#52: The 2020 State of AI report, Microsoft's Lobe visual ML model editor, and the Open Catalyst Project

Hey everyone, welcome to Dynamically Typed #52! I have quite a few links for you today. In productized AI, I’ve got the 2020 State of AI report, Google’s Duplex updates, and Microsoft’s Lobe (re)launch. For ML research, there’s Julian Togelius’ excellent new AGI essay and a tool to make 3D visualizations of deep learning models. On the climate AI side, Facebook and CMU have a new ML battery research project, and I found a systemic review of AI approaches to energy demand-side response. Finally, for cool things there’s an interesting EU-funded computer vision experience and a huge online gallery of AI art.

A request for Gmail users: A few readers let me know that Gmail demoted the last issue of DT to the Promotions tab for them, and my open rates show that a lot of people didn’t see the email at all because of that. That’s a bummer, but you can help me reduce the chances of it happening again in the future in two simple steps:

1. Star this email: on the Gmail website or mobile app, just tap the star icon somewhere around the top right of this email. (Bonus points if you star old issues too; you can find them by searching “from:[email protected]”.)
2. Move this email to Primary: on the web, drag this email to the Primary category and, on the bottom left pop-up notification, click Yes; on mobile, tap the more (“…”) button, tap Move to , and choose the Primary category.

If you did both, thanks so much! Hopefully it’ll teach Gmail that people actually care to see my emails. It’s of course a bit ironic that, after I chose an email newsletter as platform for my writing largely to avoid dealing with the whims of social media newsfeed content ranking algorithms, that is precisely what I find myself doing now. Oh well. With that out of the way, let’s get into today’s newsletter.

Productized Artificial Intelligence 🔌

• 📑 Nathan Benaich and Ian Hogarth’s 2020 State of AI report came out. It covers research, talent, industry, and politics, and is once again full of great in-depth data and analysis. It touches on many of the trends I’ve covered in DT this year, including gargantuan (Transformer) language models, productized NLP, reproducibility, accelerator chips, and self-driving progress. A few topics that are a bit outside DT’s scope but that are very interesting nonetheless include their assertion that biology is experiencing its “AI moment”, their analysis of talent education and flow, and their summary of geopolitical trends surrounding AI hardware and software companies. (Google Slides link; an executive summary is on slide 7.)
• ☎️ Duplex, Google’s “AI technology that uses natural conversation to get things done,” was first launched at the company’s 2018 I/O conference as a way to automatically make phone calls for reservations at restaurants or rental car services (see DT #13). It’s now being used in a lot more places, from calling businesses listed on Google Maps to automatically confirm their opening times, to screening potential spam phone calls. Personally I’d feel a little rude having Duplex make a reservation for me, but I think the use case of double-checking opening times is very useful — especially now, during the pandemic — since that single automated call can prevent a lot of people from showing up to closed doors if opening times are wrong on Google Maps.
• 🔗 Lobe, a web app to visually edit high-level compute graphs of machine learning models and train them, has (re)launched as a Microsoft product. “Just show it examples of what you want it to learn, and it automatically trains a custom machine learning model that can be shipped in your app.” The site’s UI looks super slick, and it can export models to TensorFlow (1.15, not 2.x), TFLite, ONNX, CoreML and more. I’d be very interested to find out what kind of optimizations it applies for the mobile and edge deployment targets — anything on top of the standard TFLite conversion, for example?

Machine Learning Research 🎛

• 🤖 In his follow-up to A very short history of some times we solved AI, Julian Togelius asks How many AGIs can dance on the head of a pin? Intelligence — let alone artificial general intelligence — is an extremely ill-defined concept; and the path from current machine learning software to a self-aware Terminator is… unclear, to put it mildly. So, Togelius argues that the popular philosophical AGI debates (about how to contain or align an exponentially self-improving intelligence explosion) are completely moot. “If you don’t believe in angels, it makes no sense discussing how much space they occupy. It just becomes a word game.” It’s a good read that once again very much aligns with my views on AGI.
• ⚡️ Efemerai is a tool to visualize, inspect and debug deep learning code. The visualization bit looks coolest: “Efemarai can scan the execution of your machine learning code and automatically generate interactive 3D visualizations. All you need is a single line of Python to explore the entire computational graph of your model with all of its values, parameters and gradients.”

I’ve also collected all 74+ ML research tools previously featured in Dynamically Typed on a Notion page for quick reference. ⚡️

Artificial Intelligence for the Climate Crisis 🌍

• 🔋 Facebook AI Research and CMU’s Department of Chemical Engineering have teamed up to launch Open Catalyst Project, a collaborative battery technology research effort in “using AI to model and discover new catalysts to address the energy challenges posed by climate change.” Practically, this comes in the form of a dataset, open-source baseline models, and a leader board.
• 💡Artificial intelligence and machine learning approaches to energy demand-side response: A systematic review (Antonopoulos et al., 2020) is a new 35-page paper examining how AI can be used to balance an electrical grid that is increasingly powered by variable renewable sources like wind and solar. “This work provides an overview of AI methods utilised for [Demand Response] applications, based on a systematic review of over 160 papers, 40 companies and commercial initiatives, and 21 large-scale projects.”

Cool Things ✨

• 👁 How normal am I? is a really cool EU-sponsored experiment that runs a bunch of machine learning models on your webcam video feed to determine your perceived beauty, age, gender, BMI, life expectancy, and “face print.” This all happens locally on your laptop (no data is uploaded to a server), and during the whole experiment a video of a friendly researcher talks you through the ways these models are being used by companies and governments in the real world. It takes about five minutes to run through the whole — quite eye-opening — experience. (In the end, the project considered me to be about 75% normal: “violently average.”)
• 🎨 ML x ART is a new 340-piece collection of creative machine learning experiments, curated by Google Arts & Culture Lab resident Emil Wallner. I came across a few projects I’ve featured here on DT, and tons I hadn’t seen before — I definitely recommend spending some time scrolling through it!

Thanks for reading! As usual, you can let me know what you thought of today’s issue using the buttons below or by replying to this email. If you’re new here, check out the Dynamically Typed archives or subscribe below to get a new issues in your inbox every second Sunday.

If you enjoyed this issue of Dynamically Typed, why not forward it to a friend? It’s by far the best thing you can do to help me grow this newsletter. 🇺🇸🦅🎉