In his book TinyML, Pete Warden talks about smart stickers that can do limited AI, communicate via radio, and contain sensors so they can easily be put onto machinery or other objects. That technology is here, with disposable bluetooth stickers powered by ambient RF.
A year ago, Foster Provost said that causality is the next big thing in data science. That’s a trend that bears watching for the next few years; it’s starting to pay off with some medical AI technology that looks for cause and effect, not just correlation.
Democratization of deep learning isn’t a new idea, but we haven’t thought much about how democratization progresses. Pre-trained and partially trained models significantly reduce the cost of training and the time needed to train a system, along with the expertise required.
One of the biggest issues facing AI is the transition from research and development to production and operations. One aspect of this transition is the challenge of serving extremely large models, such as OpenAI’s GPT-2 and Google’s Meena. It’s no coincidence that both of these are natural language models.
We’re seeing more and more signs that machine learning and edge computing go together; in fact, that may be the only way that edge computing makes sense. At the O’Reilly Software Architecture Conference in New York, Mary Poppendieck talked about the absurdity of her robotic vacuum cleaner not working during a two-day internet outage. This kind of intelligence needs to be at the edge, rather than in the cloud.
The Coronavirus is already having a significant impact on the economy: shutdown of factories in China and disruption of supply chains is a big deal, even if there isn’t a pandemic.
The Coronavirus may present a significant opportunity for educational technology and online education. China is doing some work with AI in education; this may or may not pan out, but they don’t want to fall behind even as they’re closing schools.
Autonomous vehicles on the farm: we’ve known for a long time that farms are technologically sophisticated. And they may be the first place we see autonomous vehicles rolling out. You don’t have to worry about traffic laws or pedestrians.
Cornell’s College of Agriculture is hosting a digital agriculture hackathon. Schools in the Netherlands and China are playing key roles. Agriculture is already a high-tech field, and will only become more so. There’s also a startup called Tillable that thinks of itself as an “AirBnB for farmland,” and that is already running into trouble for abusing farm data.
It isn’t new or surprising that government comment websites are hacked by bots. We saw that with the FCC and the demise of network neutrality. And I suppose it’s also not a surprise that government sites are particularly open to attack.
Another Lidar: this one is an AI-based intrusion detection system for cyberattacks. Sort of a very smart honeypot. It’s a rules-based system to detect attacks, but it also has the ability to detect anomalies and collect information about new kinds of attacks.
Future of computing
Artificial qubits are potentially a significant breakthrough in practical quantum computing.
RapidQL is an open source graphical query language for consolidating APIs (single sign-on, access to multiple web APIs via REST, SOAP, and GraphQL). API consolidation is interesting in itself, but I’m also interested in it as a step away from text-based programming.
Mark Zuckerberg proposes regulation based on putting appropriate systems in place, not liability for content. This has been criticized as avoiding the real issue, which is the content; but probably the only reasonable way to regulate social media is based around capabilities and systems, rather than individual pieces of content.
zsh has been around for a long time without getting a lot of traction, but it may be coming out of the shadows. Don’t forget that it’s the default shell in OS X 10.15, replacing bash.
DNS may be the most problematic part of Internet infrastructure–and it’s becoming even more problematic with issues like the potential sale of the .org registry. The Handshake Network is a blockchain-based alternative to ICANN’s control of the centralized DNS root registry. A lot of problems have to be solved to make Handshake work, but it sounds to me like a natural application of the blockchain.