Lifeboat Foundation

In the field of artificial intelligence, reinforcement learning is a type of machine-learning strategy that rewards desirable behaviors while penalizing those which aren’t. An agent can perceive its surroundings and act accordingly through trial and error in general with this form or presence – it’s kind of like getting feedback on what works for you. However, learning rules from scratch in contexts with complex exploration problems is a big challenge in RL. Because the agent does not receive any intermediate incentives, it cannot determine how close it is to complete the goal. As a result, exploring the space at random becomes necessary until the door opens. Given the length of the task and the level of precision required, this is highly unlikely.

Exploring the state space randomly with preliminary information should be avoided while performing this activity. This prior knowledge aids the agent in determining which states of the environment are desirable and should be investigated further. Offline data collected by human demonstrations, programmed policies, or other RL agents could be used to train a policy and then initiate a new RL policy. This would include copying the pre-trained policy’s neural network to the new RL policy in the scenario where we utilize neural networks to describe the procedures. This process transforms the new RL policy into a pre-trained one. However, as seen below, naively initializing a new RL policy like this frequently fails, especially for value-based RL approaches.

Google AI researchers have developed a meta-algorithm to leverage pre-existing policy to initialize any RL algorithm. The researchers utilize two procedures to learn tasks in Jump-Start Reinforcement Learning (JSRL): a guide policy and an exploration policy. The exploration policy is an RL policy trained online using the agent’s new experiences in the environment. In contrast, the guide policy is any pre-existing policy that is not modified during online training. JSRL produces a learning curriculum by incorporating the guide policy, followed by the self-improving exploration policy, yielding results comparable to or better than competitive IL+RL approaches.

CSIS will host a public event on responsible AI in a global context, featuring a moderated discussion with Julie Sweet, Chair and CEO of Accenture, and Brad Smith, President and Vice Chair of the Microsoft Corporation, on the business perspective, followed by a conversation among a panel of experts on the best way forward for AI regulation. Dr. John J. Hamre, President and CEO of CSIS, will provide welcoming remarks.

Keynote Speakers:
Brad Smith, President and Vice Chair, Microsoft Corporation.
Julie Sweet, Chair and Chief Executive Officer, Accenture.

Continue reading “Responsible AI in a Global Context” »

Rich Heimann explores how the philosophy of mind and consciousness has affected AI research.

There are ways around this, but they don’t have the exciting scalability story and worse, they have to rely on a rather non-tech crutch: human input. Smaller language models fine-tuned with actual human-written answers are ultimately better at generating less biased text than a much larger, more powerful system.

And further complicating matters is that models like OpenAI’s GPT-3 don’t always generate text that’s particularly useful because they’re trained to basically “autocomplete” sentences based on a huge trove of text scraped from the internet. They have no knowledge of what a user is asking it to do and what responses they are looking for. “In other words, these models aren’t aligned with their users,” OpenAI said.

Any test of this idea would be to see what happens with pared-down models and a little human input to keep those trimmed neural networks more…humane. This is exactly what OpenAI did with GPT-3 recently when it contracted 40 human contractors to help steer the model’s behavior.

Our next challenge, then, was to determine the evolutionary connections between these genes. The more similar the two genes were, the more likely viruses with those genes were closely related. Because these sequences had evolved so long ago (possibly predating the first cell), the genetic signposts indicating where new viruses may have split off from a common ancestor had been lost to time. A form of artificial intelligence called machine learning, however, allowed us to systematically organize these sequences and detect differences more objectively than if the task were done manually.

We identified a total of 5,504 new marine RNA viruses and doubled the number of known RNA virus phyla from five to 10. Mapping these new sequences geographically revealed that two of the new phyla were particularly abundant across vast oceanic regions, with regional preferences in either temperate and tropical waters (the Taraviricota, named after the Tara Oceans expeditions) or the Arctic Ocean (the Arctiviricota).

MIT has launched a new academia and industry partnership called the AI Hardware Program that aims to boost research and development.

“A sharp focus on AI hardware manufacturing, research, and design is critical to meet the demands of the world’s evolving devices, architectures, and systems,” says Anantha Chandrakasan, dean of the MIT School of Engineering, and Vannevar Bush Professor of Electrical Engineering and Computer Science.

Microsoft and Hewlett Packard Enterprise (HSE) are working with NASA scientists to develop an AI system for inspecting astronauts’ gloves.

Space is an unforgiving environment and equipment failures can be catastrophic. Gloves are particularly prone to wear and tear as they’re used for just about everything, including repairing equipment and installing new equipment.

Continue reading “AI system inspects astronauts’ gloves for damage in real-time” »

University of Alberta researchers have trained a machine learning model to identify people with post-traumatic stress disorder with 80 per cent accuracy by analyzing text data.

University of Alberta researchers have trained a machine learning model to identify people with post-traumatic stress disorder with 80 per cent accuracy by analyzing text data.