OpenAI Disbands Its Robotics Research Team

Posted by Peter Rudin on 23. July 2021 in News No Comments

OpenAI has disbanded its robotics team after years of research into machines that can learn to perform tasks like solving a Rubik’s Cube, shifting its focus to other domains, where data is more readily available.

“Because of the rapid progress in AI and its capabilities, we’ve found that other approaches, such as reinforcement learning with human feedback, lead to faster progress in our reinforcement learning research”, OpenAI cofounder Zaremba said.

Moving away from robotics might also reflect the economic realities the company faces. DeepMind, the Alphabet-owned AI research lab, has undergone a similar shift due to rising R&D cost in favour of work with applications, like protein shape prediction.

Read More

Robots May Soon Be Able To Reproduce

Posted by Peter Rudin on 9. July 2021 in News No Comments

A team of researchers have recently demonstrated a fully automated technology to allow physical robots to repeatedly breed, evolving their artificial genetic code over time to better adapt to their environment.

The idea of digital evolution – imitating biological evolution in software to successively breed better and better solutions to a problem over time – is not new.

While scientists have always been confident that digital evolution could be effective as an optimisation tool, its creativity in producing original and unusual design has been more surprising. Unlike natural evolution which is driven simply by the goals of “survival and reproduction”, artificial evolution can be driven by specific targets.

Read More

Will Reinforcement Learning Achieve General AI ?

Posted by Peter Rudin on 25. June 2021 in News No Comments

In a new paper, scientists at Google’s DeepMind AI research unit argue that intelligence and its associated abilities will emerge not from formulating and solving complicated problems but by sticking to a simple but powerful principle: reward maximization.

Reinforcement learning formalises the problem of goal-seeking intelligence. The general problem may be instantiated with a wide and realistic range of goals and worlds, corresponding to different reward signals to maximise in different environments.

The researchers also acknowledge that learning mechanisms for reward maximization is an unsolved problem that remains a central question to be further studied in reinforcement learning.

Read More

Will AI Succeed? Quantum Theory Suggests Otherwise

Posted by Peter Rudin on 11. June 2021 in News No Comments

Will artificial intelligence one day surpass human thinking? The rapid progress of AI has raised concerns that its abilities will grow uncontrollably, eventually wiping out humanity.

One argument against the possibility reaching this level of intelligence comes from the fact that AI would need to accurately predict the future.

Quantum Theory, one of modern science’s key concepts for explaining the universe, says that predicting the future may not be possible because the universe is random. The next Singularity2030 Essay ‘Quantum Theory to Overcome the limits of AI’, to be published on June 18, 2021, augments this discussion. 

Read More

Google with a New Species of Search Engine

Posted by Peter Rudin on 28. May 2021 in News No Comments

Search results have improved leaps and bounds over the years. Still, the approach is far from perfect.

Though they have their own shortcomings large language models like GPT-3 are much more flexible and can construct novel replies in natural language to any query or prompt.

A Google team suggests that the next generation of search engines might synthesize the best of all worlds, folding today’s top information retrieval systems into large-scale AI.

A key advance would be moving beyond algorithms that only model the relationships between terms (such as individual words) to algorithms that also model the relationship between words in an article.

Read More

The Brain can teach AI a Lesson or Two

Posted by Peter Rudin on 14. May 2021 in News No Comments

The brain is considered an autonomous learning system. It can detect patterns and acquire new knowledge without external guidance. Until recently, this was not the case for AI — any data being fed to machine learning systems needed to be tagged first.

Some presumed that no such mechanism existed in biological neural networks. However, based on a combination of considerations from the current practice of A and computational neuroscience researchers came to a different conclusion.

In a recent scientific publication, the authors show that self-supervision and error backpropagation co-exists in the brain, and with a very specific area of the brain involved: the hippocampus.

Read More

Latest Survey: Only 6% of Companies have Adopted AI

Posted by Peter Rudin on 30. April 2021 in News No Comments

In a new survey Juniper Networks found that 95% of respondents believe their organization would benefit from embedding AI into their daily operations. However, only 6% of those reported adoption of AI-powered solutions across their business.

It  is Jupiter’s second recent report to peg data issues as the reason organizations fail to successfully deploy AI. Inherent biases in the data being used create compliance risks for their organizations.

73%  of Juniper survey respondents said that their organizations were struggling with expanding their workforce to integrate with AI-systems. Executives reported that they feel it is more of a priority to hire people than to develop their own AI-Talents.

Read More

Issues and Risks to Society with large Language Models

Posted by Peter Rudin on 16. April 2021 in News No Comments

Researchers at Google’s DeepMind have discovered major flaws in the output of large language models like GPT-3 and warn these could have serious consequences for society, like enabling deception and reinforcing bias.

The DeepMind paper joins a series of works that highlight NLP shortcomings. In late March, nearly 30 businesses and universities from around the world found major issues in an audit of five popular multilingual datasets used for machine translation.

A paper written about that audit found that in a significant fraction of the major dataset portions evaluated, less than 50% of the sentences were of acceptable quality, according to more than 50 volunteers from the NLP community. 

Read More

Researcher Finds a Better Way to Explore the Brain

Posted by Peter Rudin on 2. April 2021 in News No Comments

Using a new class of nanoparticles, Sakhrat Khizroev, a professor of computer engineering at the University of Miami, hopes to unlock the secrets of the brain.

Using a novel class of ultrafine units called magnetoelectric nanoparticles (MENPs), he and his research group are perfecting a method to talk to the brain without wires or implants as applied by Neuralink, Elon Musk’s Neurotechnology company.

Once the MENPs are inside the brain and positioned next to neurons, they can stimulate them with an external magnetic field, which in turn produces an electric field for communication and control without the use of wires. To extract the information in real time, his team plans to use a special helmet with magnetic transducers to send and pick up signals.

Read More

This Digital NFT Painting Just Sold for 69 Million USD

Posted by Peter Rudin on 19. March 2021 in News No Comments

A work by digital artist Mike Winkelmann, better known as “Beeple,” has sold for an astonishing $69 million at world-renowned auction house Christie’s.
The work, titled “The First 5000 Days” was, according to Christie’s, the “first purely digital work of art ever offered by a major auction house.”

NFT stands for “non-fungible token,” and it can technically contain anything digital, including drawings, animated GIFs, songs etc. NFTs allow you to buy and sell ownership of unique digital items and keep track of who owns them using the blockchain.

Critics of the rise of NFTs question the trend and see it as a monetizing hype surrounding digital art. As NFTs allow artists to sell unique pieces of art, digital works can also be sold off as authentic even if they are not.

Read More