Radar trends to watch: August 2020
Trends in COVID-19, AI, data, robotics, programming, VR, technology and society, and security.
I thought July was going to be a dull month, but I’m wrong again. COVID-specific technology seems to be drying up, though there’s a fascinating report about a DIY vaccine. (Developed by serious scientists, so don’t try this at home.) There’s a lot of news about AI, and specifically, about the GPT-2 and GPT-3 language models. And a few things that are just fun, like Festo’s Bionic Swifts.
- A DIY vaccine for COVID-19 is circulating among a group of scientists at Harvard and MIT. It comes from a group called RADVAC (Rapid Deployment Vaccine Collaborative); it was developed by Preston Estep, a protege of George Church.
- Wearables (particularly Apple Watch and Fitbit) may be able to detect COVID-19 infections in their users by constantly monitoring heart rate, temperature, and other parameters with a good understanding of the wearer’s baseline metrics.
- OpenAI’s GPT-2 natural language generation has now been trained to generate images, a significant step forward in the creation of realistic fake video.
- OpenAI has released GPT-3, the next generation of their language model. It’s impressively good, and may unlock many new applications that require accurate parsing and generation of human language. Still, it’s important to remember that it is just a model, and that it is not intelligent.
- Using AI to generate pseudocode from human source code may be a better route to AI-assisted programming than AI-generated source code. Pseudocode places AI in the position of an assistant or a helper that helps you analyze what the code does, rather than an oracle.
- Another development in AI-assisted programming is a neural network that compares the code being written to a body of existing code to detect possible bugs.
- A promising new voice separation model allows voice recognition to distinguish up to five voices speaking simultaneously without knowing the number of speakers in advance.
- What does “too autonomous” mean for spacecraft? raises important questions about the interface between humans and AI. What problems arise from reliance on autonomous systems? It’s also worth looking at this article about Basecamp’s Hey! and the uncanny valley.
- Building Anti-Racist Products is the start of a series from ProjectsByIf that explores the shortcomings of human-centered design, and suggests alternatives that don’t encode biases into our technology.
- One of the biggest issues facing machine learning is fitting it into current practices for deploying software. CML is an open source project developing tools for continuous integration and continuous deployment that are appropriate for machine learning.
- Google has introduced a toolkit for creating model cards. Model cards are essentially data sheets for models, recording benchmarks about how the model performs for different groups, describing known biases, situations in which the model should or shouldn’t be used, and more.
- Europe has struck down the law enabling the EU-US Data Privacy Shield because of the US’ use of data for surveillance. This means it will be much more difficult for European companies to legally store data in the US.
- Google’s BigQuery isn’t just for Google: it’s a multicloud tool. BigQuery Omni can execute queries across multiple clouds, including AWS and Azure.
- R jumps from #20 to #8 on the Tiobe index–possibly because of analysis of COVID data. The Redmonk rankings don’t have a similarly dramatic shift, but have R at #13 based on their analysis of StackOverflow posts and Github projects.
- Festo’s BionicSwifts: Festo has long done amazing things with biomimicry, ranging from jellyfish and dragonflies to seagulls. This is the latest. It’s a Swift (related to swallows) that weighs in at 42 grams, including a 6 gram battery. It has 7 minutes flying time. And it isn’t as scary as Boston Dynamics’ creations.
- Spacial perception for robotics: Identifying objects is easy, but building a model of a 3D scene is difficult, and an essential problem for general-purpose robotics.
- Google has created the “Open Usage Commons” as a “trademark office” for open source trademarks. (Allison Randal and Chris DiBona are on the board.) It’s worth watching because Adam Jacob has identified trademarks as a significant asset in an open source business model. (And yes, this commons is an important part of the ongoing fuss over Istio and the CNCF.)
- Perl 7 has been announced, and should be available by the end of the year. Perl 7 is essentially a return to Perl 5 with some updates. Will this reignite interest in Perl, which recently showed up on a Stack Overflow survey as the third most dreaded programming language?
- No code may be the next generation of software development. What does this mean for training? What new paradigms will be developed for making computers do what we want? We could be headed for a big change in how people interact with computers.
- A lightweight wristband is capable of sensing all of the hand’s motions in 3D using thermal sensing and machine learning. Gestural interfaces have so far been rather lame; this could give the field a new start.
- Here’s a glimpse at Facebook’s VR glasses. They are still proof-of-concept, but this prototype (along with Apple’s rumored product, and Google’s continued interest) means that nobody can count VR out. And that Google was on the right track with Google Glass a decade ago.
Technology and Society
- TikTok, the social media site for short videos, has over 800 million active users. What’s especially important isn’t that it’s a Facebook-scale social media application, but that it is based in China. China-based technology companies making serious inroads into the US market may be the shape of the future.
- Facebook advertisers boycott: Unlike other actions against Facebook (remember #DeleteFacebook?), #StopHateForProfit looks like it might have an effect.
- Microsoft’s Project Freta is about trusted sensing (malware detection) for the cloud: full non-invasive memory audits of thousands of VMs in the cloud.