AI & Robotics News and Discussions

[Yuli Ban]

POLAR MANUFACTURING has been making ​metal ​hinges, locks, and brackets ​in south Chicago for more than 100 years. Some of the company’s metal presses—hulking great machines that loom over a worker—date from the 1950s. Last year, to meet rising demand amid a shortage of workers, Polar hired its first robot employee.

[caltrek]

When Should Someone Trust an AI Assistant's Predictions?
by Adam Zewe
January 20, 2022

https://www.eurekalert.org/news-releases/940756

Introduction to News Release:

(MIT via EurekAlert) In a busy hospital, a radiologist is using an artificial intelligence system to help her diagnose medical conditions based on patients’ X-ray images. Using the AI system can help her make faster diagnoses, but how does she know when to trust the AI’s predictions?

She doesn’t. Instead, she may rely on her expertise, a confidence level provided by the system itself, or an explanation of how the algorithm made its prediction — which may look convincing but still be wrong — to make an estimation.

To help people better understand when to trust an AI “teammate,” MIT researchers created an onboarding technique that guides humans to develop a more accurate understanding of those situations in which a machine makes correct predictions and those in which it makes incorrect predictions.

By showing people how the AI complements their abilities, the training technique could help humans make better decisions or come to conclusions faster when working with AI agents.

“We propose a teaching phase where we gradually introduce the human to this AI model so they can, for themselves, see its weaknesses and strengths,” says Hussein Mozannar, a graduate student in the Clinical Machine Learning Group of the Computer Science and Artificial Intelligence Laboratory (CSAIL). “We do this by mimicking the way the human will interact with the AI in practice, but we intervene to give them feedback to help them understand each interaction they are making with the AI.”

[caltrek]

Meta Researchers Build an AI that Learns Equally Well from Visual, Written or Spoken Materials
by David Coldewey
January 20, 2022

https://techcrunch.com/2022/01/20/meta- ... materials/

Introduction:

(TechCrunch) Advances in the AI realm are constantly coming out, but they tend to be limited to a single domain: for instance, a cool new method for producing synthetic speech isn’t also a way to recognize expressions on human faces. Meta (AKA Facebook) researchers are working on something a little more versatile: an AI that can learn capably on its own whether it does so in spoken, written, or visual materials.

The traditional way of training an AI model to correctly interpret something is to give it lots and lots (like millions) of labeled examples. A picture of a cat with the cat part labeled, a conversation with the speakers and words transcribed, etc. But that approach is no longer in vogue as researchers found that it was no longer feasible to manually create databases of the sizes needed to train next-gen AIs. Who wants to label 50 million cat pictures? Okay, a few people probably — but who wants to label 50 million pictures of common fruits and vegetables?

Currently some of the most promising AI systems are what are called self-supervised: models that can work from large quantities of unlabeled data, like books or video of people interacting, and build their own structured understanding of what the rules are of the system. For instance, by reading a thousand books it will learn the relative positions of words and ideas about grammatical structure without anyone telling it what objects or articles or commas are — it got it by drawing inferences from lots of examples.

This feels intuitively more like how people learn, which is part of why researchers like it. But the models still tend to be single-modal, and all the work you do to set up a semi-supervised learning system for speech recognition won’t apply at all to image analysis — they’re simply too different. That’s where Facebook/Meta’s latest research, the catchily named data2vec, comes in.

The idea for data2vec was to build an AI framework that would learn in a more abstract way, meaning that starting from scratch, you could give it books to read or images to scan or speech to sound out, and after a bit of training it would learn any of those things. It’s a bit like starting with a single seed, but depending on what plant food you give it, it grows into an daffodil, pansy, or tulip.

[Ozzie guy]

I have no idea what this means in the practical sense and if it is related to progress in the longterm development of human level AGI, does anyone here know if the news is important or not?

"The Human Brain-Scale AI Supercomputer Is Coming ...
Funded by the Slovakian government using funds allocated by the EU, the I4DI consortium is behind the initiative to build a 64 AI exaflop machine (that’s 64 billion, billion AI operations per second) on our platform by the end of 2022. This will enable Slovakia and the EU to deliver for the first time in the history of humanity a human brain-scale AI supercomputer. Meanwhile, almost a dozen other countries are watching this project closely, with interest in replicating this supercomputer in their own countries."

https://www.forbes.com/sites/forbestech ... f6e65e75a5

[wjfox]

[caltrek]

https://futureoflife.org/

[caltrek]

Future of Life Institute Opens Worldbuilding Contest
January 2022 Newsletter

https://futureoflife.org/2022/01/19/fli ... sletter-2/

Introduction:

The Future of Life Institute (FLI) has launched a Worldbuilding competition with a prize purse worth $100,000 plus. Individuals and teams are invited to design visions of a plausible, aspirational future which includes artificial general intelligence.

Worldbuilding is the art and science of constructing a coherent and relatively detailed fictitious world. It is frequently practised by creative writers and scriptwriters, providing the context and backdrop for stories that take place in future, fantasy or alternative realities. FLI’s contest challenges entrants to use worldbuilding to explore possible futures for our own world.

The contest is designed to appeal broadly, with artistic components as well as more conceptual ones. Whether you’re a writer, an economist, a scientist, an AI researcher, a student, a filmmaker, an expert in geopolitics or simply a Sci-Fi enthusiast, we believe there’s something here for you.

FLI hopes to encourage people to start thinking about the future in more positive terms, particularly with regards to powerful new technologies. In order to steer the technology’s trajectory in a positive direction, we need to know what we’re aiming for. To know what future we would most like, we must first imagine the kinds of futures we could plausibly have. Unfortunately, not nearly enough effort goes into imagining what a good future might look like. Mainstream media tends to focus on the dystopias we could end up in. This contest seeks to change that.

Applications are due on 15th April 2022. If you’d like to attend a worldbuilding workshop or you need help finding team members, visit this page. For more information about the contest and to enter, visit this website.*

*https://worldbuild.ai/?utm_source=Activ ... Newsletter

[caltrek]

Argonne Scientists Use Artificial Intelligence to Improve Airplane Manufacturing

https://www.eurekalert.org/news-releases/941064

Introduction:

(Department of Energy/Argonne National Laboratory via EurekAlert) “It’s like programming a coffeemaker.”

When it comes to manufacturing new lightweight, yet strong components for new passenger jets, scientists are treating the process like trying to brew the most delicious cup of coffee.

By using artificial intelligence (AI) and machine learning, researchers at the U.S. Department of Energy’s (DOE) Argonne National Laboratory are intelligently and automatically selecting the perfect settings for a different kind of hot brew — the process of friction stir welding, a common ingredient needed to manufacture airplane components.

In a new collaboration with GE Research, Edison Welding Institute and GKN Aerospace, Argonne computer scientists are putting the power of the laboratory’s automated machine learning expertise and supercomputers to use. By reducing the number of costly experiments and time-consuming simulations with a new machine learning approach, they can generate accurate models that provide valuable information about the welding process in much less time and at a fraction of the cost.

This approach, called DeepHyper, is a scalable automated machine learning package developed by Argonne computational scientist Prasanna Balaprakash and his colleagues at Argonne. Machine learning is a process by which a computer can train itself to find the best answers to a particular question.

[caltrek]

Boston Dynamics Warehouse Robot Gets a $15 Million Gig Working for DHL
by Brian Heater
January 26, 2022

https://techcrunch.com/2022/01/26/bosto ... g-for-dhl/

Introduction:

(TechCrunch) Back in March of last year, Boston Dynamics unveiled its second commercial robot, Stretch. The system, built from its impressive box-moving Handle concept, is designed to bring the company’s advanced robotics technologies into a warehouse/logistics setting – easily one of the hottest categories in robotics, these days.

Today the Hyundai-owned firm announced its first commercial customer – and it’s a big one. Logistics giant DHL has committed to a multi-year, $15 million deal (or “investment” as the parties are referring to it) set to bring the robot to its North American facilities. Specific details on the number of robots being purchased haven’t been revealed, but Boston Dynamics says it’s going to be bringing a “fleet” of the robots to DHL logistics centers over the next three years.

Stretch will get to work unloading trucks to start – a feature its creators have highlighted as a key part of its initial rollout. Additional tasks will be added, over the course of the roll out, in an effort to further automate the package handling process.

Says CEO Robert Playter, “Stretch is Boston Dynamics’ newest robot, designed specifically to remedy challenges within the warehouse space. We are thrilled to be working with DHL Supply Chain to deliver a fleet of robots that will further automate warehousing and improve safety for its associates. We believe Stretch can make a measurable impact on DHL’s business operations, and we’re excited to see the robot in action at scale.”

[weatheriscool]

Robot performs first laparoscopic surgery without human help

by Johns Hopkins University

A robot has performed laparoscopic surgery on the soft tissue of a pig without the guiding hand of a human—a significant step in robotics toward fully automated surgery on humans. Designed by a team of Johns Hopkins University researchers, the Smart Tissue Autonomous Robot (STAR) is described today in Science Robotics.

"Our findings show that we can automate one of the most intricate and delicate tasks in surgery: the reconnection of two ends of an intestine. The STAR performed the procedure in four animals and it produced significantly better results than humans performing the same procedure," said senior author Axel Krieger, an assistant professor of mechanical engineering at Johns Hopkins' Whiting School of Engineering.

The robot excelled at intestinal anastomosis, a procedure that requires a high level of repetitive motion and precision. Connecting two ends of an intestine is arguably the most challenging step in gastrointestinal surgery, requiring a surgeon to suture with high accuracy and consistency. Even the slightest hand tremor or misplaced stitch can result in a leak that could have catastrophic complications for the patient.

https://techxplore.com/news/2022-01-rob ... human.html

[caltrek]

CausaLens Obtains $45 Million No-code Technology That Introduces Cause and Effect into AI Decision Making
by Ingrid Lunden
January 28, 2022

https://techcrunch.com/2022/01/28/causa ... on-making/

Introduction:

(TechCrunch) One of the most popular applications of artificial intelligence to date has been to use it to predict things, using algorithms trained with historical data to determine a future outcome. But popularity doesn’t always mean success: predictive AI leaves out a lot of the nuance, context and cause-and-effect reasoning that goes into an outcome; and as some have pointed out (and as we have seen), this means that sometimes the “logical” answers produced by predictive AI can prove disastrous. A startup called causaLens has developed causal inference technology — presented as a no-code tool that doesn’t require a data scientist to use to introduce more nuance, reasoning and cause-and-effect sensibility into an AI-based system — which it believes can solve this problem.

CausaLens’s aim, CEO and co-founder Darko Matovski said, is for AI “to start to understand the world as humans understand it.”

Today the startup is announcing $45 million in funding after seeing some early success with its approach, growing revenues 500% since coming out of stealth a year ago. This is being described as a “first close” of the round, meaning it’s still open and potentially going to grow in size.

Dorilton Ventures and Molten Ventures (the VC that rebranded from Draper Esprit) led the round, with previous backers Generation Ventures and IQ Capital, and new backer GP Bullhound also participating. Sources tell us the round values London-based causaLens at around $250 million.

CausaLens’s customers and partners currently include organizations in healthcare, financial services and government, among a number of other verticals, where its technology is used not just for AI-based decision making but to bring in more cause-and-effect nuance when arriving at outcomes.

You can read more about AI and cause and effect reasoning here: https://www.causalens.com/white-paper/h ... nd-effect/

[Yuli Ban]

[Yuli Ban]

Musk: Robots to be bigger business than Tesla cars

Elon Musk likes to have a focus - and this year, it looks like it might be robots.

He told investors on a Tesla earnings call his nascent robot plans had "the potential to be more significant than the vehicle business, over time".

And they would be the most important things Tesla worked on this year.

The robot in question, part of a project dubbed Optimus, was previewed last year - to raised eyebrows - by a human in a robot suit dancing on stage.

And the performance became a popular internet meme.

The Tesla Bot, as it was dubbed, would use the same artificial-intelligence (AI) systems that helped power Tesla vehicles, Mr Musk said at the event last August - but no prototype has yet been made.

[Yuli Ban]

"Chain of Thought Prompting Elicits Reasoning in Large Language Models", Wei (LaMDA inner monologue)

Although scaling up language model size has reliably improved performance on a range of NLP tasks, even the largest models currently struggle with certain reasoning tasks such as math word problems, symbolic manipulation, and commonsense reasoning. This paper explores the ability of language models to generate a coherent chain of thought -- a series of short sentences that mimic the reasoning process a person might have when responding to a question. Experiments show that inducing a chain of thought via prompting can enable sufficiently large language models to better perform reasoning tasks that otherwise have flat scaling curves.

This, if you ask me, is going to be the next area of improvement for transformers: inner voices. It'll improve these models vastly more than merely adding more parameters will.

[Yuli Ban]

How robots learn to hike

Steep sections on slippery ground, high steps, scree and forest trails full of roots: the path up the 1,098-​metre-high Mount Etzel at the southern end of Lake Zurich is peppered with numerous obstacles. But ANYmal, the quadrupedal robot from the Robotic Systems Lab at ETH Zurich, overcomes the 120 vertical metres effortlessly in a 31-minute hike. That’s 4 minutes faster than the estimated duration for human hikers – and with no falls or missteps.
This is made possible by a new control technology, which researchers at ETH Zurich led by robotics professor Marco Hutter recently presented in the journal Science Robotics ("Learning robust perceptive locomotion for quadrupedal robots in the wild").

[caltrek]

Artificial Intelligence System Rapidly Predicts How Two Proteins Will Attach
February 1, 2022

https://www.eurekalert.org/news-releases/942017

Introduction:

(EurekAlert) Antibodies, small proteins produced by the immune system, can attach to specific parts of a virus to neutralize it. As scientists continue to battle SARS-CoV-2, the virus that causes Covid-19, one possible weapon is a synthetic antibody that binds with the virus’ spike proteins to prevent the virus from entering a human cell.

To develop a successful synthetic antibody, researchers must understand exactly how that attachment will happen. Proteins, with lumpy 3D structures containing many folds, can stick together in millions of combinations, so finding the right protein complex among almost countless candidates is extremely time-consuming.

To streamline the process, MIT researchers created a machine-learning model that can directly predict the complex that will form when two proteins bind together. Their technique is between 80 and 500 times faster than state-of-the-art software methods, and often predicts protein structures that are closer to actual structures that have been observed experimentally.

This technique could help scientists better understand some biological processes that involve protein interactions, like DNA replication and repair; it could also speed up the process of developing new medicines.

[Yuli Ban]

An Autonomous Delivery Robot Is the Next Big Thing, Experts Say

In a few years, it will probably be pretty ordinary to see robots delivering packages and transporting goods in cities.

Delivery companies are always on the lookout for the next big innovation in the sector.

In the US, companies like Starship Technologies have seen huge growth as a result of the pandemic, and others like Coco are also scaling quickly.

Some big names have also developed their own autonomous delivery robots, like Alibaba, whose robots are becoming ever more prominent in daily life in China.

Many companies have already launched their first prototypes, but the robots developed by the Polytechnic University of Catalonia (UPC) and CARNET have a standout feature — they're able to go up and down stairs, according to the press release.

It has six wheels and can go up or down steps up to 7.9 inches high. Universitat Politècnica de Catalunya/upc.edu

I'd like to believe that this wasn't going to be the "Next Big Thing" in a world without the COVID-19 pandemic. Perhaps we'd see some delivery robot experiments in the 2020s, but there'd be no overwhelming demand for it driving rapid commercialization until at least the 2030s. Yet we're probably just 3 to 5 years away from this being a major thing in most major (and even some minor) cities.
As always, I might be wrong.

[Yuli Ban]

Many potential applications of artificial intelligence involve making real-time decisions in physical systems while interacting with humans. Automobile racing represents an extreme example of these conditions; drivers must execute complex tactical manoeuvres to pass or block opponents while operating their vehicles at their traction limits1. Racing simulations, such as the PlayStation game Gran Turismo, faithfully reproduce the non-linear control challenges of real race cars while also encapsulating the complex multi-agent interactions. Here we describe how we trained agents for Gran Turismo that can compete with the world’s best e-sports drivers. We combine state-of-the-art, model-free, deep reinforcement learning algorithms with mixed-scenario training to learn an integrated control policy that combines exceptional speed with impressive tactics. In addition, we construct a reward function that enables the agent to be competitive while adhering to racing’s important, but under-specified, sportsmanship rules. We demonstrate the capabilities of our agent, Gran Turismo Sophy, by winning a head-to-head competition against four of the world’s best Gran Turismo drivers. By describing how we trained championship-level racers, we demonstrate the possibilities and challenges of using these techniques to control complex dynamical systems in domains where agents must respect imprecisely defined human norms.

[Yuli Ban]

Over the last two years, Sony AI trained a computer system to play Polyphony Digital's Gran Turismo Sport, a popular and realistic car racing game, and beat some of the world's best human competitors, Sony said on Wednesday.

The AI, named GT Sophy, defeated top humans only in time trials when there were no other cars on the track during a July competition. But by October, GT Sophy beat the humans even with a scrum of virtual race cars.

[caltrek]

Self-healing Materials for Robotics Made from ‘Jelly’ and Salt
February 18, 2022

https://www.eurekalert.org/news-releases/944019

Introduction:

(EurekAlert) Researchers have developed self-healing, biodegradable, 3D-printed materials that could be used in the development of realistic artificial hands and other soft robotics applications.

The low-cost jelly-like materials, developed by researchers at the University of Cambridge, can sense strain, temperature and humidity. And unlike earlier self-healing robots, they can also partially repair themselves at room temperature.

The results are reported in the journal NPG Asia Materials.

Soft sensing technologies could transform robotics, tactile interfaces and wearable devices, among other applications. However, most soft sensing technologies aren’t durable and consume high amounts of energy.

“Incorporating soft sensors into robotics allows us to get a lot more information from them, like how strain on our muscles allows our brains to get information about the state of our bodies,” said David Hardman from Cambridge’s Department of Engineering, the paper’s first author.