Comprehensive software reviews to make better IT decisions
Why Did Facebook Acquire Another AI Startup (Atlas ML)?
“Facebook quietly acquired another UK AI startup and almost no one noticed.” reported TechCrunch on February 10. The startup in question is Atlas ML (Deeptide Inc.), which offers rather esoteric products aimed at machine learning (ML) researchers and developers. Other recent acquisitions include Bloomsbury AI (natural language processing [NLP]) in 2018 and Scape Technologies (computer vision for augmented reality [AR]) just this month.
While the previous acquisitions made sense – NLP to combat fake news and AR to build out immersive AR experiences – this analyst was puzzled by the most recent acquisition. What does Facebook want with Atlas ML? I decided to investigate.
PWC organizes most recent ML research by linking research papers published across a range of journals and open access repositories such as arXiv with the source code on which the research is based (and that usually resides elsewhere). So if you read a research paper and want to get your hands on the source code, PWC saves a lot of time. Click on the button and it will take you there; no need to search.
PWC also ranks research papers, and anyone can contribute. It is open source, and it runs on GitHub, a hosting platform for software development commonly used to host open source projects. (It was acquired by Microsoft in 2018.)
The idea behind PWC, writes cofounder Robert Stojnic, was “helping the community track newly published machine learning papers with source code and quickly understand the current state of the art.”
In addition to linking research papers and source code, PWC also extracts metrics on the code itself. And here’s where sotabench comes in. It is a twin site to PWC in the sense that it continuously executes code libraries submitted and evaluates them against common benchmarks used in ML. Think of it as a “free benchmarking service for all open source ML repositories that is integrated with GitHub. It is like Continuous Integration, but instead of running unit tests, it runs benchmarks.”
So if you liked someone else’s ML research, you can obtain their source code via PWC, modify it to your heart’s content, hopefully improving on it, and then test it using sotabench. This saves a lot of time and speeds up development in a highly competitive field where improvements are measured in fractions of a percentage, not unlike milliseconds in Olympic sports.
And it’s not only for academia. Industry researchers benefit for the same reasons: they can shorten development cycles and put out improved ML models faster. (And hence more accurate decisions.)
But going back to our initial question – how does Facebook stand to benefit from this acquisition? By the way, it only paid US$40 million for Atlas ML. For comparison, Microsoft paid $7.5 billion for GitHub. Google acquired Kaggle, an online community of data scientists and ML practitioners known for its competitions, in 2017 for an undisclosed amount, but in 2011 (the most recent amount available), Kaggle was worth $25 million.
We think the answer is three-part:
- Automation of source code testing/benchmarking will help Facebook to shorten its internal product development cycles, bringing new AI-enabled capabilities to market quickly. And speed to market is speed to value.
- Facebook will have access to a lot of talent who are drawn to PWC and sotabench because these products help developers focus on what they like best – tinkering with code. And hence Facebook will have a pool from which to recruit. Back in 2018, a story on Medium claimed that “80% of all machine learning engineers work at Google or Facebook.” If anything, that proportion has only intensified. And with the market shifting to AI and ML as a service, having access to talent means pulling ahead of the competition.
- And finally, all this talent and AI code is in one place. And Atlas ML – pardon me, now Facebook – controls access to it. It reminds me of Google’s corporate mission “to organize the world’s information and make it universally accessible and useful.” But who cares about information (access) anymore? In the future, AI will be making all decisions, right? And so whoever controls the source AI code controls the future of the world. And for all of $40 million!
And so the race between the AI giants is on to the next level.
Want to Know More?
Read our other tech notes on AI and ML.
Using AI/ML software or tools? Leave a review.
Transparency, explainability, and trust are pressing topics in AI/ML today. While much has been written about why these are important and what organizations should do, no tools to help implement these principles have existed – until now.
Recently I attended the inaugural Emotion AI conference, organized by Seth Grimes, a leading analyst and business consultant in the areas of natural language processing, text analytics, sentiment analysis, and their business applications. So, what is emotion AI, why is it relevant, and what do you need to know about it?
SortSpoke’s novel approach to machine learning answers a longstanding problem in financial services – how to efficiently extract critical data from inbound, unstructured documents at 100% data quality.
Amazon is offering its cashierless store technology to other retailers. The technology known as “Just Walk Out” eliminates checkout lines, offering an “effortless” shopping experience and shifting store associates to “more valuable activities”.
As the COVID-19 pandemic is shutting down whole countries, a few of you may be wondering whether AI can help create a vaccine for the virus responsible. After all, AI is magic, right?
Alphabet is facing backlash from its shareholders over its approach to digital privacy, reports the Financial Times. And not for the first time. This time, however, things will need to change.
The EU plans to invest €6 billion to build a single European data space, reports EURACTIV. The envisioned space will house personal, business, and “high-quality industrial data” and create the infrastructure for data sharing and use across businesses and nations.
In a landmark ruling, a Dutch court has ordered an immediate halt to the government’s use of an automated system for detection of welfare fraud.
Databricks, a data processing and analytics platform with a strong focus on AI and ML, has partnered with Immuta to deliver automated end-to-end data governance for AI, data science, and ML projects.