Monthly Archives: March 2022

As we saw in Part 1, its possible to get started on a number of important problems by building augmentation infrastructure around the strengths of an artificial intelligence model. Does the model generate text? Build around text. Can it accurately predict 3D structures? Build around 3D structures. But taking an artificial intelligence system completely at face value comes with its own limitations.

Douglas Engelbart used the term co-evolution to describe the way in which humanitys tools and its processes for using those tools adapt and evolve together. Models like GPT-3 and DALL-E represent a large step in the evolution of tools, but its only one half of the equation. When you build around the model, without also building new tools and processes for the model, youre stuck with what you get. The models weaknesses become your weaknesses. If you dont like the result, its up to you to fix it. And since training any of the large, complex AI systems weve discussed so far requires massive data and computation, you likely dont have the resources to change the model all that much.

This is a bit of a conundrum: On the one hand, we dont have the resources to change the model significantly. On the other hand, we need to change the model, or at least come up with better ways of working with it, to solve for our specific use case. For prompt-based models like GPT-3 and DALL-E, the two easiest ways to tackle this fixed-model conundrum are prompt-hacking and fine-tuning neither of which are particularly efficient:

The goal of augmented intelligence is to make manual processes like these more efficient so humans can spend more time on the things they are good at, like reasoning and strategizing. The inefficiency of prompt-hacking and fine-tuning show that the time is ripe for a reciprocal step in process evolution. So, in this section, well explore some examples of a new theme building for the model and the role it plays in creating more effective augmentation tools.

As a working example, lets say youre an up-and-coming game developer working on the next online gaming franchise. Youve seen how games like Call of Duty and Fortnite have created massively successful (and lucrative) marketplaces for custom skins and in-game assets, but youre a resource-constrained startup. So, instead of developing these assets yourself, you offload content generation to DALL-E, which can generate any number of skins and asset styles for a fraction of the cost. This is a great start, but prompt-hacking your way to a fully stocked asset store is inefficient.

To make things less manual, you can turn prompting over to a text generation model like GPT-3. The key to the virality of a game like Fortnite is the combination of a number of key game assets weapons, vehicles, armor with a variety of unique styles and references, such as eye-catching patterns/colors, superheroes, and the latest pop culture trends. When you seed GPT-3 with your asset types, it can generate any number of these combinations into a prompt. Pass that prompt over to DALL-E, and out comes your skin design.

This GPT-3 to DALL-E handoff sounds great, but it only really works if it produces stimulating, high-quality skin designs for your users. Combing through each of the design candidates manually is not an option, especially at scale. The key here is to build tools that let the marketplace do the work for you. Users flock to good content and have no patience for bad content apps like TikTok are based entirely on this concept. User engagement will therefore be a strong signal for which DALL-E prompts are working (i.e., leading to interesting skin designs) and which are not.

To let your users do the work for you, youll want to build a recursive loop that cross-references user activity with each prompt and translates user engagement metrics into a ranking of your active content prompts. Once you have that, normal A/B testing will automatically surface prompt insights and you can prioritize good prompts, remove bad prompts, and even compare the similarity of newly generated prompts to those you have tested before.

But thats not all the same user engagement signal can also be used for fine-tuning.

Lets move one more step backward and focus on GPT-3s performance. As long as you keep track of the inputs you are giving to GPT-3 (asset types + candidate themes), you can join that data with the quality rankings you have just gotten from further down in your content pipeline to create a dataset of successful and unsuccessful input-output pairs. This dataset can be used to fine-tune GPT-3 on game-design-focused prompt generation, making it even better at generating prompts for your application.

This user-driven cyclical pipeline helps DALL-E generate better content for your users by surfacing the best prompts, and helps GPT-3 generate better prompts by fine-tuning on examples generated from your own user activity. Without having to worry about prompt-hacking and fine-tuning, you are free to work on bigger-ticket items, like which assets are next in the pipeline, and which new content themes might lead to even more interesting skins down the road.

There also exists a huge opportunity to build middleware connecting creative industries and creative, personalized content-generating models.AI models and the services they enable (e.g. Copilot) could help for use cases that require novel content creation. This, again, requires using our understanding of the AI system and how it works to think of ways in which we can modify its behavior ever so slightly to create new and better experiences.

Imagine you are building a service for learning to code that uses Copilot under the hood to generate programming exercises. Out of the box, Copilot will generate anywhere from a single line of code to a whole function, depending on the docstring its given as input. This is great you can construct a bunch of exercises really quickly!

To make this educational experience more engaging, though, youll probably want to tailor the exercises generated by Copilot to the needs and interests of your users. For example, you might want to personalize across dimensions such as:

Generating docstrings yourself is tedious and manual, so personalizing Copilots outputs should be as automated as possible. Well, we know of another AI system, GPT-3, that is great at generating virtually any type of text so maybe we can offload the docstring creation to GPT-3.

This can be done in one of two ways. One approach is to ask GPT-3 to generate generic docstrings that correspond to a particular skill or concept (e.g., looping, recursion, etc.). With one prompt, you can generate any number of boilerplate docstrings. Then, using a curated list of target themes and keywords (a slight manual effort), you can replace variable names in the boilerplate to your target audience. Alternatively, you can try feeding both target skills/concepts and themes to GPT-3 at the same time and let GPT-3 tailor the docstrings to your themes automatically.

The success of this idea, of course, comes down to the quality of GPT-3s content. For one, youll want to make sure the exercises generated by this GPT/Copilot combination are age-appropriate. Perhaps an aligned model like InstructGPT would be better here.

We are now over a decade into the latest AI summer. The flurry of activity in the AI community has led to incredible breakthroughs that will have significant impact across a number of industries and, possibly, on the trajectory of humanity as a whole. Augmented intelligence represents an opportunity to kickstart this progress, and all it takes is a slight reframing of our design principles for building AI systems. In addition to building models to solve problems, we can think of new ways to build infrastructure around models and for models; and even ways in which foundation models might work together (like GPT-3 and DALL-E or GPT-3 + CoPilot).

Maybe one day we will be able to offload all of the dirty work of life to some artificial general intelligence and live hakuna-matata style, but until that day comes we should think of Engelbart focusing less on machines that replace human intelligence and more about those that are savvy enough to enhance it.

Posted March 8, 2022

Technology, innovation, and the future, as told by those building it.

Check your inbox for a welcome note.

Go here to read the rest:

AlphaFold, GPT-3 and How to Augment Intelligence with AI (Pt. 2) - A16Z Future

Research to focus on AI, ML, Routing and Quantum Networking for advanced communication

SUNNYVALE, Calif., March 08, 2022--(BUSINESS WIRE)--Juniper Networks (NYSE: JNPR), a leader in secure, AI-driven networks, today announced a university funding initiative to fuel strategic research to advance network technologies for the next decade. Junipers goal is to enable universities, including Dartmouth, Purdue, Stanford and the University of Arizona, to explore next-generation network solutions in the fields of artificial intelligence (AI) and machine learning (ML), intelligent multipath routing and quantum communications.

Investing now in these technologies, as organizations encounter new levels of complexity across enterprise, cloud and 5G networks, is critical to replace tedious, manual operations as networks become mission critical for nearly every business. This can be done through automated, closed-loop workflows that use AI and ML-driven operations to scale and cope with the exponential growth of new cloud-based services and applications.

The universities Juniper selected in support of this initiative are now beginning the research that, once completed, will be shared with the networking community. In addition, Juniper joined the Center for Quantum Networks Industrial Partners Program to fund industry research being spearheaded by the University of Arizona.

Supporting Quotes:

"Cloud services will continue to proliferate in the coming years, increasing network traffic and requiring the industry to push forward on innovation to manage the required scale out architectures. Junipers commitment to deliver better, simpler networks requires us to engage and get ahead of these shifts and work with experts in all areas in order to trailblaze. I look forward to collaborating with these leading universities to reach new milestones for the network of the future."

- Raj Yavatkar, CTO, Juniper Networks

"With internet traffic continuing to grow and evolve, we must find new ways to ensure the scalability and reliability of networks. We look forward to exploring next-generation traffic engineering approaches with Juniper to meet these challenges."

Story continues

- Sonia Fahmy, Professor of Computer Science, Purdue University

"It is an exciting opportunity to work with a world-class partner like Juniper on cutting edge approaches to next-generation, intelligent multipath routing. Dartmouth's close collaboration with Juniper will combine world-class skills and technologies to advance multipath routing performance."

- George Cybenko, Professor of Engineering, Dartmouth University

"As network technology continues to evolve, so do operational complexities. The ability to utilize AI and machine learning will be critical in keeping up with future demands. We look forward to partnering with Juniper on this research initiative and finding new ways to drive AI forward to make the network experience better for end users and network operators."

- Jure Leskovec, Associate Professor of Computer Science, Stanford University

"The internet of today will be transformed through quantum technology which will enable new industries to sprout and create new innovative ecosystems of quantum devices, service providers and applications. With Juniper's strong reputation and its commitment to open networking, this makes them a terrific addition to building this future as part of the Center for Quantum Networks family."

- Saikat Guha, Director, NSF Center for Quantum Networks, University of Arizona

About Juniper Networks

Juniper Networks is dedicated to dramatically simplifying network operations and driving superior experiences for end users. Our solutions deliver industry-leading insight, automation, security and AI to drive real business results. We believe that powering connections will bring us closer together while empowering us all to solve the worlds greatest challenges of well-being, sustainability and equality. Additional information can be found at Juniper Networks (www.juniper.net) or connect with Juniper on Twitter, LinkedIn and Facebook.

Juniper Networks, the Juniper Networks logo, Juniper, Junos, and other trademarks listed here are registered trademarks of Juniper Networks, Inc. and/or its affiliates in the United States and other countries. Other names may be trademarks of their respective owners.

category-corporate

View source version on businesswire.com: https://www.businesswire.com/news/home/20220308005449/en/

Contacts

Dan MuozJuniper Networks+ 1 (408) 936-2145dmunoz@juniper.net

Continue reading here:

Juniper Networks Announces University Research Funding Initiative to Advance Artificial Intelligence and Network Innovation - Yahoo Finance

Researchers at Pacific Northwest National Laboratory (PNNL) and Micron are are developing an advanced memory system to support AI for scientific computing. The work is designed to address AIs insatiable demand for live data to push the boundaries of memory-bound AI applications by connecting memory across processors in a technology strategy utilizing the Compute Express Link (CXL) data interface, according to a recent edition of the ASCR Discovery publication.

Most of the performance improvements have been on the processor side, said James Ang, PNNLs chief scientist for computing and the labs project leader. But recently, weve been falling short on performance improvements, and its because were actually more memory-bound. That bottleneck increases the urgency and priority in memory resource research.

Boise-Idaho-based memory and storage semiconductor company Micron is collaborating with PNNL, Richland, WA, on this effort, sponsored by the Advanced Scientific Computing Research (ASCR) program in the Department of Energy (DOE), to help assess emerging memory technologies for DOE Office of Science projects that employ artificial intelligence. The partners say they will apply CXL to join memory from various processing units deployed for scientific simulations.

Tony Brewer, Microns chief architect of near-data computing, says the collaboration aims to blend old and new memory technologies to boost high-performance computing (HPC) workloads. We have efforts that look at how we could improve the memory devices themselves and efforts that look at how we can take traditional high-performance memory devices and run applications more efficiently.

Part of the strategy is to implement a centralized memory pool would help mitigate the issue of over-provisioning the memory.

In HPC systems that deploy AI, high performance but low-capacity memory (typically gigabytes in capacity) is typically coupled to the GPUs, whereas a conventional system with low-performance but high capacity memory (terabytes) is loosely coupled via the traditional HPC workhorses, central processing units (CPUs), PNNL said. With PNNL, Micron will create proof-of-concept shared GPU and CPU systems and combine them with additional external storage devices in the hundreds of terabytes range. Future systems will need rapid access to petabytes of memory a thousand times more capacity than on a single GPU or CPU.

The intent is to create a third level of memory hierarchy, Brewer explains. The host would have some local memory, the GPU would have some local memory, but the main capacity memory is accessible to all compute resources across a switch, which would allow scaling of much larger systems. This unified memory would let researchers using deep-learning algorithms to run a simulation while its results simultaneously feed back to the algorithm.

A centralized memory system could also benefit operations because an algorithm or scientific simulation can share data with, say, another program thats tasked with analyzing those data. These converged application workflows are typical in DOEs scientific discovery challenges. Sharing memory and moving it around involves other technical resources, says Andrs Mrquez, a PNNL senior computer scientist. This centralized memory pool, on the other hand, would help mitigate the issue of over-provisioning the memory.

Because AI-aided data-driven science drives up demand for memory, an application cant afford to partition and strand the memory. The result: memory keeps piling up underutilized at various processing units. Having the capability of reducing that over-provisioning and getting more bang out of your buck by sharing that data across all those devices and different stages of workflow cannot be overemphasized, Mrquez explained.

Some of PNNLs AI algorithms can underperform when memory is slow to access, Mrquez says. In PNNLs computational chemistry group, for instance, researchers use AIto study waters molecular dynamics to see how it aggregates and interacts with other compounds. Water is a common solvent for commercial processes, so running simulations to understand how it acts with a molecule of interest is important. A separate research team at Richland is using AI and neural networks to modernize the power grids transmission lines.

Microns Brewer said he looks forward not only to the development of tools with PNNL but also for commercial use by any company working on large-scale data analysis. We are looking at algorithms, he said, and understanding how we can advance these memory technologies to better meet the needs of those applications.

PNNLs computational science problems provide Micron a way to observe applications that will most stress the memory.Those findings will help Brewer and colleagues develop products that help industry meet its memory requirements.

Ang, too, said he expects the project to help AI at large, pointing out that the Micron partnership isnt just a specialized one-off for DOE or scientific computing. The hope is that were going to break new ground and understand how we can support applications with pooled memory in a way that can be communicated to the community through enhancements to the CXL standard.

See the original post here:

PNNL and Micron Partner to Push Memory Boundaries for HPC and AI - insideHPC

(TNS) The Pittsburgh Public Schools could soon implement an artificial intelligence system that captures information from vehicles that illegally pass stopped school buses and sends fines to their owners in an attempt to deter repeat infractions.

The city school board this month could agree to launch a pilot program with BusPatrol, a tech company that uses artificial and machine learning to promote safety for students traveling to and from school.

According to the company, which is based in Lorton, Va., and recently opened a Pennsylvania office in the city of Allentown, more than 136,000 school bus-related injuries and more than 1,000 fatalities have occurred in the past decade.

BusPatrol has partnered with school systems in several states and began working with districts in Pennsylvania a couple of years ago. The company installs software on school buses that has the capability of monitoring the vehicle's surroundings.

If a vehicle illegally passes a stopped bus, the device collects video of the offense and other information, including a license plate number, and turns it into an evidence package that it sends to police. If police then approve the citation, BusPatrol prints it and mails it to the vehicle owner, who can go online and view a video of their vehicle passing a stopped bus.

BusPatrol said the program has proven to increase safety because 98% of offenders do not get cited by the company a second time.

Mr. Souliere said the cost of the program is completely paid for by the ticket revenue, and the school district gets a large chunk of the $300 citation. A conservative estimate showed the district would receive $500,000 per 100 buses per year that it can invest back into schools, he said. About 500 buses carry students throughout the district.

School board member Pam Harbin said she had a "very strong reaction" to the idea that the district would have a program paid for by fining people who may simply make a mistake.

"I'm not going to agree to a system that's going to put people in that position," Ms. Harbin said. "I would rather have months and months of education and pay for that to improve behavior rather than saying we're going to harm people in a different way."

Mr. Souliere said the citation is a civil monetary penalty, no points are put on an offender's license, and vehicle insurance is not impacted. If necessary, he said, the fine can be paid over a certain period of time through a payment plan. But he noted that failure to pay the fine could result in the suspension of a license or plates not being renewed.

Before citations start being issued, the company blitzes media in communities where the program is implemented to provide an education or reminder of the law. The company places informational television commercials, works with local media and creates educational videos for schools and other entities.

School board member Tracey Reed said she was concerned about police using the information that the system collects for other purposes, such as fining individuals for other issues with their vehicles. Mr. Souliere, though, said police are not allowed to do that.

"The scope of use by law is exclusively for the enforcement of this specific stop arm infraction," he said. "In fact, if someone were to get caught by a police officer in passing a stopped school bus, the civil penalty would no longer apply it would be the criminal one that would apply."

The software has the ability to film inside the bus and can provide video in the instance of a fight or an accident, according to Mr. Souliere.

Michael McNamara, the district's chief operating officer, said that if the board approves the pilot program, the software will be placed on about 20 buses in various areas of the city. No citations would be issued during the pilot.

If the board approves the full program after the pilot period, the technology would be installed on all school buses with stop arms that serve Pittsburgh students.

"If this is successful and the board is on board with it no pun intended there we would then deploy [the software] over the summer to the rest of our buses that have stop arms, and then start collecting ticket revenue beginning the first day of the new school year next year," Mr. McNamara said. "No tickets would be issued, only warnings, the rest of this year, and then we would be able to educate the rest of the summer and have full deployment first day of school next fall."

2022 the Pittsburgh Post-Gazette, Distributed by Tribune Content Agency, LLC.

Go here to see the original:

Pittsburgh May Use AI to Fine Those Who Pass School Buses - Government Technology

PALO ALTO, Calif., March 8, 2022 /PRNewswire/ -- Wolfpack, the free mobile app with 12,000+ users on its waitlist, announced its iPhone app is available for download here. The app makes it easy for investors to discover investment opportunities by selecting several investment filters suited to their personal investment needs. An Artificial Intelligence (AI)-driven technology provides a list of investment opportunities.

"With Wolfpack, beginner investors discover, trade, and grow their wealth over time with an AI Discovery Engine."

"There are many investment apps out there, but they all fail to solve one fundamental problem, 'What should I invest in?' Wolfpack's proprietary AI allows users to discover investment opportunities that enables them to find securities they might otherwise miss," said George Parthimos, Chief Executive Officer (CEO) and founder of Wolfpack.

"With Wolfpack, beginner investors discover, trade, and grow their wealth over time with an AI Discovery Engine that finds trending investments. Users receive daily notifications that align with their personal investment goals," according to Nicholas Kapes, Chairman of Wolfpack.

In addition to the AI Discovery Engine, the app offers investors the ability to follow top-performing wolves and receive their top picks, join discussion boards, and receive referral credits when referring friends. The AI tool also ranks the most popular investments across the entire Wolfpack community with the free service.

Brokerage products and services are provided by Apex Clearing Corporation. Apex Clearing is the licensed broker-dealer for Wolfpack and is one of the largest clearing houses in the United States.

WOLFPACK'S TOP FEATURES:

Users can invest as little as $5. Wolfpack is designed for beginnersright through to the most experienced of investors.

"Wolfpack is an incredibly powerful tool with ground-breaking features that can change lives for investors looking to build long-term wealth," said Parthimos.

DOWNLOAD NOW

Wolfpack is available for download from the App Store today.

Android app coming end-March 2022.

About Wolfpack Financial Inc.

Headquartered in Palo Alto, California, Wolfpack Financial Inc. has launched its ingenious mobile app, Wolfpack. Designed to empower the millennial generation to build sustained wealth, the app offers investors the opportunity to discover stocks and ETFs which perfectly match their selection criteria.

Contact: Monica Matulich PRHollywood [emailprotected] 310-383-9502

SOURCE Wolfpack Financial Inc.

Original post:

Wolfpack Uses AI to Solve the Biggest Problem Facing Every Investor: What Should I Invest In? - PR Newswire

Social media giant Meta Platforms, formerly known as Facebook, is only the latest company to draw legal heat over its technology -- specifically, its artificial intelligence (AI) innovations. In this episode of "The AI/ML Show" on Motley Fool Live, recorded on Feb. 16, Fool.com contributors Toby Bordelon and Jason Hall discuss how the debate of AI versus privacy continues to rage on.

Toby Bordelon: We talked about data protection and privacy, I think, a decent amount with Facebook, and you can see what happens when that goes badly. If you don't follow those rules, $650 million with maybe more to come, and that can put a damper on what you can do. You want data to train AI well. You want data to be free-flowing, but then how does that work with our existing laws? Do we need to change them? Do we as a society need to get to a point where we say, you know what, we have to just allow use of personal data or it's not going to get us to where we want to be. We don't have to do that. It's a choice to be made. But there are trade-offs each way.

Jason Hall: It's like somebody refusing to not use cash or write checks. You can, but you're also unable to participate fully or easily in the way that most people do commerce.

Bordelon: I think with AI, too, there's a level beyond that. Because if say, Jose says, "I don't want my data being used to train this AI, I don't want it to be used at all." Does that impact how good the AI is, and does that impact my experience with the AI? Where is that line? It's not a new debate. It's a classic debate about where does individual rights end and communal rights begin, or what are you required to give up as an individual to live in the communal society. We've been having that debate...

Hall: As long as there's been a society.

Bordelon: Thousands of years. Exactly, and this is just another iteration of that, that we need to have a conversation around and struggle with, I think. You think about pace of innovation, which you touched on, Jason. The innovation in this field gets ahead of the law. We see when we have that a little bit with Facebook, but what ends up happening is that courts decide issues without a great legal framework because it's an issue of first impression. They have never seen it before. They are being asked to interpret existing laws that were written before the technology exists, and they have to wing it. That's not awesome. As a society, I don't think we want judges being forced into making decisions that have particularly billion-dollar impacts, and real impacts on people's lives using 30-year-old laws.

Hall: Using 30-year-old laws that weren't written to apply to a thing that didn't exist, and having no basis for understanding what they're ruling on.

Bordelon: Right. People yell at judges for getting that wrong, but that's unfair to put them in that position to begin with, I think. The laws just have to keep up. We have to find a way to anticipate things better, I think, not always react with our legislation, so that when things come up in the court, the judge can say, "OK, I have been given a framework by legislatures, with which I can work to try to find a resolution dispute." I've been saying I'm going to use a framework that's 50 years old because that's all anyone's given me. That's what the law is. We're just going to go with it and make the best of it we can. That's not the best way to do things. But that's kind of where we fall into a lot of technology, including AI. That's got to be addressed at some point.

The rest is here:

Will We Have to Relinquish Some Privacy for the Best AI? - The Motley Fool

The green energy revolution promised by nuclear fusion is now a step closer, thanks to the first successful use of a cutting-edge artificial intelligence system to shape the superheated hydrogen plasmas inside a fusion reactor.

The successful trial indicates that the use of AI could be a breakthrough in the long-running search for electricity generated from nuclear fusion bringing its introduction to replace fossil fuels and nuclear fission on modern power grids tantalizingly closer.

"I think AI will play a very big role in the future control of tokamaks and in fusion science in general," Federico Felici, a physicist at the Swiss Federal Institute of Technology in Lausanne (EPFL) and one of the leaders on the project, told Live Science. "There's a huge potential to unleash AI to get better control and to figure out how to operate such devices in a more effective way."

Related: Fission vs. fusion: What's the difference?

Felici is a lead author of a new study describing the project published in the journal Nature. He said future experiments at the Variable Configuration Tokamak (TCV) in Lausanne will look for further ways to integrate AI into the control of fusion reactors. "What we did was really a kind of proof of principle," he said. "We are very happy with this first step."

Felici and his colleagues at the EPFL's Swiss Plasma Center (SPC) collaborated with scientists and engineers at the British company DeepMind a subsidiary of Google owners Alphabet to test the artificial intelligence system on the TCV.

The doughnut-shaped fusion reactor is the type that seems most promising for controlling nuclear fusion; a tokamak design is being used for the massive international ITER ("the way" in Latin) project being built in France, and some proponents think they'll have a tokamak in commercial operation as soon as 2030.

The tokamak is principally controlled by 19 magnetic coils that can be used to shape and position the hydrogen plasma inside the fusion chamber, while directing an electric current through it, Felici explained.

The coils are usually governed by a set of independent computerized controllers one for each aspect of the plasma that features in an experiment that are programmed according to complex control engineering calculations, depending on the particular conditions being tested. But the new AI system was able to manipulate the plasma with a single controller, he said.

The AI a "deep reinforcement learning" (RL) system developed by DeepMind was first trained on simulations of the tokamak a cheaper and much safer alternative to the real thing.

But the computer simulations are slow: It takes several hours to simulate just a few seconds of real-time tokamak operation. In addition, the experimental condition of the TCV can change from day to day, and so the AI developers needed to take those changes into account in the simulations.

When the simulated training process was complete, however, the AI was coupled to the actual tokamak.

The TCV can sustain a superheated hydrogen plasma, typically at more than 216 million degrees Fahrenheit (120 million degrees Celsius), for a maximum of 3 seconds. After that, it needs 15 minutes to cool down and reset, and between 30 and 35 such "shots" are usually done each day, Felici said.

A total of about 100 shots were done with the TCV under AI control over several days, he said: "We wanted some kind of variety in the different plasma shapes we could get, and to try it under various conditions."

Related: Science fact or fiction? The plausibility of 10 sci-fi concepts

Although the TCV wasn't using plasmas of neutron-heavy hydrogen that would yield high levels of nuclear fusion, the AI experiments resulted in new ways of shaping plasmas inside the tokamak that could lead to much greater control of the entire fusion process, he said.

The AI proved adept at positioning and shaping the plasma inside the tokamak's fusion chamber in the most common configurations, including the so-called snowflake shape thought to be the most efficient configuration for fusion, Felici said.

In addition, it was able to shape the plasma into "droplets" separate upper and lower rings of plasma within the chamber which had never been attempted before, although standard control engineering techniques could also have worked, he said.

Creating the droplet shape "was very easy to do with the machine learning," Felici said. "We could just ask the controller to make the plasma like that, and the AI figured out how to do it."

The researchers also saw that the AI was using the magnetic coils to control the plasmas inside the chamber in a different way than would have resulted from the standard control system, he said.

"We can now try to apply the same concepts to much more complicated problems," he said. "Because we are getting much better models of how the tokamak behaves, we can apply these kinds of tools to more advanced problems."

The plasma experiments at the TCV will support the ITER project, a massive tokamak that's projected to achieve full-scale fusion in about 2035. Proponents hope ITER will pioneer new ways of using nuclear fusion to generate usable electricity without carbon emissions and with only low levels of radioactivity.

The TCV experiments will also inform designs for DEMO fusion reactors, which are seen as successors to ITER that will supply electricity to power grids something that ITER is not designed to do. Several countries are working on designs for DEMO reactors; one of the most advanced, Europe's EUROfusion reactor, is projected to begin operations in 2051.

Originally published on Live Science.

Go here to read the rest:

Nuclear fusion is one step closer with new AI breakthrough - Livescience.com