Artifical Intelligence News

AI Meta News. Explore the latest news, insights and use cases on how Meta and Facebook are using AI here at AI News.

Researchers from the University of Toronto have used AI to generate a “recipe” for an exciting new catalyst needed to produce green hydrogen fuel.

As the effects of climate change begin to become more apparent in our everyday lives, research like this could open the door to green hydrogen fuel that could be used for everything from transportation to residential and commercial heating.

While the entire process of creating hydrogen fuel is quite complicated, it can be summarized more simply.

Scientists take water and pass electricity from renewable sources between two pieces of metal called electrodes that are submerged in the water. These electrodes are coated in a catalyst that speed up the process of splitting the water into its two parts – hydrogen gas and oxygen gas. This hydrogen gas is then taken and can be used for fuel.

Until now, iridium oxide was the most widely used catalyst that could withstand the harsh acidic conditions in this reaction. Sadly, iridium is extremely scarce and expensive. This makes it an unsustainable source for large-scale hydrogen production. Ruthenium-based catalysts are more abundant and less expensive than iridium, but they suffer from instability due to the overoxidation of ruthenium atoms during the reaction.

Therefore, the scientists at the University of Toronto endeavored to use AI resources to solve this problem. The team created an AI program to speed up the search for an optimal alloy combination that would act as a catalyst in the water-splitting reaction. This program analyzed over 36,000 different metal oxide combinations through virtual simulations. Traditionally, such a search would require trial and error in the lab.

The Canadian Light Source at the University of Saskatchewan allows scientists to look at things that we can\’t see with our eyes or regular microscopes.

“We’re talking about hundreds of millions or billions of alloy candidates, and one of them could be the right answer,” Jehad Abed, who was part of the team that developed the AI program, said in a blog post about the research.

After letting the AI program chew through the 36,000 different metal oxide combinations, the program recommended a combination of ruthenium, chromium, and titanium. Aben then tested the program’s top candidate in the lab to see if the program was accurate. Then, the team used the Canadian Light Source (CLS) at the University of Saskatchewan to help researchers understand how the new catalyst works.

Think of the CLS as a super-powerful microscope that uses bright X-rays to look at things we can’t see with our eyes or regular microscopes. The scientists shined these bright X-ray lights on their new catalyst while the catalyst was working to split the water. This allowed them to see how the atoms in the catalyst were arranged and how they moved during the reaction. Specifically, they wanted to ensure the ruthenium didn’t dissolve, which can often happen due to overoxidation.

After this process, the scientists found that their new catalyst was better at keeping the ruthenium from dissolving than other candidates. Additionally, they saw that this catalyst was less likely to break apart oxygen from its own structure, thereby helping it last longer. What’s more, they witnessed some pretty incredible results.

“The computer’s recommended alloy performed 20 times better than our benchmark metal in terms of stability and durability,” said Abed. “It lasted a long time and worked efficiently.”

This research is considered a major success, but scientists still have a long way to go before the ruthenium, chromium, and titanium alloy can be used in large-scale hydrogen production. The scientists stated that they would need a lot of testing to ensure it would last under “real world” conditions.

The work performed here is an example of how AI can offer faster routes to finding answers to the complex questions surrounding the climate crisis.

Researchers from the University of Toronto have used AI to generate a “recipe” for an exciting new catalyst needed to produce green hydrogen fuel.

As the effects of climate change begin to become more apparent in our everyday lives, research like this could open the door to green hydrogen fuel that could be used for everything from transportation to residential and commercial heating.

While the entire process of creating hydrogen fuel is quite complicated, it can be summarized more simply.

Scientists take water and pass electricity from renewable sources between two pieces of metal called electrodes that are submerged in the water. These electrodes are coated in a catalyst that speed up the process of splitting the water into its two parts – hydrogen gas and oxygen gas. This hydrogen gas is then taken and can be used for fuel.

Until now, iridium oxide was the most widely used catalyst that could withstand the harsh acidic conditions in this reaction. Sadly, iridium is extremely scarce and expensive. This makes it an unsustainable source for large-scale hydrogen production. Ruthenium-based catalysts are more abundant and less expensive than iridium, but they suffer from instability due to the overoxidation of ruthenium atoms during the reaction.

Therefore, the scientists at the University of Toronto endeavored to use AI resources to solve this problem. The team created an AI program to speed up the search for an optimal alloy combination that would act as a catalyst in the water-splitting reaction. This program analyzed over 36,000 different metal oxide combinations through virtual simulations. Traditionally, such a search would require trial and error in the lab.

The Canadian Light Source at the University of Saskatchewan allows scientists to look at things that we can\’t see with our eyes or regular microscopes.

“We’re talking about hundreds of millions or billions of alloy candidates, and one of them could be the right answer,” Jehad Abed, who was part of the team that developed the AI program, said in a blog post about the research.

After letting the AI program chew through the 36,000 different metal oxide combinations, the program recommended a combination of ruthenium, chromium, and titanium. Aben then tested the program’s top candidate in the lab to see if the program was accurate. Then, the team used the Canadian Light Source (CLS) at the University of Saskatchewan to help researchers understand how the new catalyst works.

Think of the CLS as a super-powerful microscope that uses bright X-rays to look at things we can’t see with our eyes or regular microscopes. The scientists shined these bright X-ray lights on their new catalyst while the catalyst was working to split the water. This allowed them to see how the atoms in the catalyst were arranged and how they moved during the reaction. Specifically, they wanted to ensure the ruthenium didn’t dissolve, which can often happen due to overoxidation.

After this process, the scientists found that their new catalyst was better at keeping the ruthenium from dissolving than other candidates. Additionally, they saw that this catalyst was less likely to break apart oxygen from its own structure, thereby helping it last longer. What’s more, they witnessed some pretty incredible results.

“The computer’s recommended alloy performed 20 times better than our benchmark metal in terms of stability and durability,” said Abed. “It lasted a long time and worked efficiently.”

This research is considered a major success, but scientists still have a long way to go before the ruthenium, chromium, and titanium alloy can be used in large-scale hydrogen production. The scientists stated that they would need a lot of testing to ensure it would last under “real world” conditions.

The work performed here is an example of how AI can offer faster routes to finding answers to the complex questions surrounding the climate crisis.

Lattice structures, characterized by their complex patterns and hierarchical designs, offer immense potential across various industries, including automotive, aerospace, and biomedical engineering. 

With their outstanding high strength-to-weight ratio, customizability, and versatility, lattice structures enable the development of lightweight, durable components that can be precisely tailored to meet specific functional requirements. 

However, the complexity of the structure and the vastness of the design space encompassed by lattice structures makes it challenging for traditional methods to thoroughly explore all possible configurations and pinpoint the optimal solution for the application. With each additional design variable, the possible configurations grow exponentially, making the design space intractable.

Lawrence Livermore National Laboratory (LLNL) engineers are looking to address these challenges by harnessing the power of machine learning (ML) and artificial intelligence (AI). Advanced computational tools powered by ML and AI have enabled LLNL researchers to accelerate and enhance the optimization of lattice structure designs significantly.

In a study published by Scientific Reports, LLNL researchers detailed how they used a combination of ML algorithms and traditional methods to optimize design variables, predict mechanical performance, and accelerate the design process for lattices with millions of potential configurations.

“By leveraging machine learning-based approaches in the design workflow, we can accelerate the design process to truly leverage the design freedom afforded by lattice structures and take advantage of their diverse mechanical properties,” said lead author and LLNL engineer Aldair Gongora.

“This work advances the field of design because it demonstrates a viable way of integrating iterative ML-based approaches in the design workflow and underscores the critical role ML and artificial intelligence (AI) can play in accelerating design processes.”  

The LLNL researchers used ML to tackle two main challenges in designing lattice structures. First, they developed a model that helped them understand the impact of various design choices on the lattice’s mechanical performance. Second, they created a method to efficiently identify which designs are the most effective. 

At the core of the research was the creation of ML-driven surrogate models that act as digital prototypes for investigating the mechanical properties of lattice structures. These models were trained on a vast dataset that included various lattice design variables. 

The surrogate models were able to deliver valuable insights into design parameters and their impact on mechanical performance. According to Gongora, the accuracy of the surrogate models exceeded 95% and enabled the researchers to optimize lattice design by exploring only 1% of the design space size. 

Using Bayesian optimization and Shapley additive explanation (SHAP) analysis, the researchers efficiently explored lattice design options, reducing both computational load and the number of simulations required to identify optimal designs. They claim that their custom active-learning approach to finding optimal lattice structures required 82% fewer simulations compared to traditional grid-based search methods.

The research has set a new benchmark for intelligent design systems using computational modeling and ML algorithms. It also highlights AI’s pivotal role in designing lattice structures for a variety of applications

Looking ahead, Gongora is hopeful that his research will have an impact that goes beyond the realm of lattice structures. He believes that the approach can be applied to various design challenges, which often rely on expensive simulations. 

Related Items 

LLNL Could Reverse Ocean Acidification

Researchers Use Machine Learning To Optimize High-Power Laser Experiments 

Generative AI to Account for 1.5% of World’s Power Consumption by 2029 

IBM, a leader in global hybrid cloud and AI solutions, unveiled the specifications of its new Telum II processors and Spyre AI accelerators, which are set to drive AI workloads on the latest IBM Z mainframe systems.

The announcement was made at the Hot Chips 2024 event happening this week at Stanford University. 

As GenAI initiatives transition from proof of concept to production, there is an increasing demand for power-efficient, secure, and scalable solutions. According to research from Morgan Stanley, GenAI power demands will rise 70% annually for the next few years. The research indicates that by 2027, GenAI can consume as much energy as Spain needed to power itself in 2022. 

The new Telum II process and Spyre AI accelerators are engineered to address these escalating demands effectively. 

The Telum II processor builds on its predecessor, the first-generation Telum chip, with several key improvements. The new processor features a completely new data processing unit (DPU) designed to accelerate complex IO protocols for networking and storage on the mainframe and to improve key component performance.

IBM claims the new DPU offers increased frequency, memory capacity, and an integrated AI accelerator core. This allows it to handle larger and more complex datasets efficiently. 

The Spyre accelerator complements the Telum II processor by offering additional AI compute capabilities and supports what IBM calls “ensemble methods of AI modeling” – which combines multiple models to potentially boost prediction accuracy. 

Ensemble methods enhance the robustness of AI predictions, making them more reliable and less sensitive to errors or variations compared to individual models. The IBM Spyre accelerator chip will be delivered as an add-on option. Each Spyre chip contains 32 compute cores for AI applications, which reduces latency and enhances throughput across various AI tasks.

\”Our robust, multi-generation roadmap positions us to remain ahead of the curve on technology trends, including escalating demands of AI,\” said Tina Tarquinio, VP, Product Management, IBM Z, and LinuxONE. 

\”The Telum II processor and Spyre accelerator are designed to deliver high-performance, secured, and more power-efficient enterprise computing solutions. After years in development, these innovations will be introduced in our next-generation IBM Z platform so clients can leverage LLMs and generative AI at scale.\”

The new Telum II processor is a major upgrade to the original Telum processor that debuted in 2021. With 8 high-performance cores running at 5.5 gigahertz and with 36 megabytes of memory per core, Telum II processors offer an increase of 40% in on-chip cache capacity for a total of 360MB. The integrated AI accelerator enables low-latency, high-throughput AI inferencing during transactions. 

The powerful specifications of the two new technologies translate to better efficiency and security for AI-powered applications. While the  Telum II processor is designed to efficiently manage large-scale AI workloads and data-intensive business needs, the Spyre accelerator is geared toward handling complex AI models and generative AI use cases. 

The integration of Telum II and Spyre accelerators eliminates the need to transfer data to external GPU-equipped servers, thereby enhancing the mainframe\’s reliability and security.

Both technologies will be manufactured by IBM\’s long-standing fabrication partner, Samsung Foundry, using a 5 nm process. IBM expects the Telum II processor to be available to LinuxOnE and IBM Z clients in 2025. The Spyre accelerator is also expected to be available in 2025. 

Related Items 

Groq’s $640 Million Funding Poised to Disrupt AI Chip Market 

Is the GenAI Bubble Finally Popping? 

AWS and Fujitsu Expand Partnership to Modernize Legacy Cloud Applications 

A new bill that has advanced to the California Senate Assembly floor represents both a significant step forward in AI governance as well as a risk to the technology’s innovative growth. Officially called California Senate Bill 1047 – and also known as the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act – this bill is meant to regulate large-scale AI models in the state of California.

Authored by State Senator Scott Wiener, this bill would require AI companies to test their models for safety. Specifically, the bill targets “covered models,” which are AI models that exceed certain compute and cost thresholds. Any model that costs more than $100 million to train would fall under the jurisdiction of this bill.

Elon Musk has thrown his support behind this bill.

As of August 27, 2024, the bill has passed the California Assembly Appropriations Committee and will be soon advancing to the Assembly floor for a final vote.

California Senate Bill 1047 has a variety of requirements for builders of large AI models. One of these is to create a “full shutdown” ability that enables someone in authority to immediately shut down an unsafe model during nefarious or dangerous circumstances.

On top of this, developers will be required to generate a written safety and security protocol in the event of a worst-case scenario with the AI model. Companies such as Amazon, Google, Meta, and OpenAI have already made voluntary pledges to the Biden Administration to ensure the safety of their AI products. That said, this new bill would give the Californian government certain powers to enforce the bill’s regulations.

Additionally, California Senate Bill 1047 would require companies to retain an unredacted and unchanged copy of the safety and security protocol for the model for as long as the model is in use, plus five years. This is meant to ensure that developers maintain a complete and accurate record of their safety measures, thereby allowing for thorough audits and investigations if needed. If an adverse event were to occur with the model, this regulation should help developers prove they were adhering to safety standards – or that they weren’t.

In short, the bill is meant to prohibit companies from making a model commercially available if there is an unreasonable risk of causing or enabling harm.

Of course, it’s easy to put some words on a page. It’s much more difficult to actually follow through with the promises made in those words. The bill would also create the Board of Frontier Models within the Government Operations Agency. The functions of this group would be to provide high-level guidance on AI policy and regulation, approve regulations proposed by the Frontier Model Division, and ensure that oversight measures keep pace with the explosion of AI technology.

This bill also gives the California Attorney General power to address potential harms caused by Ai models. The Attorney General would have the authority to take action against developers whose AI models cause sever harm or pose imminent public safety threats. This person would also have the ability to bring civil actions against non-compliant developers as well as the power to enforce penalties for violations.

If the bill passes, developers will have until January 1, 2026 to begin annually retaining a third-party auditor to perform an independent compliance audit. Developers will be required to retain an unredacted copy of the audit report and grant access to the Attorney General upon request.

As one might imagine, this bill is causing an uproar among Silicon Valley elite. One of the biggest concerns is that this bill could hamper innovation in the AI community. Many of the U.S.’s AI companies reside within California, and as such this bill would have major ramifications on the entire U.S. tech industry. Certain critics believe the regulations will slow companies down, and allow foreign organizations to gain ground.

Additionally, there seems to be debate around the definition of “covered models” and “critical harm.” While both of these phrases appear many times within the bill, their actual definitions are considered by some to be too broad or vague. This could potentially lead to overregulation.

That said, there are also many supporters for the bill. In fact, Elon Musk has thrown his support behind the bill, stating on X that he has been “an advocate for AI regulation, just as we regulate any product/technology that is a potential risk.”

As of right now, we don’t know if and when the bill will pass the Assembly floor’s final vote. When it does, it will go to the Governor for either a signature or a veto.

California has an opportunity to shape the future of AI development with this bill, and it will be interesting to see which way the decision swings.

AI has changed everything, including HPC. A recent HPCwire article discussed how AI-augmented HPC is creating a large paradigm shift in many aspects of the market and community.

Supporting this shift, Hyperion Research announced today that it has updated its market sizing for the HPC (technical computing servers) market precipitated by AI’s unprecedented impact on the market. In addition to revenue from traditional HPC suppliers, revenue attributed to non-traditional HPC suppliers, often from non-traditional HPC buyers, is now included. These new types of servers are called “AI-centric Servers” in the figure and table below.

This change in scope increases the overall HPC server market growth rate from our January 2024 forecast to 36.7% in 2023 and is projected to add $13.6B by 2028. According to Earl Joseph, CEO of Hyperion Research:

“As AI is being integrated into traditional HPC workflows, and HPC-class machines are being purchased to adopt AI into enterprise datacenter applications, we are seeing a great increase in new buyers and new suppliers that have entered the HPC market.“

The changes to the HPC market sizing and tracking are based on multiple surveys conducted over the last two years and represent a major adjustment to the overall HPC market, heavily driven by a combination of the use of AI, Large Language Models, and the purchases of large GPU/accelerator-based systems. HPC systems are used as the backbone for many AI and AI-related applications, which has driven the overall HPC market to a new level of growth.

Market Segment Definitions

Hyperion has provided updated market segmentation criteria.

AI-centric Servers (new revenues added to the previous HPC market sizing)

These are on-premises AI-centric HPC servers provided by non-traditional HPC suppliers like NVIDIA, Cerebras, SambaNova, SuperMicro, etc., frequently at non-traditional HPC user sites like large enterprise sites adding AI capabilities.

• These servers are designed primarily to run AI and AI-related workloads.
• These servers are a subsegment of the overall HPC market but haven’t historically been accounted for within prior HPC market numbers.

HPC & Advanced AI Servers (previously counted in the HPC market sizing)

These are on-premises servers that are used for highly computational or data-intensive tasks:

• Hyperion Research uses the terms technical computing and high-performance computing (HPC) to encompass the entire market for computer servers used by scientists, engineers, analysts, and other groups using computationally and/or data-intensive modeling and simulation applications. An on-premises system primarily used for HPC workloads (at least 50%) can be referred to as an HPC/AI system or simply an HPC. These servers also include AI-centric servers used by traditional HPC end-users and are typically sold by traditional server vendors.
• In addition to scientific and engineering applications, technical computing includes related markets/applications areas such as economic analysis, financial analysis, animation, server-based gaming, digital content creation and management, business intelligence modeling, and homeland security database applications.
• Systems acquired by cloud service providers for the purpose of hosting cloud workloads are excluded, as Hyperion Research separately tracks spending for HPC usage in cloud environments.
• The Advanced AI servers previously tracked are primarily sold by traditional HPC system vendors that go to traditional HPC sites.

Find More HPC/AI Discussions at the Fall HPC User Forum

The latest information and expert discussions on HPC and AI are part of the fall Hyperion Research HPC/AI User Forum. According to the agenda many topical HPC and AI talks will be presented including “The Future of AI” by Rick Stevens of ANL and a session on Real-world Large Scale AI Use Cases and Examples. The forum will be held at Argonne National Laboratory from September 4-5, 2024

About Hyperion Research

Hyperion Research is the premier industry analysis and market intelligence firm for high-performance computing (HPC) and associated emerging markets. Hyperion Research analysts provide timely, in-depth mission-critical insight across a broad portfolio of advanced computing market segments, including High-Performance Computing (HPC), Advanced Artificial Intelligence (AI), High-Performance Data Analysis (HPDA), Quantum Computing (QC), Cloud, and advanced technologies.

For close to 35 years, the industry analysts at Hyperion Research have been helping private, public, and government organizations make intelligent, fact-based decisions related to business impact and technology direction in the complex and competitive landscape of advanced computing and emerging technologies.

Doubt is creeping into discussion over generative AI, as industry analysts begin to publicly question whether the huge investments in GenAI will ever pay off. The lack of a “killer app” besides coding co-pilots and chatbots is the most pressing concern, critics in a Goldman Sachs Research letter say, while data availability, chip shortages, and power concerns also provide headwinds. However, many remain bullish on the long-term prospects of GenAI for business and society.

The amount of sheer, unadulterated hype layered onto GenAI over the past year and a half certainly caught the attention of seasoned tech journalists, particularly those who lived through the dot-com boom and ensuing bust at the turn of the century, not to mention the subsequent rise of cloud computing and smartphones with the introduction of Amazon Web Services and the Apple iPhone in 2006 and 2007, respectively.

The big data boom of the early 2010s was the next tech obsession, culminating with the coronation of Hadoop as The New New Thing, to paraphrase Michael Lewis’ illuminating 1999 book into Silicon Valley’s fixation on continuous technological reinvention. After the collapse of Hadoop–slowly at first, and then all of a sudden in 2019–the big data marketing machine subtly shifted gears and AI was the hot new thing. Several other new (new) things made valiant runs for attention and VC dollars along the way–Blockchain will change the world! 5G will turbocharge edge computing! Self-driving cars are almost here! Smart dust is new oil!–but nothing really seemed to really gain traction, and the big data world made incremental gains with traditional machine learning while wondering what these newfangled neural networks would ever be good for.

That is, until OpenAI dropped a new large language model (LLM) called ChatGPT onto the world in late 2022. Since then, the hype level for neural network-powered AI, and transformer network-based GenAI in particular, has been eerily reminiscent of these previous Big Moments In Tech. It’s worth pointing out that some of these big moments turned out to be actual inflection points, such as mobile and cloud, some had us asking ourselves “What were we thinking (blockchain, 5G), while it took years for the full lessons from other technological breakthroughs to become apparent (the dot-com boom, even Hadoop-style computing).

So the big question for us now is: Which of those categories will we be putting GenAI into in five years? One of the voices suggesting AI may go the way of 5G and blockchain is none other than Goldman Sachs. In a much-read report from the June edition of the Goldman Sachs Research Newsletter titled “Gen AI: too much spend, too little benefit?” Editor Allison Nathan ponders whether AI will pan out.

“The promise of generative AI technology to transform companies, industries, and societies continues to be touted, leading tech giants, other companies, and utilities to spend an estimated ~$1tn on capex in coming years, including significant investments in data centers, chips, other AI infrastructure, and the power grid,” she writes. “But this spending has little to show for it so far beyond reports of efficiency gains among developers.”

Nathan interviewed MIT Professor Daron Acemoglu, who said that only a quarter of tasks that AI is supposed to automate will actually be automated in a cost-effective manner. Overall, Acemoglu estimates that only 5% of all tasks will be automated within 10 years, raising the overall productivity of the United States by less than 1% over that time.

“Generative AI has the potential to fundamentally change the process of scientific discovery, research and development, innovation, new product and material testing, etc. as well as create new products and platforms,” Acemoglu told Nathan. “But given the focus and architecture of generative AI technology today, these truly transformative changes won’t happen quickly and few–if any–will likely occur within the next 10 years.”

Accelerating GenAI progress by ramping up production of its two core ingredients–data and GPUs–probably won’t work, as data quality is a big piece of the equation, Acemoglu said.

“Including twice as much data from Reddit into the next version of GPT may improve its ability to predict the next word when engaging in an informal conversation,” he said, “but it won’t necessarily improve a customer service representative’s ability to help a customer troubleshoot problems with their video service.”

A shortage in chips suitable for training GenAI models is another factor in Goldman’s pessimistic (some would say realistic) take on GenAI. That has benefited Nvidia enormously, which saw revenue grow by more than 260%, to $26 billion, for the quarter ended April 28. That helped pump its market cap over the $3-trillion market, joining Microsoft and Apple as the most valuable companies in the world.

“Today, Nvidia is the only company currently capable of producing the GPUs that power AI,” Jim Covello, Goldman’s head of global equity research, wrote in the newsletter. “Some people believe that competitors to Nvidia from within the semiconductor industry or from the hyperscalers–Google, Amazon, and Microsoft–themselves will emerge, which is possible. But that’s a big leap from where we are today given that chip companies have tried and failed to dethrone Nvidia from its dominant GPU position for the last 10 years.”

The huge costs involved in training and using GenAI act as headwinds against any productivity or efficiency gains that the GenAI may ultimately deliver, Covello said.

“Currently, AI has shown the most promise in making existing processes–like coding–more efficient, although estimates of even these efficiency improvements have declined, and the cost of utilizing the technology to solve tasks is much higher than existing methods,” he wrote.

Covello was semiconductor analyst when smartphones were first introduced, and learned a few lessons about what it takes to actually realize monetary gains from technological innovation. For instance, the smartphone makers promised to integrate global positioning systems (GPS) into the phones, he said, and they had a roadmap that proved prescient.

“No comparable roadmap exists today” for AI, he said. “AI bulls seem to just trust that use cases will proliferate as the technology evolves. But eighteen months after the introduction of generative AI to the world, not one truly transformative–let alone cost-effective–application has been found.”

Finally, the amount of power required to train LLMs and other GenAI models has to be factored into the equation. It’s been estimated that AI currently consumes about 0.5% of the world’s energy, and that amount is expected to increase in the future.

“Utilities are fielding hundreds of requests for huge amounts of power as everyone chases the AI wave, but only a fraction of that demand will ultimately be realized,” says Brian Janous, the Co-founder of Cloverleaf Infrastructure and formerly the VP of energy at Microsoft.

The total capacity of power projects waiting to connect to the grid grew nearly 30% last year, with wait times currently ranging from 40-70 months, Janous said. With so many projects waiting for power, data centers looking for more power to fuel AI training will become “easy targets.”

The US needs to expand its grid to handle expected increase for power demand, but that isn’t likely to be done cheaply or efficiently, he said. “The US has unfortunately lost the ability to build large infrastructure projects–this is a task better suited for 1930s America, not 2030s America,” Janous said. “So, that leaves me a bit pessimistic.”

But not everyone is pessimistic about AI’s future. One GenAI optimist is Joseph Briggs, Goldman’s senior global economist. In his article countering  Acemoglu, Briggs estimates that GenAI ultimately will automate 25% of all work tasks and raise US productivity by 9% and GDP growth by 6.1% cumulatively over the next decade. What’s more, GenAI will not only automate some existing tasks currently done by humans, but will spur the creation of new tasks, he said.

“…[T]he full automation of AI exposed tasks that are likely to occur over a longer horizon could generate significant cost savings to the tune of several thousands of dollars per worker per year,” he wrote. “The cost of new technologies also tends to fall rapidly over time. Given that cost-saving applications of generative AI will likely follow a similar pattern, and that the marginal cost of deployment will likely be very small once applications are developed, we expect AI adoption and automation rates to ultimately far exceed Acemoglu’s 4.6% estimate.”

Kash Rangan is another GenAI believer. In an interview with the Goldman editor Nathan, the senior equity research analyst said he’s amazed at the pace of GenAI innovation and impressed at the infrastructure buildout of the cloud bigs. He acknowledged that GenAI hasn’t discovered its killer app yet, in the way that ERP dominated the 1990s, search and e-commerce dominated the 2000s, and cloud applications dominated the 2010s.

“But this shouldn’t come as a surprise given that every computing cycle follows a progression known as IPA—infrastructure first, platforms next, and applications last,” Rangan said. “The AI cycle is still very much in the infrastructure buildout phase, so finding the killer application will take more time, but I believe we’ll get there.”

His colleague, Eric Sheridan, joined him in a bullish stance.

“So, the technology is still very much a work in progress. But it’s impossible to sit through demonstrations of generative AI’s capabilities at company events or developer conferences and not come away excited about its long-term potential,” he said.

“So, while I would never say I’m not concerned about the possibility of no payback, I’m not particularly worried about it today, though I could become more concerned if scaled consumer applications don’t emerge over the next 6-18 [months],” Sheridan said.

The promise of GenAI remains high, if unfulfilled at the end of the day. The big question right now is whether GenAI’s returns will go up before the clock runs out. The clock is ticking.

Governments are scrambling to stay ahead of the AI tsunami, and for good reason. Like any other useful technology, AI presents a gigantic economic opportunity for governments worldwide.

In fact, PricewaterhouseCoopers (PwC) has estimated that AI could contribute up to $15.7 trillion to the global economy in 2030 – which would be more than the current economic output of both China and India combined. PwC has also provided a breakdown of where this money is likely to come from. The organizations stated that about $6.6 trillion will come from increased productivity, while an additional $9.1 trillion will likely come from “consumption-side effects.”

To learn more about how governments spend on AI and what we can expect from government AI spending in the next few years, let’s dive into some specific numbers.

The Opportunity AI Presents

Clearly, AI is going to have a massive effect on the economy as a whole. As such, governments are looking to invest in ways that give their citizens a leg-up in future economic endeavors. However, government officials also have their eyes on solving government-specific problems.

We’ll go into more detail about each government\’s investments in sectors like healthcare and national security in future articles, but one of the more interesting specific benefits of AI in government is helping overcome bureaucracy and regulation.

One might argue that a government’s sole reason for existence is to uphold laws, but many of these regulations can be difficult to navigate – both for average citizens and experienced government workers.

For the former, look at the “Ask Jamie” virtual assistant in Singapore. This AI tool is meant to help citizens and businesses navigate government-provided services across nearly 70 government agencies. “Ask Jamie” is powered through both chat and voice and is meant to make life easier for Singaporean citizens.

Singapore\’s government has made past commitments to staying on the cutting edge of technology.

For the latter, we can look to a recent interview I conducted with Argonne National Laboratory’s own Rick Stevens. While we discussed a lot of topics relating to Argonne’s AI for Energy report, one of the more interesting aspects of our conversation was about nuclear reactors.

As you might imagine, nuclear reactors are some of the most complicated systems humans have ever created. While they are magnificent achievements of innovation that will help wean us off fossil fuels, nuclear reactors can also be extremely dangerous when implemented incorrectly.

The AI for Energy report stated that building advanced nuclear reactors in the U.S. is a “slow, expensive, and convoluted regulatory process.” While the report states that obtaining a construction permit and operating license for a new reactor in the U.S. generally takes five years, the process can sometimes stretch into multiple decades.

By training on datasets of scientific literature, technical documents, and operational data, multi-modal LLMs could help streamline and expedite the nuclear regulatory licensing and compliance process. Considering much of government work is finding ways to cut through bureaucratic red tape to actually accomplish something, AI stands to completely change how our governments operate.

Regional Investment Strategies

Although AI is a widely used tool at this point, its implementation will differ based on region and local government. In later articles, we’ll break down global spending by region much more granularly.

Here, we’ll give a quick overview of what’s happening around the world with government AI spending.

China

• In July 2017, China’s State Council announced the New Generation Artifiical Intelligence Development Plan. Since then, total Chinese national and local government spending to implement the plan has not been publicly disclosed.
• By 2022, the Chinese government had reportedly created 2,107 guidance funds with a registered target size of $1.86 trillion. However, by 2023, a report from Zero2IPO stated that these funds had only raised a total of $940 billion.
• One of the specific regional numbers we do have is from 2018, where Shanghai announced it would launch a fund of about 100 billion yuan (about $14.6 billion at the time) for developing China’s AI industry.

European Union

• Like China, the European Union announced a national plan for AI investment called the AI Innovation Strategy. It includes a public and private investment package of around €4 billion through 2027 dedicated specifically to generative AI.
• The AI Innovation strategy calls for a variety of measures, beginning with the intention to create “AI Factories” across the European Union that will bring together supercomputing infrastructure and human resources to further develop AI applications
• Additionally, the Commission intends to make data available through the development of “Common European Data Spaces.” The objective here is to improve the availability of and access to high-quality data for start-ups and other innovation organizations to train AI systems, models, and applications.
• On top of the AI Innovation Strategy, the European Union has also established the AI Act to establish a comprehensive legal framework for AI. While not specifically aimed at innovation and growth, the AI Act is meant to foster trustworthy AI that respects fundamental rights and ethical principles.

United States

• Like the European Union and China, the United States also has a plan in place concerning AI in the form of the U.S. National AI R&D Strategic Plan. Updated in 2023, it outlines the federal government’s roadmap for AI research and development. The United States also has the National AI Initiative Act
• In 2022 fiscal year, Federal government spending on AI hit $3.3 billion, which is a 2.5 times increase over 2017’s $1.3 billion.
• The overall United States Federal IT Budget for 2025 is projected to be $75.13 billion, with a heavy focus on cybersecurity and AI.
• The Department of Defense has been a major driver of AI spending, with AI-related federal contracts increasing by almost 1,200% from $355 million in August 2022 to $4.6 billion in August 2023.
• The United States will also be the largest market for AI-centric systems, accounting for more than 50% of all AI spending worldwide.

Notable Mentions

• Japan: Spearheaded by the Ministry of Economy, Trade, and Industry, Japan is working with Nvidia as well as other Japanese companies to unlock the economic potential of AI in the country. Japan has allocated approximately ¥114.6 billion ($740 million) to subsidize the AI computing industry in the country.
• India: The Indian government has recently announced the IndiaAi Mission initiative to advance the country’s AI ecosystem. Out of the investment of 74 billion Indian rupee (US$1.25 billion), around 45 billion Indian rupee ($543 million) will be used to build computer infrastructure while 20 billion Indian rupee ($241 million) will be used to finance startups.
• South Korea: The South Korean government plans to invest 9.4 trillion won ($6.94 billion) in AI by 2027. This money is meant to help the country retain an edge in the semiconductor industry and develop AI chips, such as artificial neural processing units and next-generation high-bandwidth memory chips.

This may not represent a global roundup of all investments in AI, but the amount of money the major players are putting up toward AI development is certainly noteworthy.

Challenges and Barriers

While governments are seeing the benefits of investing in AI, there will be some challenges to overcome on both national and international scales. To begin, some social issues will need to be solved over time.

There is a notable AI skills shortage in most countries, and many of the existing workforce may be resistant to adopting new AI technologies. While both of these problems can be assisted by government-funded educational efforts, there are more specific problems for nations to address.

A larger problem that will require a lot of money to address is that many of the legacy systems that government agencies use aren’t designed to work with AI/ML implementations. Solving this will involve modernizing data, networks, the cloud, and cybersecurity capabilities on an enormous scale.

Finally, the overall cost of AI infrastructure will hamper a government\’s ability to implement these tools quickly. A recent poll found that 55% of respondents reported that the most significant barrier to the adoption of AI-enabled tools was cost. The various hardware required for AI work has exploded in price recently, and those costs aren’t going to go down anytime soon. Any investment in AI on a governmental scale will demand a hefty upfront cost that some nations simply cannot afford on their own.

Proteins, life’s building blocks, perform a wide range of functions based on their unique shapes. The molecules fold into specific forms and shapes that define their roles, from catalyzing biochemical reactions to providing structural support and enabling cellular communication.

Predicting the protein structure is challenging due to the complexity of the folds and shapes. Even slight variations in folding can significantly alter a protein\’s function.

To address this complexity, researchers have developed a new open-source software tool called OpenFold that leverages the power of supercomputers and AI to predict protection structures. This can help scientists gain a deeper understanding of misfolded proteins associated with neurodegenerative diseases, such as Parkinson’s and Alzheimer’s disease, and develop new medicines. 

OpenFold, which was announced in a study published in the Nature Methods journal, builds on the success of AlphaFold2, an AI program developed by DeepMind that predicts the structure and interactions between biological molecules with unprecedented accuracy. 

AlphaFold2 is being used by over two million researchers for protein predictions in various fields, including drug discovery and medical treatments. While AlphaFold2 offers exceptional accuracy, it is limited by its lack of accessible code and data for training new models. 

(Shutterstock)

This restricts its application to new tasks, like protein-ligand complex structure prediction, understanding its learning process, or assessing the model’s capacity for unseen regions of fold space.

The research for OpenFold was initiated by Dr. Nazim Bouatta, a senior research fellow at Harvard Medical School, and his colleague Mohammed AlQuraishi, formerly at Harvard but now at Columbia University. The project was supported by several other researchers from Harvard and Columbia. 

The project eventually grew into the OpenFold Consortium, a non-profit AI research and development consortium developing free and open-source software tools for biology and drug discovery.

A core component of AI-based research is large language models (LLMs), which can process vast amounts of data to generate new and meaningful insights. The ability to use natural language to interact with AI has greatly enhanced accessibility and usability, allowing users to communicate with these systems more intuitively and effectively. 

One of the earliest applications of OpenFold was by Meta AI, formerly known as Facebook. Meta AI recently used OpenFold to integrate a ‘protein language model’ to launch an atlas featuring over 600 million proteins from bacteria, viruses, and other microorganisms that had not yet been characterized. 

Bouatta explained that living organizations are also organized in a language, referring to the four bases of DNA – adenine, cytosine, guanine, and thymine. \”This is the language that nature picked to build these sophisticated living organisms.\”

He further elaborated that proteins have a second layer of language, represented by the 20 amino acids that make up all proteins in the human body and determine their functions. While genome sequencing has gathered extensive data on these biological “letters”, a crucial piece that has been missing is a “dictionary” that can translate this data into predicting shapes. 

“Machine learning allows us to take a string of letters, the amino acids that describe any kind of protein that you can think of, run a sophisticated algorithm, and return an exquisite three-dimensional structure that is close to what we get using experiments. The OpenFold algorithm is very sophisticated and uses new developments that we\’re familiar with from ChatGPT and others,” said Bouatta. 

The research was supported by Flatiron Institute, OpenBioML, Stability AI, the Texas Advanced Computing Center (TACC), and NVIDIA, all of whom provided the resources needed for the experiments described in this paper.

TACC provided the OpenFold team access to Lonestar6 and Frontera supercomputers, enabling large-scale machine learning and AI deployments that significantly accelerated their research and computational capabilities. 

Related Items 

NCSA’s SEAS Team Makes Advanced Computing More Efficient and Accessible 

The Path to Insight Is Changing: The AI-HPC Paradigm Shift 

Nvidia Taps Into Generative AI Fervor with Unveiling of AI Foundations Cloud Services