Lifeboat Foundation

Learnings For Regenerative Morphogenesis, Astro-Biology And The Evolution Of Minds — Dr. Michael Levin, Tufts University, and Dr. Josh Bongard, University of Vermont.

Xenobots are living micro-robots, built from cells, designed and programmed by a computer (an evolutionary algorithm) and have been demonstrated to date in the laboratory to move towards a target, pick up a payload, heal themselves after being cut, and reproduce via a process called kinematic self-replication.

Continue reading “Xenobots — Novel Synthetic Life Forms At The Intersection Of Biology & Information Science” »

Though Meta didn’t give numbers on RSC’s current top speed, in terms of raw processing power it appears comparable to the Perlmutter supercomputer, ranked fifth fastest in the world. At the moment, RSC runs on 6,800 NVIDIA A100 graphics processing units (GPUs), a specialized chip once limited to gaming but now used more widely, especially in AI. Already, the machine is processing computer vision workflows 20 times faster and large language models (like, GPT-3) 3 times faster. The more quickly a company can train models, the more it can complete and further improve in any given year.

In addition to pure speed, RSC will give Meta the ability to train algorithms on its massive hoard of user data. In a blog post, the company said that they previously trained AI on public, open-source datasets, but RSC will use real-world, user-generated data from Meta’s production servers. This detail may make more than a few people blanch, given the numerous privacy and security controversies Meta has faced in recent years. In the post, the company took pains to note the data will be carefully anonymized and encrypted end-to-end. And, they said, RSC won’t have any direct connection to the larger internet.

To accommodate Meta’s enormous training data sets and further increase training speed, the installation will grow to include 16,000 GPUs and an exabyte of storage—equivalent to 36,000 years of high-quality video—later this year. Once complete, Meta says RSC will serve training data at 16 terabytes per second and operate at a top speed of 5 exaflops.

The RVFL was used in combination with four different techniques: the Jellyfish Search Algorithm (JFSA); the Artificial Ecosystem-based Optimization (AEO); the Manta Ray Foraging Optimization (MRFO) model; and the Sine Cosine Algorithm (SCA). Through the four models, the academics assessed the PV-fed current, the cooling power, the average air chamber temperature, and the coefficient of performance (COP) of a PV-powered STEACS for air conditioning of a 1m³ test chamber under diversified cooling loads varying from 65 to 260W.

The system was built with six solar panels, an air duct system, four batteries, a charge controller, TECs, an inverter, heat sinks, a test chamber, and condenser fans. “The TECs were mainly connected with the air duct arrangement and placed close to each other [and] were placed between the air duct and heat sinks,” the researchers explained. “When direct PV current was fed to TECs arranged on the sheet of the air duct system, one face [became] cold, defined as a cold air duct, and another side [became] hot, called “hot air.” The air ducts were composed of an acrylic enclosure wrapped with a protection sheet.”

It could hardly be more complicated: tiny particles whir around wildly with extremely high energy, countless interactions occur in the tangled mess of quantum particles, and this results in a state of matter known as “quark-gluon plasma”. Immediately after the Big Bang, the entire universe was in this state; today it is produced by high-energy atomic nucleus collisions, for example at CERN.

Such processes can only be studied using high-performance computers and highly complex computer simulations whose results are difficult to evaluate. Therefore, using artificial intelligence or machine learning for this purpose seems like an obvious idea. Ordinary machine-learning algorithms, however, are not suitable for this task. The mathematical properties of particle physics require a very special structure of neural networks. At TU Wien (Vienna), it has now been shown how neural networks can be successfully used for these challenging tasks in particle physics.

There are few environments as unforgiving as the ocean. Its unpredictable weather patterns and limitations in terms of communications have left large swaths of the ocean unexplored and shrouded in mystery.

“The ocean is a fascinating environment with a number of current challenges like microplastics, algae blooms, coral bleaching, and rising temperatures,” says Wim van Rees, the ABS Career Development Professor at MIT. “At the same time, the ocean holds countless opportunities — from aquaculture to energy harvesting and exploring the many ocean creatures we haven’t discovered yet.”

Ocean engineers and mechanical engineers, like van Rees, are using advances in scientific computing to address the ocean’s many challenges, and seize its opportunities. These researchers are developing technologies to better understand our oceans, and how both organisms and human-made vehicles can move within them, from the micro scale to the macro scale.

The more a technology or concept permeates and gets normalized in our day-to-day lives, the more we grow to expect from it. About two decades ago, a sub-56kpbs dial-up internet connection seemed miraculous. Today, with internet speeds as high as 2000Mbps becoming normal, the 56Kbps connection would be considered a failure of sorts—in the developed world, at least. This shift in expectation also applies to AI. Having seen numerous practical AI applications aid human convenience and progress, both the general population and the AI research community now expects every new breakthrough in the field to be more earth-shattering than the previous one. Similarly, what qualifies as AI failure has also seen a massive shift in recent years, especially from a problem owner’s perspective. failure, in most cases, is attributed to technology-centric factors like the quality of data or the capabilities of algorithms and hardware used, ignoring the most crucial aspect of AI success—the end user.

One of the questions that I always get when I talk about credit risk modeling (Loan payment default, credit card payment default) is about the algorithms’ or models’ prediction limitations.

How can we implement a solution if the prediction probability is lower? How can we use the model or algorithm effectively for real-world problems?

Have chalked out what are all the available methods to predict the probability of default, while not getting into them detail since that’s not what this article’s intent is.

The single technique for teaching neural networks multiple skills is a step towards general-purpose AI.

“No AI technology ‘where training or transactional data is known to be of poor quality, carry bias, or where the quality of such data is unknown’ should ever be considered for use, and thus should be deemed Extreme Risk, not High Risk. Any AI technology based on poor quality or biased data is inherently compromised.”

“No AI technology that assists in “identifying, categorizing, prioritizing or otherwise making decisions pertaining to members of the public” should be deemed Low Risk. Automating such actions through technology, even with the inclusion of a human-in-the-loop, is an intrinsically risky activity, and should be categorized as such by the Policy.”

Full Story:

Continue reading “Community Of AI Researchers, Practitioners Calls For Stringency In Toronto Police Services Board’s Use Of AI Technologies Policy” »