For predicting carbon capture
Measuring how much forests are capturing carbon in the atmosphere is usually a labor-intensive, manual process—making it difficult to verify the carbon offsets market. But Pachama is using machine learning to analyze satellite and lidar imagery of forests to predict the amount of carbon a forest is capturing, with up to 90% accuracy compared to traditional models. The company has also built a model to track annual trends over time in the Amazon rainforest in particular. By accurately representing the impact of reforesting and carbon capture projects, Pachama has been able to bring clarity to which initiatives are making a real difference in the fight against climate change—leading to partnerships with tech companies like Microsoft, Shopify, and Amazon. For more on why Pachama is one of the Most Innovative Companies of 2021, click here.
For fighting misinformation
Logically is using AI to scale up the fight against one of the biggest problems plaguing the internet: misinformation. The company’s algorithms cross-reference information related to a claim, feeding evidence, reference materials, and analysis from more than 100,000 sources to human fact-checkers. In August 2020, Logically launched a browser extension and app in the United States that enable users to quickly fact-check misinformation as it arises on social media. The company has also worked with governments in India and the U.K., where it is based, to identify COVID-19 misinformation threats. Logically’s internal investigations team was able to identify a previously anonymous operative within the QAnon conspiracy that led to the subsequent shutdown of a popular website within the conspiracy theory community.
For pushing AI to the edge
In 2020, renowned semiconductor maker Arm launched two new chips that are designed to bring AI beyond smartphones and tablets. The chips, called the Arm Cortex-M55 and the Ethos-U55, can power machine learning on billions more devices and sensors within the internet of things. This is part of Arm’s push into what’s known as “TinyML,” where computationally intensive software can run entirely on small, super low-powered devices without needing access to the internet or the cloud. This kind of AI is already showing up in small wearables like the Oura ring and in smart asthma inhalers, but it has other applications in farming, biometrics, and the smart home. Now, pending regulatory approval, these capabilities will be part of the Nvidia family: In September 2020, the U.S. chipmaker agreed to acquire Arm for $40 billion.
4. Cerebras Systems
For building a dedicated AI supercomputer
Cerebras’s CS-1 computer was designed specifically for artificial intelligence. It includes a trillion-transistor processor that’s as large as a dinner plate called the Wafer Scale Engine, which reduces the amount of time it takes to train and run AI models from months to minutes. Cerebras says its purpose-built system, which also includes a software platform that integrates with the company’s hardware, can accelerate model training by 1,000 times. The Wafer Scale Engine has resulted in 37 patents with dozens more pending, and Cerebras’s system is being used by researchers to study COVID-19 therapeutics, black holes, and nuclear fusion by such institutions as the U.S. Department of Energy’s Argonne National Laboratory. Cerebras is now also partnering with the pharma giant GSK to speed up drug discovery.
5. Algorithmic Justice League
For crusading against programmatic bias
Even as research and deployment of AI systems have continued to accelerate, the nonprofit Algorithmic Justice League is dedicated to auditing these systems to make sure they are free from racial, gender, and other kinds of bias. Based on founder Joy Buolamwini’s pioneering research that exposed how facial-recognition algorithms from top tech companies are significantly better at detecting light-skinned male faces as compared to dark-skinned female faces, AJL’s advocacy has helped convince giants like Amazon, IBM, and Microsoft to hold back on developing facial-recognition algorithms. A documentary about the organization’s work that examines how people can be harmed by technology, Coded Bias, premiered at Sundance in early 2020.
For being a smart shopper
When the online grocery rush began during the pandemic, Walmart built an AI-based feature called “Customer Choice” that uses historical and real-time shopping data to predict when items might be out of stock. If a customer who is shopping online chooses an item that may need a substitution, Walmart now shows them up to five options so that people can choose which alternative they prefer. Those choices are then stored for future shopping orders. The company built a working prototype of the feature within two weeks. The quick deployment helped Walmart nearly double its e-commerce sales year over year in Q2 2020, during the heart of the pandemic.
For mimicking human speech
Introduced by AI laboratory OpenAI in May 2020, the natural language model GPT-3 has quickly become a revolutionary technology for its ability to mimic human speech. After being trained on hundreds of billions of words from sources like books and Wikipedia, the model—which at the time of its launch was the largest language model in the world with 175 billion parameters—has led to an explosion of creative projects and startups built on top of the technology. Microsoft has exclusively licensed the source code, but GPT-3 is now available through an API, allowing developers without any machine-learning expertise to play with its powerful language-prediction capabilities. Despite the AI’s impressive abilities, it has also sparked concerns about bias and raised questions about whether AI can ever truly comprehend meaning.
For opening the black box
AI models are often black boxes—making it difficult for people to understand and trust their decision-making ability in the real world. But Truera is aiming to make AI truly explainable and trustworthy by preventing errors and bias from creeping into machine-learning models when they are deployed and as data changes over time. In 2020, the company deployed its explainable AI platform at Fortune 100 companies in insurance and banking, including Standard Chartered, where it aimed to ensure the bank’s credit algorithms were built fairly and equitably.
For bringing computing intelligence to the camera viewfinder
In June 2020, Adobe launched Photoshop Camera, an app that brings the company’s expertise in computationally enhanced photography to the viewfinder. The app is built on top of Adobe Sensei, the company’s machine-learning platform, and automatically adjusts a photo when you take it to bring out its best qualities. In addition, the app enables you to layer hundreds of filters and lenses on top of your images in real time—saving you the trouble of painstakingly pushing pixels around in Photoshop. Since launch, the app has been downloaded nearly 3 million times, and users have downloaded more than 3.5 million lenses.
For optimizing data storage
Building and deploying AI models and other types of high-performance computing requires an unbelievable amount of data. But storing and managing that data can be a nightmare for IT teams. WekaIO’s product is an efficient data-storage system that is optimized for machine learning, allowing companies to get the most out of GPU servers. The Weka File System enables IT teams to manage millions of files in the same directory, ensuring all of the data is available at the same time for training AI models. With clients in autonomous vehicles, life sciences, and banking, WekaIO has built a system that can handle the intensity of modern data science in corporations.