AI & Robotics News and Discussions

[Yuli Ban]

Azure AI milestone: Microsoft KEAR surpasses human performance on CommonsenseQA benchmark

Last month, our Azure Cognitive Services team, comprising researchers and engineers with expertise in AI, achieved a groundbreaking milestone by advancing commonsense language understanding. When given a question that requires drawing on prior knowledge and five answer choices, our latest model— KEAR, Knowledgeable External Attention for commonsense Reasoning—performs better than people answering the same question, calculated as the majority vote among five individuals. KEAR reaches an accuracy of 89.4 percent on the CommonsenseQA leaderboard compared with 88.9 percent human accuracy. While the CommonsenseQA benchmark is in English, we follow a similar technique for multilingual commonsense reasoning and topped the X-CSR leaderboard.

Although recent large deep learning models trained with big data have made significant breakthroughs in natural language understanding, they still struggle with commonsense knowledge about the world, information that we, as people, have gathered in our day-to-day lives over time. Commonsense knowledge is often absent from task input but is crucial for language understanding. For example, take the question “What is a treat that your dog will enjoy?” To select an answer from the choices salad, petted, affection, bone, and lots of attention, we need to know that dogs generally enjoy food such as bones for a treat. Thus, the best answer would be “bone.” Without this external knowledge, even large-scale models may generate incorrect answers. For example, the DeBERTa language model selects “lots of attention,” which is not as good an answer as “bone.”

On the other hand, expert systems with lots of rules and domain knowledge and little data have failed to deliver their promise of AI that understands and reasons more like people do. We revisit the rules and knowledge approach and find that deep learning models and knowledge can be organically combined via an external attention mechanism to achieve breakthroughs in AI. With KEAR, we specifically equip language models with commonsense knowledge from a knowledge graph, dictionary, and publicly available machine learning data.

[weatheriscool]

Machine learning used to predict synthesis of complex novel materials
https://phys.org/news/2021-12-machine-s ... rials.html
by Northwestern University
Machine learning enables materials discovery. Credit: Northwestern University

Scientists and institutions dedicate more resources each year to the discovery of novel materials to fuel the world. As natural resources diminish and the demand for higher value and advanced performance products grows, researchers have increasingly looked to nanomaterials.

Nanoparticles have already found their way into applications ranging from energy storage and conversion to quantum computing and therapeutics. But given the vast compositional and structural tunability nanochemistry enables, serial experimental approaches to identify new materials impose insurmountable limits on discovery.

Now, researchers at Northwestern University and the Toyota Research Institute (TRI) have successfully applied machine learning to guide the synthesis of new nanomaterials, eliminating barriers associated with materials discovery. The highly trained algorithm combed through a defined dataset to accurately predict new structures that could fuel processes in clean energy, chemical and automotive industries.

"We asked the model to tell us what mixtures of up to seven elements would make something that hasn't been made before," said Chad Mirkin, a Northwestern nanotechnology expert and the paper's corresponding author. "The machine predicted 19 possibilities, and, after testing each experimentally, we found 18 of the predictions were correct."

The study, "Machine learning-accelerated design and synthesis of polyelemental heterostructures," will be published December 22 in the journal Science Advances.

[Yuli Ban]

CEO of Serve Robotics, Ali Kashani, joins Yahoo Finance to talk about the goal of replacing 5% of food delivery with robots.

- So one thing we've learned during the pandemic is that not only have traditional businesses had to adapt, but the whole concept of delivery, whether it be food, retail delivery, it is adapting, and technology is leading the way. We want to bring back into the stream, Ali Kashani, Serve Robotics CEO and former head of Postmates X at Uber, it's good to have you back. In fact, the last time we talked about this issue, the goal was to replace I think it was 5% of food delivery with the robots. And let's just face the facts, your robots are so darn cute, those little bots that you're testing. When is this really going to take place? Because many of us in New York City are tired of dodging the bicyclists who are going to run us over when they run the red lights.

[Yuli Ban]

Robot 509 from Starship Technologies is both patient and skittish. The autonomous machine, which resembles a Yeti cooler crossed with a Waymo minivan, moves around like a mammal near the bottom of the food chain. It freezes up in crowds and, even when utterly alone, scoots forward in halting spurts, seemingly suspicious of fallen leaves.

A few days after Thanksgiving, Robot 509 ferries some cargo across the campus of James Madison University in Harrisonburg, Va. It stops on a deserted road, waiting for about a minute before finally scooting across, then trundles over a stretch of pavement and some railroad track before coming to a full stop.

Rachael Haberstroh, a James Madison administrator whose office is a long walk from most of the campus’s cafes, comes over and taps on her phone. “It’s supposed to play me a song,” she says, watching skeptically as the entire top of the robot swings open. As 509 serenades her with Adele’s Easy on Me, she reaches in and grabs a large, iced Starbucks drink.

James Madison University has about 50 Starship Technologies autonomous robots rolling across campus. PHOTOGRAPHER: GREG KAHN FOR BLOOMBERG BUSINESSWEEK

[weatheriscool]

Using game theory to thwart multistage privacy intrusions when sharing data
https://phys.org/news/2021-12-game-theo ... ivacy.html
by Thamarasee Jeewandara , Phys.org

Biomedical data is widely collected in the field of medicine, although sharing such data can raise privacy concerns about the re-identification of seemingly anonymous records. Risk assessment frameworks for formal re-identification can inform decisions on the process of sharing data, and current methods focus on scenarios where data recipients use only one resource to identify purposes. However, this can affect privacy where adversaries can access multiple resources to enhance the chance of their success. In a new report now in Science Advances, Zhiyu Wan and a team of scientists in electrical engineering and computer engineering and biomedical informatics in the U.S. represented a re-identification game using a two-player Stackelberg game of perfect information to assess risk. They suggest an optimal data-sharing strategy based on a privacy-utility trade-off. The team used experiments with large-scale genomic datasets and game theoretic models to induce adversarial capabilities to effectively share data with low re-identification risk.

[Yuli Ban]

New footage of a lifelike android fitted with cutting-edge artificial intelligence reacting to engineers invading its “personal space” has spooked the internet.

Britain-based tech company Engineered Arts this month revealed a scarily realistic robot with lifelike facial features and expressions that can “develop interaction” with people. The android, nicknamed Ameca, uses high resolution cameras for eyes to scan its surroundings and react accordingly.

Now available for purchase or rent for an undisclosed sum, the relatable android has a grey face and movable arms and is billed as “the perfect humanoid robot platform for human-robot interaction.”

It can smile, routinely blink its eyes, gasp in shock, scratch its nose — or even have a staring contest with an owner, just for the heck of it — along with plenty more hi-tech features.

According to the company, the can “strike an instant rapport with anybody” due to its personlike nature.

“Human-like artificial intelligence needs a human-like artificial body,” Engineered Arts wrote of Ameca.

[Yuli Ban]

[Yuli Ban]

Researchers in China say they have achieved a world first by developing a machine that can charge people with crimes using artificial intelligence.
The AI “prosecutor” can file a charge with more than 97 per cent accuracy based on a verbal description of the case, according to the researchers.
The machine was built and tested by the Shanghai Pudong People’s Procuratorate, the country’s largest and busiest district prosecution office.
The technology could reduce prosecutors’ daily workload, allowing them to focus on more difficult tasks, according to Professor Shi Yong, director of the Chinese Academy of Sciences’ big data and knowledge management laboratory, who is the project’s lead scientist.

[Nanotechandmorefuture]

It would be nice to see that in action. I will say that would probably be freaky in a way.

[Lurking]

It would be nice to see that in action. I will say that would probably be freaky in a way.

It seems to be a joke

[Yuli Ban]

Tang Jie, the Tsinghua University professor leading the Wu Dao project, said in a recent interview that the group built an even bigger, 100 trillion-parameter model in June, though it has not trained it to “convergence,” the point at which the model stops improving. “We just wanted to prove that we have the ability to do that,” Tang said.

This isn’t simple one-upmanship. On the one hand, it’s how research progresses. But on the other, it is emblematic of an intensifying competition between the world’s two technology superpowers. Whether the researchers involved like it or not, their governments are eager to adopt each AI advance into their national security infrastructure and military capabilities.

That matters, because dominance in the technology means probable victory in any future war. Even more important, such an advantage likely guarantees the longevity and global influence of the government that wields it. Already, China is exporting its AI-enabled surveillance technology—which can be used to quash dissent—to client states and is espousing an authoritarian model that promises economic prosperity as a counter to democracy, something that the Soviet Union was never able to do.

Ironically, China is a competitor that the United States abetted. It’s well known that the U.S. consumer market fed China’s export engine, itself outfitted with U.S. machines, and led to the fastest-growing economy in the world since the 1980s. What’s less well-known is how a handful of technology companies transferred the know-how and trained the experts now giving the United States a run for its money in AI.

gwern

They obviously don't have the compute to train even a 100t MoE (yet, maybe next year), but simply training a 100t-parameterfor a few steps is still quite impressive. (Note that it's not necessarily that impressive as there's already been embeddings/recommender models around that scale.)

[Yuli Ban]

Artificial Intelligence Can Now Craft Original Jokes—And That’s No Laughing Matter

Don’t you hate it,” says Jon the Robot, gesturing with tiny articulated arms at an expectant crowd, “when you’re trying to solve inverse kinematics equations to pick up a cup and then you get ‘Error 453, no solution found’?” The crowd laughs. “Don’t you hate that?”

An experiment billed as a comedy act, Jon is the brainchild of Naomi Fitter, an assistant professor in the School of Mechanical, Industrial and Manufacturing Engineering at Oregon State University. The tiny android performs when a handler (who must also hold the mic) presses a button, then tells the same jokes in the same order, like a grizzled veteran comic at a down-market Vegas casino.

But the robot’s act is more human than it might first appear. Jon is learning how to respond to its audience—it can now vary the timing of its delivery based on the length of the audience’s laughter, and append different responses to jokes based on the level of noise in the room. It can deliver one line if a joke gets a roar of laughter (“Please tell the booking agents how funny that joke was”) and another if there are crickets (“Sorry about that. I think I got caught in a loop. Please tell the booking agents that you like me … that you like me … that you like me”).

The prospect of an AI that understands why we are laughing, and that can generate its own genuinely funny material, is sort of a holy grail for a subset of AI researchers. Artificial intelligence can diagnose tumors, read maps and play games, often faster and with more accuracy than humans can.

[caltrek]

The Steep Cost of Capture
by Meredith Whittaker

https://mags.acm.org/interactions/novem ... Id1742681

Introduction:

(Association for Computer Machinery) This is a perilous moment. Private computational systems marketed as artificial intelligence (AI) are threading through our public life and institutions, concentrating industrial power, compounding marginalization, and quietly shaping access to resources and information.

In considering how to tackle this onslaught of industrial AI, we must first recognize that the “advances” in AI celebrated over the past decade were not due to fundamental scientific breakthroughs in AI techniques. They were and are primarily the product of significantly concentrated data and compute resources that reside in the hands of a few large tech corporations. Modern AI is fundamentally dependent on corporate resources and business practices, and our increasing reliance on such AI cedes inordinate power over our lives and institutions to a handful of tech firms. It also gives these firms significant influence over both the direction of AI development and the academic institutions wishing to research it. Meaning that tech firms are startlingly well positioned to shape what we do—and do not—know about AI and the business behind it, at the same time that their AI products are working to shape our lives and institutions.

Examining the history of the U.S. military’s influence over scientific research during the Cold War, we see parallels to the tech industry’s current influence over AI. This history also offers alarming examples of the way in which U.S. military dominance worked to shape academic knowledge production, and to punish those who dissented.

Today, the tech industry is facing mounting regulatory pressure, and is increasing its efforts to create tech-positive narratives and to silence and sideline critics in much the same way the U.S. military and its allies did in the past. Taken as a whole, we see that the tech industry’s dominance in AI research and knowledge production puts critical researchers and advocates within, and beyond, academia in a treacherous position. This threatens to deprive frontline communities, policymakers, and the public of vital knowledge about the costs and consequences of AI and the industry responsible for it—right at the time that this work is most needed.

[Yuli Ban]

[Lurking]

It seems to be a joke

I think it is "safe" to say that this post aged like spoiled milk.

[Ozzie guy]

I checked out a cold fusion video on Ameca from a month ago.

Unfortunately from my understanding it lacks the AI ability to communicate and a human is telling it what to say.

It may be much dumber than we think.

[weatheriscool]

A new framework that could simplify imitation learning in robotics
https://techxplore.com/news/2022-01-fra ... otics.html
by Ingrid Fadelli , Tech Xplore

Over the past few decades, computer scientists have been trying to train robots to tackle a variety of tasks, including house chores and manufacturing processes. One of the most renowned strategies used to train robots on manual tasks is imitation learning.

As suggested by its name, imitation learning entails teaching a robot how to do something using human demonstrations. While in some studies this training strategy achieved very promising results, it often requires large and annotated datasets containing hundreds of videos where humans complete a given task.

Researchers at New York University have recently developed VINN, an alternative imitation learning framework that does not necessarily require large training datasets. This new approach, presented in a paper pre-published on arXiv, works by decoupling two different aspects of imitation learning, namely learning a task's visual representations and the associated actions.

"I was interested in seeing how we can simplify imitation learning," Jyo Pari, one of the researchers who carried out the study, told TechXplore. "Imitation learning requires two fundamental components; one is learning what is relevant in your scene and the other is how you can take the relevant features to perform a task. We wanted to decouple these components, which are traditionally coupled into one system, and understand the role and importance of each of them."

[Yuli Ban]

As the hospitality industry continues to experience major staffing shortages, one restaurant is attempting to continue serving customers in a creative way.

Eat District in Boca Raton, Florida, recently introduced a new server named Bella. Developed by Pudu Robotics, she is touted as providing "an unprecedented food delivery robot experience." She debuted this week to Eat District patrons, who reportedly loved the new waiter.

"I thought that was really cool how the robot served the food," one patron told local news station WPBF 25. Another commented on how the robot was able to efficiently deliver food to their table without error.

Bella costs between $10,000 and $20,000 and is one of several similar robots made by Pudu Products.

[Ozzie guy]

Moores law is slow compared to alot of AI predictions and if anything I would assume AI develops slower than moores law so I think most AI predictions are unfortunately overly optimistic. I will still give tiny weight to Dr_singularity and massive weight to Yuli Ban, Ray Kurzweil and starspawn0. For example at moores laws rates AI will be less than 3 times better by the start of 2025 (probably about 2.75 times) and a tini bit under 16 times better by the start of 2030 and we are doubling something that is very small/weak.

[Yuli Ban]

The pursuit of perfect poker goes back at least as far as the 1944 publication of “Theory of Games and Economic Behavior,” by the mathematician John von Neumann and the economist Oskar Morgenstern. The two men wanted to correct what they saw as a fundamental imprecision in the field of economics. “We wish,” they wrote, “to find the mathematically complete principles which define ‘rational behavior’ for the participants in a social economy, and to derive from them the general characteristics of that behavior.” Economic life, they suggested, should be thought of as a series of maximization problems in which individual actors compete to wring as much utility as possible from their daily toil. If von Neumann and Morgenstern could quantify the way good decisions were made, the idea went, they would then be able to build a science of economics on firm ground.

It was this desire to model economic decision-making that led them to game play. Von Neumann rejected most games as unsuitable to the task, especially those like checkers or chess in which both players can see all the pieces on the board and share the same information. “Real life is not like that,” he explained to Jacob Bronowski, a fellow mathematician. “Real life consists of bluffing, of little tactics of deception, of asking yourself what is the other man going to think I mean to do. And that is what games are about in my theory.” Real life, von Neumann thought, was like poker.

“Theory of Games” pointed the way to a future in which all manner of competitive interactions could be modeled mathematically: auctions, submarine warfare, even the way species compete to pass their genes on to future generations. But in strategic terms, poker itself barely advanced in response to von Neumann’s proof until it was taken up by members of the Department of Computing Science at the University of Alberta more than five decades later. The early star of the department’s games research was a professor named Jonathan Schaeffer, who, after 18 years of work, discovered the solution to checkers. Alberta faculty and students also made significant progress on games as diverse as go, Othello, StarCraft and the Canadian pastime of curling. Poker, though, remained a particularly thorny problem, for precisely the reason von Neumann was attracted to it in the first place: the way hidden information in the game acts as an impediment to good decision making.