helps cultivate companies’ relationships with consumers.

As a researcher who studies technology in marketing, I joined several colleagues in conducting new research that shows AI marketing overwhelmingly neglects its potential negative consequences.

Our peer-reviewed study reviewed 290 articles that had been published over the past 10 years from 15 high-ranking marketing journals. We found that only 33 of them addressed the potential “dark side” of AI marketing.

This matters because the imbalance creates a critical gap in understanding the full impact of AI.

AI marketing can perpetuate harmful stereotypes, such as producing hypersexualized depictions of women, for example. AI can also infringe on the individual rights of artists. And it can spread misinformation through deepfakes and “hallucinations,” which occur when AI presents false information as if it were true, such as inventing historical events.

It can also negatively affect mental health. The prevalence of AI-powered beauty filters on social media, for instance, can foster unrealistic ideals and trigger depression.

These concerns loom large, prompting anxiety about the potential misuse of this powerful technology. Many people experience these worries, but young women are notably vulnerable. As AI apps gain acceptance, beauty standards are moving further from reality.

Our research finds there is an urgent need to address AI’s ethical considerations and potential negative consequences. Our intent is not to discredit AI. It’s to make sure that AI marketing benefits everyone, not just a handful of powerful companies.

I believe researchers should consider exploring the ethical problems with AI more thoroughly, and how to use it safely and responsibly.

This is important because AI is suddenly being used everywhere – from social media to self-driving cars to making health decisions. Understanding its potential negative effects empowers the public to be informed consumers and call for responsible AI use.

About the Author:

Lauren Labrecque, Professor of Marketing, University of Rhode Island

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

By Jaeyeon Chung, Rice University 

Think back to a time when you needed a quick answer, maybe for a recipe or a DIY project. A few years ago, most people’s first instinct was to “Google it.” Today, however, many people are more likely to reach for ChatGPT, OpenAI’s conversational AI, which is changing the way people look for information.

Rather than simply providing lists of websites, ChatGPT gives more direct, conversational responses. But can ChatGPT do more than just answer straightforward questions? Can it actually help people be more creative?

I study new technologies and consumer interaction with social media. My colleague Byung Lee and I set out to explore this question: Can ChatGPT genuinely assist people in creatively solving problems, and does it perform better at this than traditional search engines like Google?

Across a series of experiments in a study published in the journal Nature Human Behavour, we found that ChatGPT does boost creativity, especially in everyday, practical tasks. Here’s what we learned about how this technology is changing the way people solve problems, brainstorm ideas and think creatively.

ChatGPT and creative tasks

Imagine you’re searching for a creative gift idea for a teenage niece. Previously, you might have googled “creative gifts for teens” and then browsed articles until something clicked. Now, if you ask ChatGPT, it generates a direct response based on its analysis of patterns across the web. It might suggest a custom DIY project or a unique experience, crafting the idea in real time.

To explore whether ChatGPT surpasses Google in creative thinking tasks, we conducted five experiments where participants tackled various creative tasks. For example, we randomly assigned participants to either use ChatGPT for assistance, use Google search, or generate ideas on their own. Once the ideas were collected, external judges, unaware of the participants’ assigned conditions, rated each idea for creativity. We averaged the judges’ scores to provide an overall creativity rating.

One task involved brainstorming ways to repurpose everyday items, such as turning an old tennis racket and a garden hose into something new. Another asked participants to design an innovative dining table. The goal was to test whether ChatGPT could help people come up with more creative solutions compared with using a web search engine or just their own imagination.

The results were clear: Judges rated ideas generated with ChatGPT’s assistance as more creative than those generated with Google searches or without any assistance. Interestingly, ideas generated with ChatGPT – even without any human modification – scored higher in creativity than those generated with Google.

One notable finding was ChatGPT’s ability to generate incrementally creative ideas: those that improve or build on what already exists. While truly radical ideas might still be challenging for AI, ChatGPT excelled at suggesting practical yet innovative approaches. In the toy-design experiment, for example, participants using ChatGPT came up with imaginative designs, such as turning a leftover fan and a paper bag into a wind-powered craft.

Limits of AI creativity

ChatGPT’s strength lies in its ability to combine unrelated concepts into a cohesive response. Unlike Google, which requires users to sift through links and piece together information, ChatGPT offers an integrated answer that helps users articulate and refine ideas in a polished format. This makes ChatGPT promising as a creativity tool, especially for tasks that connect disparate ideas or generate new concepts.

It’s important to note, however, that ChatGPT doesn’t generate truly novel ideas. It recognizes and combines linguistic patterns from its training data, subsequently generating outputs with the most probable sequences based on its training. If you’re looking for a way to make an existing idea better or adapt it in a new way, ChatGPT can be a helpful resource. For something groundbreaking, though, human ingenuity and imagination are still essential.

Additionally, while ChatGPT can generate creative suggestions, these aren’t always practical or scalable without expert input. Steps such as screening, feasibility checks, fact-checking and market validation require human expertise. Given that ChatGPT’s responses may reflect biases in its training data, people should exercise caution in sensitive contexts such as those involving race or gender.

We also tested whether ChatGPT could assist with tasks often seen as requiring empathy, such as repurposing items cherished by a loved one. Surprisingly, ChatGPT enhanced creativity even in these scenarios, generating ideas that users found relevant and thoughtful. This result challenges the belief that AI cannot assist with emotionally driven tasks.

Future of AI and creativity

As ChatGPT and similar AI tools become more accessible, they open up new possibilities for creative tasks. Whether in the workplace or at home, AI could assist in brainstorming, problem-solving and enhancing creative projects. However, our research also points to the need for caution: While ChatGPT can augment human creativity, it doesn’t replace the unique human capacity for truly radical, out-of-the-box thinking.

This shift from Googling to asking ChatGPT represents more than just a new way to access information. It marks a transformation in how people collaborate with technology to think, create and innovate.

About the Author:

Jaeyeon Chung, Assistant Professor of Business, Rice University

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

By Gertjan Verdickt, University of Auckland, Waipapa Taumata Rau 

When it comes to investing and planning your financial future, are you more willing to trust a person or a computer?

This isn’t a hypothetical question any more.

Big banks and investment firms are using artificial intelligence (AI) to help make financial predictions and give advice to clients.

Morgan Stanley uses AI to mitigate the potential biases of its financial analysts when it comes to stock market predictions. And one of the world’s biggest investment banks, Goldman Sachs, recently announced it was trialling the use of AI to help write computer code, though the bank declined to say which division it was being used in. Other companies are using AI to predict which stocks might go up or down.

But do people actually trust these AI advisers with their money?

Our new research examines this question. We found it really depends on who you are and your prior knowledge of AI and how it works.

Trust differences

To examine the question of trust when it comes to using AI for investment, we asked 3,600 people in the United States to imagine they were getting advice about the stock market.

In these imagined scenarios, some people got advice from human experts. Others got advice from AI. And some got advice from humans working together with AI.

In general, people were less likely to follow advice if they knew AI was involved in making it. They seemed to trust the human experts more.

But the distrust of AI wasn’t universal. Some groups of people were more open to AI advice than others.

For example, women were more likely to trust AI advice than men (by 7.5%). People who knew more about AI were more willing to listen to the advice it provided (by 10.1%). And politics mattered – people who supported the Democratic Party were more open to AI advice than others (by 7.3%).

We also found people were more likely to trust simpler AI methods.

When we told our research participants the AI was using something called “ordinary least squares” (a basic mathematics technique in which a straight line is used to estimate the relationship between two variables), they were more likely to trust it than when we said it was using “deep learning” (a more complex AI method).

This might be because people tend to trust things they understand. Much like how a person might trust a simple calculator more than a complex scientific instrument they have never seen before.

Trust in the future of finance

As AI becomes more common in the financial world, companies will need to find ways to improve levels of trust.

This might involve teaching people more about how the AI systems work, being clear about when and how AI is being used, and finding the right balance between human experts and AI.

Furthermore, we need to tailor how AI advice is presented to different groups of people and show how well AI performs over time compared to human experts.

The future of finance might involve a lot more AI, but only if people learn to trust it. It’s a bit like learning to trust self-driving cars. The technology might be great, but if people don’t feel comfortable using it, it won’t catch on.

Our research shows that building this trust isn’t just about making better AI. It’s about understanding how people think and feel about AI. It’s about bridging the gap between what AI can do and what people believe it can do.

As we move forward, we’ll need to keep studying how people react to AI in finance. We’ll need to find ways to make AI not just a powerful tool, but a trusted advisor that people feel comfortable relying on for important financial decisions.

The world of finance is changing fast, and AI is a big part of that change. But in the end, it’s still people who decide where to put their money. Understanding how to build trust between humans and AI will be key to shaping the future of finance.

About the Author:

Gertjan Verdickt, Lecturer, Business School, University of Auckland, Waipapa Taumata Rau

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

By Veera Sundararaghavan, University of Michigan 

John J. Hopfield and Geoffrey E. Hinton received the Nobel Prize in physics on Oct. 8, 2024, for their research on machine learning algorithms and neural networks that help computers learn. Their work has been fundamental in developing neural network theories that underpin generative artificial intelligence.

A neural network is a computational model consisting of layers of interconnected neurons. Like the neurons in your brain, these neurons process and send along a piece of information. Each neural layer receives a piece of data, processes it and passes the result to the next layer. By the end of the sequence, the network has processed and refined the data into something more useful.

While it might seem surprising that Hopfield and Hinton received the physics prize for their contributions to neural networks, used in computer science, their work is deeply rooted in the principles of physics, particularly a subfield called statistical mechanics.

As a computational materials scientist, I was excited to see this area of research recognized with the prize. Hopfield and Hinton’s work has allowed my colleagues and me to study a process called generative learning for materials sciences, a method that is behind many popular technologies like ChatGPT.

What is statistical mechanics?

Statistical mechanics is a branch of physics that uses statistical methods to explain the behavior of systems made up of a large number of particles.

Instead of focusing on individual particles, researchers using statistical mechanics look at the collective behavior of many particles. Seeing how they all act together helps researchers understand the system’s large-scale macroscopic properties like temperature, pressure and magnetization.

For example, physicist Ernst Ising developed a statistical mechanics model for magnetism in the 1920s. Ising imagined magnetism as the collective behavior of atomic spins interacting with their neighbors.

In Ising’s model, there are higher and lower energy states for the system, and the material is more likely to exist in the lowest energy state.

One key idea in statistical mechanics is the Boltzmann distribution, which quantifies how likely a given state is. This distribution describes the probability of a system being in a particular state – like solid, liquid or gas – based on its energy and temperature.

Ising exactly predicted the phase transition of a magnet using the Boltzmann distribution. He figured out the temperature at which the material changed from being magnetic to nonmagnetic.

Phase changes happen at predictable temperatures. Ice melts to water at a specific temperature because the Boltzmann distribution predicts that when it gets warm, the water molecules are more likely to take on a disordered – or liquid – state.

In materials, atoms arrange themselves into specific crystal structures that use the lowest amount of energy. When it’s cold, water molecules freeze into ice crystals with low energy states.

Similarly, in biology, proteins fold into low energy shapes, which allow them to function as specific antibodies – like a lock and key – targeting a virus.

Neural networks and statistical mechanics

Fundamentally, all neural networks work on a similar principle – to minimize energy. Neural networks use this principle to solve computing problems.

For example, imagine an image made up of pixels where you only can see a part of the picture. Some pixels are visible, while the rest are hidden. To determine what the image is, you consider all possible ways the hidden pixels could fit together with the visible pieces. From there, you would choose from among what statistical mechanics would say are the most likely states out of all the possible options.

Hopfield and Hinton developed a theory for neural networks based on the idea of statistical mechanics. Just like Ising before them, who modeled the collective interaction of atomic spins to solve the photo problem with a neural network, Hopfield and Hinton imagined collective interactions of pixels. They represented these pixels as neurons.

Just as in statistical physics, the energy of an image refers to how likely a particular configuration of pixels is. A Hopfield network would solve this problem by finding the lowest energy arrangements of hidden pixels.

However, unlike in statistical mechanics – where the energy is determined by known atomic interactions – neural networks learn these energies from data.

Hinton popularized the development of a technique called backpropagation. This technique helps the model figure out the interaction energies between these neurons, and this algorithm underpins much of modern AI learning.

The Boltzmann machine

Building upon Hopfield’s work, Hinton imagined another neural network, called the Boltzmann machine. It consists of visible neurons, which we can observe, and hidden neurons, which help the network learn complex patterns.

In a Boltzmann machine, you can determine the probability that the picture looks a certain way. To figure out this probability, you can sum up all the possible states the hidden pixels could be in. This gives you the total probability of the visible pixels being in a specific arrangement.

My group has worked on implementing Boltzmann machines in quantum computers for generative learning.

In generative learning, the network learns to generate new data samples that resemble the data the researchers fed the network to train it. For example, it might generate new images of handwritten numbers after being trained on similar images. The network can generate these by sampling from the learned probability distribution.

Generative learning underpins modern AI – it’s what allows the generation of AI art, videos and text.

Hopfield and Hinton have significantly influenced AI research by leveraging tools from statistical physics. Their work draws parallels between how nature determines the physical states of a material and how neural networks predict the likelihood of solutions to complex computer science problems.

About the Author:

Veera Sundararaghavan, Professor of Aerospace Engineering, University of Michigan

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

By Samantha Keppler, University of Michigan and Clare Snyder, University of Michigan 

Teachers can use generative AI in a variety of ways. They may use it to develop lesson plans and quizzes. Or teachers may rely on a generative AI tool, such as ChatGPT, for insight on how to teach a concept more effectively.

In our new research, only the teachers doing both of those things reported feeling that they were getting more done. They also told us that their teaching was more effective with AI.

Over the course of the 2023-2024 school year, we followed 24 teachers at K-12 schools throughout the United States as they wrestled with whether and how to use generative AI for their work. We gave them a standard training session on generative AI in the fall of 2023. We then conducted multiple observations, interviews and surveys throughout the year.

We found that teachers felt more productive and effective with generative AI when they turned to it for advice. The standard methods to teach to state standards that work for one student, or in one school year, might not work as well in another. Teachers may get stuck and need to try a different approach. Generative AI, it turns out, can be a source of ideas for those alternative approaches.

While many focus on the productivity benefits of how generative AI can help teachers make quizzes or activities faster, our study points to something different. Teachers feel more productive and effective when their students are learning, and generative AI seems to help some teachers get new ideas about how to advance student learning.

Why it matters

K-12 teaching requires creativity, particularly when it comes to tasks such as lesson plans or how to integrate technology into the classroom. Teachers are under pressure to work quickly, however, because they have so many things to do, such as prepare teaching materials, meet with parents and grade students’ schoolwork. Teachers do not have enough time each day to do all of the work that they need to.

We know that such pressure often makes creativity difficult. This can make teachers feel stuck. Some people, in particular AI experts, view generative AI as a solution to this problem; generative AI is always on call, it works quickly, and it never tires.

However, this view assumes that teachers will know how to use generative AI effectively to get the solutions they are seeking. Our research reveals that for many teachers, the time it takes to get a satisfactory output from the technology – and revise it to fit their needs – is no shorter than the time it would take to create the materials from scratch on their own. This is why using generative AI to create materials is not enough to get more done.

By understanding how teachers can effectively use generative AI for advice, schools can make more informed decisions about how to invest in AI for their teachers and how to support teachers in using these new tools. Further, this feeds back to the scientists creating AI tools, who can make better decisions about how to design these systems.

What still isn’t known

Many teachers face roadblocks that prevent them from seeing the benefits of generative AI tools such as ChatGPT. These benefits include being able to create better materials faster. The teachers we talked to, however, were all new users of the technology. Teachers who are more familiar with ways to prompt generative AI – we call them “power users” – might have other ways of interacting with the technology that we did not see. We also do not yet know exactly why some teachers move from being new users to proficient users but others do not.

About the Authors:

The Research Brief is a short take on interesting academic work.

Samantha Keppler, Assistant Professor of Technology and Operations, Stephen M. Ross School of Business, University of Michigan and Clare Snyder, PhD Candidate in Business Administration, University of Michigan

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

By Mahshid Ahmadi, University of Tennessee 

Many human activities release pollutants into the air, water and soil. These harmful chemicals threaten the health of both people and the ecosystem. According to the World Health Organization, air pollution causes an estimated 4.2 million deaths annually.

Scientists are looking into solutions, and one potential avenue is a class of materials called photocatalysts. When triggered by light, these materials undergo chemical reactions that initial studies have shown can break down common toxic pollutants.

I am a materials science and engineering researcher at the University of Tennessee. With the help of robots and artificial intelligence, my colleagues and I are making and testing new photocatalysts with the goal of mitigating air pollution.

Breaking down pollutants

The photocatalysts work by generating charged carriers in the presence of light. These charged carriers are tiny particles that can move around and cause chemical reactions. When they come into contact with water and oxygen in the environment, they produce substances called reactive oxygen species. These highly active reactive oxygen species can bond to parts of the pollutants and then either decompose the pollutants or turn them into harmless – or even useful – products.

But some materials used in the photocatalytic process have limitations. For example, they can’t start the reaction unless the light has enough energy – infrared rays with lower energy light, or visible light, won’t trigger the reaction.

Another problem is that the charged particles involved in the reaction can recombine too quickly, which means they join back together before finishing the job. In these cases, the pollutants either do not decompose completely or the process takes a long time to accomplish.

Additionally, the surface of these photocatalysts can sometimes change during or after the photocatalytic reaction, which affects how they work and how efficient they are.

To overcome these limitations, scientists on my team are trying to develop new photocatalytic materials that work efficiently to break down pollutants. We also focus on making sure these materials are nontoxic so that our pollution-cleaning materials aren’t causing further pollution.

Teeny tiny crystals

Scientists on my team use automated experimentation and artificial intelligence to figure out which photocatalytic materials could be the best candidates to quickly break down pollutants. We’re making and testing materials called hybrid perovskites, which are tiny crystals – they’re about a 10th the thickness of a strand of hair.

These nanocrystals are made of a blend of organic (carbon-based) and inorganic (non-carbon-based) components.

They have a few unique qualities, like their excellent light-absorbing properties, which come from how they’re structured at the atomic level. They’re tiny, but mighty. Optically, they’re amazing too – they interact with light in fascinating ways to generate a large number of tiny charge carriers and trigger photocatalytic reactions.

These materials efficiently transport electrical charges, which allows them to transport light energy and drive the chemical reactions. They’re also used to make solar panels more efficient and in LED lights, which create the vibrant displays you see on TV screens.

There are thousands of potential types of hybrid nanocrystals. So, my team wanted to figure out how to make and test as many as we can quickly, to see which are the best candidates for cleaning up toxic pollutants.

Bringing in robots

Instead of making and testing samples by hand – which takes weeks or months – we’re using smart robots, which can produce and test at least 100 different materials within an hour. These small liquid-handling robots can precisely move, mix and transfer tiny amounts of liquid from one place to another. They’re controlled by a computer that guides their acceleration and accuracy.

We also use machine learning to guide this process. Machine learning algorithms can analyze test data quickly and then learn from that data for the next set of experiments executed by the robots. These machine learning algorithms can quickly identify patterns and insights in collected data that would normally take much longer for a human eye to catch.

Our approach aims to simplify and better understand complex photocatalytic systems, helping to create new strategies and materials. By using automated experimentation guided by machine learning, we can now make these systems easier to analyze and interpret, overcoming challenges that were difficult with traditional methods.

About the Author:

Mahshid Ahmadi, Assistant Professor of Materials Science and Engineering, University of Tennessee

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

By William Mark Stuckey, Elizabethtown College 

The year 2025 marks the 100th anniversary of the birth of quantum mechanics. In the century since the field’s inception, scientists and engineers have used quantum mechanics to create technologies such as lasers, MRI scanners and computer chips.

Today, researchers are looking toward building quantum computers and ways to securely transfer information using an entirely new sister field called quantum information science.

But despite creating all these breakthrough technologies, physicists and philosophers who study quantum mechanics still haven’t come up with the answers to some big questions raised by the field’s founders. Given recent developments in quantum information science, researchers like me are using quantum information theory to explore new ways of thinking about these unanswered foundational questions. And one direction we’re looking into relates Albert Einstein’s relativity principle to the qubit.

Quantum computers

Quantum information science focuses on building quantum computers based on the quantum “bit” of information, or qubit. The qubit is historically grounded in the discoveries of physicists Max Planck and Einstein. They instigated the development of quantum mechanics in 1900 and 1905, respectively, when they discovered that light exists in discrete, or “quantum,” bundles of energy.

These quanta of energy also come in small forms of matter, such as atoms and electrons, which make up everything in the universe. It is the odd properties of these tiny packets of matter and energy that are responsible for the computational advantages of the qubit.

A computer based on a quantum bit rather than a classical bit could have a significant computing advantage. And that’s because a classical bit produces a binary response – either a 1 or a 0 – to only one query.

In contrast, the qubit produces a binary response to infinitely many queries using the property of quantum superposition. This property allows researchers to connect multiple qubits in what’s called a quantum entangled state. Here, the entangled qubits act collectively in a way that arrays of classical bits cannot.

That means a quantum computer can do some calculations much faster than an ordinary computer. For example, one device reportedly used 76 entangled qubits to solve a sampling problem 100 trillion times faster than a classical computer.

But the exact force or principle of nature responsible for this quantum entangled state that underlies quantum computing is a big unanswered question. A solution that my colleagues and I in quantum information theory have proposed has to do with Einstein’s relativity principle.

Quantum information theory

The relativity principle says that the laws of physics are the same for all observers, regardless of where they are in space, how they’re oriented or how they’re moving relative to each other. My team showed how to use the relativity principle in conjunction with the principles of quantum information theory to account for quantum entangled particles.

Quantum information theorists like me think about quantum mechanics as a theory of information principles rather than a theory of forces. That’s very different than the typical approach to quantum physics, in which force and energy are important concepts for doing the calculations. In contrast, quantum information theorists don’t need to know what sort of physical force might be causing the mysterious behavior of entangled quantum particles.

That gives us an advantage for explaining quantum entanglement because, as physicist John Bell proved in 1964, any explanation for quantum entanglement in terms of forces requires what Einstein called “spooky actions at a distance.”

That’s because the measurement outcomes of the two entangled quantum particles are correlated – even if those measurements are done at the same time and the particles are physically separated by a vast distance. So, if a force is causing quantum entanglement, it would have to act faster than the speed of light. And a faster-than-light force violates Einstein’s theory of special relativity.

Many researchers are trying to find an explanation for quantum entanglement that doesn’t require spooky actions at a distance, like my team’s proposed solution.

Classical and quantum entanglement

In entanglement, you can know something about two particles collectively – call them particle 1 and particle 2 – so that when you measure particle 1, you immediately know something about particle 2.

Imagine you’re mailing two friends, whom physicists typically call Alice and Bob, each one glove from the same pair of gloves. When Alice opens her box and sees a left-hand glove, she’ll know immediately that when Bob opens the other box he will see the right-hand glove. Each box and glove combination produces one of two outcomes, either a right-hand glove or a left-hand glove. There’s only one possible measurement – opening the box – so Alice and Bob have entangled classical bits of information.

But in quantum entanglement the situation involves entangled qubits, which behave very differently than classical bits.

Qubit behavior

Consider a property of electrons called spin. When you measure an electron’s spin using magnets that are oriented vertically, you always get a spin that’s up or down, nothing in between. That’s a binary measurement outcome, so this is a bit of information.

If you turn the magnets on their sides to measure an electron’s spin horizontally, you always get a spin that’s left or right, nothing in between. The vertical and horizontal orientations of the magnets constitute two different measurements of this same bit. So, electron spin is a qubit – it produces a binary response to multiple measurements.

Quantum superposition

Now suppose you first measure an electron’s spin vertically and find it is up, then you measure its spin horizontally. When you stand straight up, you don’t move to your right or your left at all. So, if I measure how much you move side to side as you stand straight up, I’ll get zero.

That’s exactly what you might expect for the vertical spin up electrons. Since they have vertically oriented spin up, analogous to standing straight up, they should not have any spin left or right horizontally, analogous to moving side to side.

Surprisingly, physicists have found that half of them are horizontally right and half are horizontally left. Now it doesn’t seem to make sense that a vertical spin up electron has left spin (-1) and right spin (+1) outcomes when measured horizontally, just as we expect no side-to-side movement when standing straight up.

But when you add up all the left (-1) and right (+1) spin outcomes you do get zero, as we expected in the horizontal direction when our spin state is vertical spin up. So, on average, it’s like having no side-to-side or horizontal movement when we stand straight up.

This 50-50 ratio over the binary (+1 and -1) outcomes is what physicists are talking about when they say that a vertical spin up electron is in a quantum superposition of horizontal spins left and right.

Entanglement from the relativity principle

According to quantum information theory, all of quantum mechanics, to include its quantum entangled states, is based on the qubit with its quantum superposition.

What my colleagues and I proposed is that this quantum superposition results from the relativity principle, which (again) states the laws of physics are the same for all observers with different orientations in space.

If the electron with a vertical spin in the up direction were to pass straight through the horizontal magnets as you might expect, it would have no spin horizontally. This would violate the relativity principle, which says the particle should have a spin regardless of whether it’s being measured in the horizontal or vertical direction.

Because an electron with a vertical spin in the up direction does have a spin when measured horizontally, quantum information theorists can say that the relativity principle is (ultimately) responsible for quantum entanglement.

And since there is no force used in this principle explanation, there are none of the “spooky actions at a distance” that Einstein derided.

With quantum entanglement’s technological implications for quantum computing firmly established, it’s nice to know that one big question about its origin may be answered with a highly regarded physics principle.

About the Author:

William Mark Stuckey, Professor of Physics, Elizabethtown College

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