Newsletter #46 - AI addressing climate change

Welcome to Nural's newsletter where you will find a compilation of articles, news and cool companies, all focusing on how AI is being used to tackle global grand challenges.

Our aim is to make sure that you are always up to date with the most important developments in this fast-moving field.

Packed inside we have

  • AI backed by Al Gore to identify and monitor carbon emissions
  • Evolutionary development of intelligent agents
  • and latest developments in brain-computer interfaces

If you would like to support our continued work from £1 then click here!

Graham Lane & Marcel Hedman


Key Recent Developments


How satellites are finding the world’s hidden greenhouse gas emissions

How satellites are finding the world’s hidden greenhouse gas emissions
Climate Trace uses machine learning and satellite imagery to find the primary sources of the world’s greenhouse gasses
(scroll down the Quartz page to see the article)

What: Climate Trace is a coalition of 12 non-profits, academic institutions, and Al Gore.  It collects reliable, up-to-date information about the carbon emissions from specific sources, a task that is notoriously difficult and expensive. A wide range of approaches are used to identify potential sources of emissions. An AI model trained on satellite images for which the emissions are already known is then used to predict emissions from the source. The system is steadily becoming more granular. It can predict the emissions from a single plane or cargo ship, and has identified emissions from previously unknown landfill sites.

Key Takeaways: The system is vital because it provides data about carbon emissions in order to target concrete actions and to ensure compliance with commitments.


COP26 presentations about AI and climate change available for live streaming

Friday 5th November 12:00-12:30 GMT
How Breakthrough Space and AI Technologies Are Enabling Radical Climate Transparency – and Action - at Scale

Wednesday 10th November 16:00-17:00 GMT
A Clearer Picture: Towards Radical Transparency in Measurement, Reporting and Verification of Climate Action with Artificial Intelligence

For the UN webcast page: click here


New research combines reinforcement learning with evolutionary principles

New deep reinforcement learning technique helps AI to evolve
A new technique from Stanford researchers creates AI virtual agents that can evolve both in their physical structure and learning capacities.

What: In nature, body and brain evolve together. Researchers at Stanford simulated this process. Virtual agents were trained using reinforcement learning carrying out tasks in an environment that simulates the laws of physics. The best performers were selected as the next generation but only the physical characteristics were carried forward – not the AI training. The virtual agents were trained in environments of differing complexity.

Key Takeaways: The experiment generated a rich set of virtual agents with different physical characteristics. It validated the hypothesis that more complex environments give rise to more intelligent agents. Robot design typically starts with a defined form (e.g. biped or quadruped) but this research suggests that the optimal physical form could emerge during the AI training.

Paper: Embodied intelligence via learning and evolution


Brain implants could be the next computer mouse

Brain implants could be the next computer mouse
What the world’s fastest brain-typist is telling us about the future of computer interfaces.

What: BrainGate’s mission is to restore the independence of people who are paralyzed or affected by neurologic disease. They are running clinical trials inserting small brain-computer interfaces into paralyzed patients. One of the patients, Dennis DeGray, is currently the world’s fastest brain typist. Previously he achieved 8 words per minute imagining point-and-click on a keyboard. Now he has now achieved 18 words per minute using a new technique imagining hand-writing on lined paper.

Key Takeaways: A lot of venture capital money is going into consumer implants of brain-computer interfaces, giving rise to a range of concerns about medical ethics and fairness in society. In the meantime, the medical researchers at BrainGate are deploying real-world technology “to help desperate cases”.


AI Ethics

🚀 Summary of the NATO AI strategy

... but no mention of lethal autonomous weapons (“killer robots”) due to the widely divergent views of NATO members.

🚀 Microsoft acquires AI-powered moderation platform Two Hat

While Facebook has struggled with content moderation, Microsoft is implementing an AI-based platform in Xbox, Minecraft, and MSN

🚀 The perils of AI analytics for police body cameras

A new US system aiming to automatically identify misbehaviour by officers but has been greeted with skepticism.

Other interesting reads

🚀 Saving seaweed with machine learning

In addition to its nutritive value, seaweed helps fight climate change by absorbing excess carbon dioxide in the atmosphere

🚀 BenevolentAI’s drug discovery platform identifies novel target for ulcerative colitis

Machine learning identified a novel biological target for treatment of a disease without any prior references linking the gene to the disease.

🚀 Deep learning helps predict traffic crashes before they happen

Researcher develop a deep learning model that predicts very high-resolution crash risk maps.

🚀 Self-driving "Roboats" tested on Amsterdam's canals

Now ... self-driving watercraft


Cool companies found this week

End-to-end autonomous AI platform

Abacus.ai - has recently raised $50 million in funding and released Computer Vision as a Service

Improved neural network training with rigorous algorithmic methods

MosaicML - has recently come out of stealth with $37 million in seed funding

“… using AI to craft the next generation of AI ...”

Deci - AI model optimization startup Deci raises $21 million


And finally ...

Shadow Planet is a new album created by writer Robin Sloan, musician Jesse Solomon Clark, and Jukebox, a machine learning music program made by OpenAI in a process described as “like wandering in an enormous labyrinth.”

AI/ML must knows

Foundation Models - any model trained on broad data at scale that can be fine-tuned to a wide range of downstream tasks. Examples include BERT and GPT-3. (See also Transfer Learning)
Few shot learning - Supervised learning using only a small dataset to master the task.
Transfer Learning - Reusing parts or all of a model designed for one task on a new task with the aim of reducing training time and improving performance.
Generative adversarial network - Generative models that create new data instances that resemble your training data. They can be used to generate fake images.
Deep Learning - Deep learning is a form of machine learning based on artificial neural networks.

Best,

Marcel Hedman
Nural Research Founder
www.nural.cc

If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.

If you are enjoying this content and would like to support the work financially then you can amend your plan here from £1/month!