Radar trends to watch: June 2020
Trends in COVID-19, programming, machine learning & AI, payment systems, and networks.
Most of the technical news this month continues to swirl around coronavirus. Many things are happening under that rubric—for example, delivery during a pandemic could be the killer app for autonomous vehicles. And money being (literally) dirty, the pandemic could drive the development of new payment systems that are more inclusive. I haven’t seen as much interesting news from the AI world this month. There are some hints that we’re reaching the point of diminishing returns for large models, and that we’re turning the corner from research into deployment and production.
- A startup is developing a CRISPR-based fast and accurate COVID-19 test that can be used at home. At-home testing raises questions about reporting, but since the test gives quick results, it could also be used by airlines, offices, and other crowded locations.
- One consequence of the coronavirus pandemic may be that there aren’t enough people to work in call centers. Will AI take up the slack? Natural language applications for automating call centers is a significant strategic bet for a number of big AI players, including IBM.
- Coronavirus may be a boon for autonomous vehicles: we may soon see driverless delivery services and driverless taxis, both of which take the driver out of the infection disease cycle. CVS is testing a self-driving prescription delivery service.
- Permanent work-from-home at Twitter: On one hand, working from home wherever possible is a no-brainer, if for no other reason than minimizing office expense. But it has huge consequences for the real-estate and building industries if other companies follow suit. It also may have consequences for salaries.
- Covid Tracker Tracking: Quis custodiet ipsos custodes (Who watches the watchers)? MIT, evidently.
- The startup Miso has developed a tool for doing smart, AI-based searches on COVID-19 research. A number of similar tools are appearing, including Google’s COVID-19 Research Explorer and SciFact from the Allen Institute. Given the number of research papers that have been generated, AI-based search and summarization may be the only way to stay current.
- In the wake of conference cancellations, venues for online conferences and meetings are evolving rapidly; it’s not just Zoom or GoToMeeting. Discord has a lot of potential; The Deserted Island Devops conference appears to have taken place in Animal Crossing, a multiplayer video game. The conference was live-streamed on Twitch. I’ve also seen references to The Online Town and Clubhouse Voice Chat. The winner will be whoever can replicate the in-person “hallway session” experience.
- Will the open office survive COVID-19? (Personal conversation) Good question. Open offices have already lost favor, at least among employees. Viruses spread like wildfire in open spaces with shared ventilation. Keeping office buildings at 25% capacity isn’t attractive–and if HVAC is a big part of the problem, unlikely to be effective.
- NuShell is a new cross-platform shell, updating the notion of pipes and built around tables as first-class objects. It runs on Windows, Linux, and OS X. The command line will remain a powerful tool.
- The New Stack is talking about Serverless Cloud mashups. This is also the point of James Urquhart’s “flow architectures,” which I suspect has been missed: what James thinks is radical about the next generation of enterprise software is that it will be as flexible as the web.
Machine Learning and AI
- Comment-driven development: Microsoft has developed AI that writes functions based on a comment and the function signature (skip to 29:00 of the video). This is obviously a demo with extremely simple examples, but this could be a big step forward in re-thinking programming. Programming would be less about describing processes in excruciating detail, and more about analyzing the problem itself.
- Early Bird is a technique for faster and less power-hungry training for neural networks by doing network pruning (identifying the nodes that will be part of the trained network) very early in the process. They claim to use a factor of 10 less energy. There are two important questions: does that just mean that people will train larger models? And do larger models actually accomplish anything? The natural language model GPT-3, which is the largest model we know of, suggests that we’ve reached a point of diminishing returns. And that we’re asking the wrong questions, particularly for language models.
- PyCaret is a new machine learning library for Python that requires very little coding.
- Affirm (buy-now-pay later, essentially a one-shot credit card) seems to have flown under the radar. Credit is only extended for the purchases in a single shopping cart, with credit-worthiness and loan terms evaluated in real-time. Interest rates are based on your ability to pay: ⅓ of the loans are at 0%, but can be up to 30%, which seems usurious. Users are given the total payments and interest charges before they accept the loan. Payment as a microservice…
- The Mojaloop Foundation, with extensive support from Google and the Gates Foundation, is attempting to create a digital payments system with inclusion as a goal. This could be seen as competition for Facebook’s Libra—although Libra appears to be dead before it started.
- As long as we’re on payments: will the dollar continue to dominate? China continues to work on their electronic currency (the e-RMB), which could become a compelling alternative to the dollar.
- The Open RAN Policy Coalition is attempting to create open, interoperable standards for 5G. Such a coalition is needed because the major networks and vendors are all pursuing slightly different versions of a standard that isn’t really standard. The coalition includes Microsoft, Google, IBM, Cisco, AT&T, and Verizon. It may also be an attempt to exclude China, since Chinese companies are not represented.