The Chan Zuckerberg Initiative has bought an AI-powered tool for analysing and searching scientific papers and will be making it free to use for all researchers.
The acquisition of Meta is the first made by the charitable foundation, setup by Facebook founder Mark Zuckerberg and his wife Priscilla Chan in December 2015 to help "cure, prevent or manage all diseases by the end of the century".
The pair announced in September 2016 it would invest $3 billion (£2.3 billion, at the time) of their personal fortune into the initiative, to be spent on organisations and companies conducting medical research over the next ten years, ultimately passing on 99 per cent of their own money earned through Facebook. An initial $600 million (£462 million) was earmarked for a Biohub in San Francisco, which is now focusing on Cell Atlas, a project to map the cells controlling the body’s organs, and the Infectious Disease initiative, which will focus on the development of new tests and vaccines for diseases including HIV, Ebola and Zika.
Scientific search engine Meta is the first company to be bought outright by the Initiative. Anyone who wants to use the tool is being invited to “reserve a free account” while shareholders and courts approve the acquisition.
CEO of the Toronto-based company, Sam Molyneux, explains in a blogpost how he and his sister Amy, along with a team of engineers and scientists, spent six years working out “how to use artificial intelligence to analyse new scientific knowledge as it’s published - along with the majority of what has been written, throughout modern history.”
Meta’s AI analyses millions of papers, seeking out patterns and drawing insights faster than a human researcher. It helps scientists make connections in the data, find collaborators working on similar material, and helps funding bodies find investment opportunities. It was created to help scientists find relevant information faster - ranking and prioritising the data it has scoured and sourced, working like an expert Google but with a human-like eye for detail and importance - but also to make the rapidly increasing number of scientific papers published each year manageable and useful. According to Molyneux, every day more than 4,000 papers on biomedicine alone are published.
“Using current tools, most will not be read by other scientists who can learn from them. Scientists lack the means to make sense of the vast amount of research being produced around the world. To speed up progress, researchers need to be able to learn from each other's insights in real time.” Meta is already used by 1,200 institutes globally, says Molyneux, and he believes the AI will be able to make its own scientific discoveries as time progresses.
“In partnership with SRI International (the creators of Siri), we commercialised an AI technology that can read millions of papers to uncover emerging discoveries years ahead of time. We created neural network systems that look at hundreds of signals within new papers, as they are published, to project their future impact with striking accuracy.”
Under the Chan-Zuckerberg Initiative, Meta will become a “single, powerful tool that is available to everyone”.
The first step, post-acquisition, will be for Meta to open up to developers so it can be integrated on other platforms, prioritising the sectors and people “who need them most”.
“We will be working to make Meta even more powerful and useful for the entire scientific community, and are committed to offering these tools and features for free to all researchers,” Cori Bargmann, the Chan Zuckerberg Initiative’s president of science, and CTO Brian Pinkerton said in a Facebook post.
“Meta will help scientists learn from others’ discoveries in real time, find key papers that may have gone unnoticed, or even predict where their field is headed.
“The potential for this kind of platform is virtually limitless: a researcher could use Meta to help identify emerging techniques for understanding coronary artery disease; a graduate student could see that two different diseases activate the same immune defence pathway; and clinicians could find scientists working on the most promising Zika treatments sooner. In the long run, it could be extended to other areas of knowledge: for example, it could help educators stay up to date on developmental science to better understand how children learn.”
Back in 2015, Facebook made the designs for the computer server it built to run AI algorithms open source. But it is not the first Silicon Valley tech giant to promise to open up its AI tech to the world in some shape or form.
The Echo already has tie-ins with BMW, Fitbit, Jamie Oliver and others, but Amazon has ensured the AI powerhouse driving it, Alexa, will have a growing ecosystem by providing developers with the Alexa Skills Kit for free, along with Amazon Voice Service. Custom experiences can be created, and the personal speaker assistant can be built into anything. This increases the opportunities for Alexa to learn and improve faster. Amazon has even launched a $100m Alexa fund to help startups and designers with these integrations.
While the inner workings of some of Google’s most advanced AI experiments, including work done by UK acquisition DeepMind, remain secret, it has made elements of its artificial intelligence open source. In November 2015 it made TensorFlow, the software library it compiled while working on deep neural network research within the Google Brain Team, open to all.
It is the basis of the engine behind products such as Google's Photos app, search and voice recognition. Like Amazon, Google is largely doing this to speed up progress and uptake in the field, not simply to be magnanimous. Despite the emergence of connected hairbrushes and toothbrushes, and other useless inventions that need never have been dreamt up, the internet of things has not yet come close to living up to its long-slated expectations. “What we’re hoping is that the community adopts this as a good way of expressing machine learning algorithms of lots of different types, and also contributes to building and improving [TensorFlow] in lots of different and interesting ways,” Google engineer Jeff Dean told WIRED US in 2015. Within a year of going open source, 480 people had contributed to TensorFlow. Most recently, Google announced it would be working with Raspberry Pi to embed its AI in the credit card-sized computers.
Not wanting to be left out, Microsoft joined the open source efforts of its peers and made the AI powering Cortana available just one month after TensorFlow was released on Github (it had been available earlier, but only for non-commercial purposes). The voice recognition framework, CNTK, is a deep learning tool, supports Microsoft Windows, and can be used across many servers. “We want this to be useful not just for academics but for commercial artificial intelligence and deep learning companies,” said Microsoft’s chief scientist Xuedong Huang.
This article was originally published by WIRED UK
