Lifeboat Foundation

Eureka has also taught quadruped, dexterous hands, cobot arms and other robots to open drawers, use scissors, catch balls and nearly 30 different tasks. According to NVIDIA Research, the AI agent’s trial and error-based reward programs are 80 percent more effective than those written by human experts. This shift meant the robots’ performance also improved by over 50 percent. Eureka also self-evaluates based on training results, instructing changes in reward functions as it sees fit.

NVIDIA Research has published a library of its Eureka algorithms, encouraging others to try them out on NVIDIA Isaac Gym, the organization’s “physics simulation reference application for reinforcement learning research.”

The idea of robots teaching robots is seeing increased interest and success. A May 2023 paper published in the Transactions on Machine Learning Research journal presented a new system called SKILL (Shared Knowledge Lifelong Learning), which allowed AI systems to learn 102 different skills, including diagnosing diseases from chest X-rays and identifying species of flowers. The AIs shared their knowledge — acting as teachers in a way — with each other over a communication network and were able to master each of the 102 skills. Researchers at schools like MIT and the University of Bristol have also had success, specifically in using AI to teach robots how to manipulate objects.

Whenever light interacts with matter, light appears to slow down. This is not a new observation and standard wave mechanics can describe most of these daily phenomena.

For example, when light is incident on an interface, the standard wave equation is satisfied on both sides. To analytically solve such a problem, one would first find what the wave looks like at either side of the interface, and then employ electromagnetic boundary conditions to link the two sides together. This is called a piecewise continuous solution.

However, at the boundary, the incident light must experience an acceleration. So far, this has not been accounted for.

Two types of technologies could change the privacy afforded in encrypted messages, and changes to this space could impact all of us.

On October 9, I moderated a panel on encryption, privacy policy, and human rights at the United Nations’s annual Internet Governance Forum. I shared the stage with some fabulous panelists including Roger Dingledine, the director of the Tor Project; Sharon Polsky, the president of the Privacy and Access Council of Canada; and Rand Hammoud, a campaigner at Access Now, a human rights advocacy organization. All strongly believe in and champion the protection of encryption.

I want to tell you about one thing that came up in our conversation: efforts to, in some way, monitor encrypted messages.

Policy proposals have been popping up around the world (like in Australia, India, and, most recently, the UK) that call for tech companies to build in ways to gain information about encrypted messages, including through back-door access. There have also been efforts to increase moderation and safety on encrypted messaging apps, like Signal and Telegram, to try to prevent the spread of abusive content, like child sexual abuse material, criminal networking, and drug trafficking.

Continue reading “IGF 2023 WS #356 Encryption’s Critical Role in Safeguarding Human Rights” »

As Shor looked for applications for his quantum period-finding algorithm, he rediscovered a previously known but obscure mathematical theorem: For every number, there exists a periodic function whose periods are related to the number’s prime factors. So if there’s a number you want to factor, you can compute the corresponding function and then solve the problem using period finding — “exactly what quantum computers are so good at,” Regev said.

On a classical computer, this would be an agonizingly slow way to factor a large number — slower even than trying every possible factor. But Shor’s method speeds up the process exponentially, making period finding an ideal way to construct a fast quantum factoring algorithm.

Shor’s algorithm was one of a few key early results that transformed quantum computing from an obscure subfield of theoretical computer science to the juggernaut it is today. But putting the algorithm into practice is a daunting task, because quantum computers are notoriously susceptible to errors: In addition to the qubits required to perform their computations, they need many others doing extra work to keep them from failing. A recent paper by Ekerå and the Google researcher Craig Gidney estimates that using Shor’s algorithm to factor a security-standard 2,048-bit number (about 600 digits long) would require a quantum computer with 20 million qubits. Today’s state-of-the-art machines have at most a few hundred.

Jailbroken large language models (LLMs) and generative AI chatbots — the kind any hacker can access on the open Web — are capable of providing in-depth, accurate instructions for carrying out large-scale acts of destruction, including bio-weapons attacks.

An alarming new study from RAND, the US nonprofit think tank, offers a canary in the coal mine for how bad actors might weaponize this technology in the (possibly near) future.

In an experiment, experts asked an uncensored LLM to plot out theoretical biological weapons attacks against large populations. The AI algorithm was detailed in its response and more than forthcoming in its advice on how to cause the most damage possible, and acquire relevant chemicals without raising suspicion.

The findings support a diagnostic algorithm to identify the cancer subtype and guide specialized treatment.

The new book Minding the Brain from Discovery Institute Press is an anthology of 25 renowned philosophers, scientists, and mathematicians who seek to address that question. Materialism shouldn’t be the only option for how we think about ourselves or the universe at large. Contributor Angus Menuge, a philosopher from Concordia University Wisconsin, writes.

Neuroscience in particular has implicitly dualist commitments, because the correlation of brain states with mental states would be a waste of time if we did not have independent evidence that these mental states existed. It would make no sense, for example, to investigate the neural correlates of pain if we did not have independent evidence of the existence of pain from the subjective experience of what it is like to be in pain. This evidence, though, is not scientific evidence: it depends on introspection (the self becomes aware of its own thoughts and experiences), which again assumes the existence of mental subjects. Further, Richard Swinburne has argued that scientific attempts to show that mental states are epiphenomenal are self-refuting, since they require that mental states reliably cause our reports of being in those states. The idea, therefore, that science has somehow shown the irrelevance of the mind to explaining behavior is seriously confused.

The AI optimists can’t get away from the problem of consciousness. Nor can they ignore the unique capacity of human beings to reflect back on themselves and ask questions that are peripheral to their survival needs. Functions like that can’t be defined algorithmically or by a materialistic conception of the human person. To counter the idea that computers can be conscious, we must cultivate an understanding of what it means to be human. Then maybe all the technology humans create will find a more modest, realistic place in our lives.

At HLTH, Mayo Clinic Platform President John Halamka gave a window into how his health system is mitigating generative AI risks. Some of the measures Mayo is taking include running analyses on how well algorithms perform across various subgroups and training models only on internal de-identified data.

Computer vision algorithms have become increasingly advanced over the past decades, enabling the development of sophisticated technologies to monitor specific environments, detect objects of interest in video footage and uncover suspicious activities in CCTV recordings. Some of these algorithms are specifically designed to detect and isolate moving objects or people of interest in a video, a task known as moving target segmentation.

While some conventional algorithms for moving target segmentation attained promising results, most of them perform poorly in real-time (i.e., when analyzing videos that are not pre-recorded but are being captured in the present moment). Some research teams have thus been trying to tackle this problem using alternative types of algorithms, such as so-called quantum algorithms.

Researchers at Nanjing University of Information Science and Technology and Southeast University in China recently developed a new quantum algorithm for the segmentation of moving targets in grayscale videos. This algorithm, published in Advanced Quantum Technologies, was found to outperform classical approaches in tasks that involve the analysis of video footage in real-time.