While October’s news was dominated by Facebook’s (sorry Metas) continuing issues (one would think they would get tired of the apology trip), the most interesting news comes from the AI world. I am fascinated by the use of large language models to analyze whales’ “speech” and preserve endangered human languages. It is also important that machine learning seems to have taken a step (pun something intentional) forward, with robots learning to walk by trial and error, and with robots learning to assemble themselves to perform specific tasks.
- The design studio Artefact has made a game to teach middle school students about algorithmic bias.
- Researchers are building large natural language models, potentially the size of GPT-3, to decode the whales’ “speech”.
- A group in Berkeley has built a robot that uses reinforcement learning to teach itself to walk from scratch – ie. through trial and error. They used two levels of simulation before loading the model into a physical robot.
- AI reinvents computers: AI drives new kinds of CPUs, new “out of the box” form factors (doorbells, appliances), decision making rather than traditional computing. The “computer” as the computing device we know may be heading out.
- Strange creatures: Unimal or universal animals are robots that can use AI to develop their body shapes so that they can solve problems more efficiently. Future generations of robots may not be designed with fixed bodies, but have the ability to adapt their shape as needed.
- Would a National AI Cloud be a grant to Google, Facebook, et al., A threat to privacy or a valuable academic research tool?
- I have been skeptical of digital twins; they appear to be a technology looking for an application. However, digital twins (AI models of real systems, used to predict their behavior) seem like a useful technology to optimize the performance of large batteries.
- Digital twins could provide a way to predict supply chain problems and circumvent shortages. They could allow manufacturers to find a compromise between just-in-time storage processes, which are vulnerable to shortages, and resilience.
- Modulate is a startup that is currently testing real-time voice change software. They provide realistic, human-sounding voices that replace the user’s own voice. They are aimed at gambling, but the software is useful in many situations where harassment is a risk.
- Voice copying algorithms were able to fool both humans and voice-activated devices about 50% of the time (30% for Azure’s voice recognition service, 62% for Alexa). This is a new front in deep forgery.
- Facebook AI Research has made a set of first-person (main camera) videos called Ego4D for AI training. They want to build AI models that see the world “as a person sees it”, and be able to answer questions like “where did I leave my keys.” Basically, this means that they will have to compile literally everything a subscriber does. Although Facebook denies that they are thinking of commercial applications, there are obvious links to Ray-Ban Stories and their interest in augmented reality.
- DeepMind is working on a deep learning model that can mimic the output of any algorithm. This is called Neuro Algorithmic Reasoning; it can be a step towards a “general AI.”
- Microsoft and NVIDIA announce a native language model of 530 billion parameters called Megatron-Turing NLG 530B. It is larger than GPT-3 (175B parameters).
- Can machine learning be used to document endangered native languages and help with language recovery?
- Beethoven’s 10th Symphony completed by AI: I’m not convinced that this is what Beethoven would have written, but this is better than other (human) attempts to complete the 10th that I’ve heard. It mostly sounds like Beethoven, though it quickly becomes pointless.
- I’m still fascinated by techniques to prevent face recognition. Here is an article on an AI system that designs minimal, natural-looking makeup that reshapes the parts of the face that face recognition algorithms are most sensitive to, without significantly changing a person’s appearance.
- Thoughtworks’ Responsible Tech Playbook is a curated collection of tools and techniques to help organizations become more aware of bias and become more inclusive and transparent.
- Kerla is a Linux-like operating system kernel written in Rust that can run most Linux executable files. I doubt this will ever be integrated into Linux, but it’s another sign that Rust has joined the big time.
- Turing Complete is a game about computer science. That says it all …
- wasmCloud is a runtime environment that can be used to build distributed systems with wasm in the cloud. WebAssembly was designed as a programming language neutral virtual machine for browsers, but it increasingly looks like it will also find a home on the server side.
- Adobe Photoshop now runs in the browser using wasm and the Emscript (C ++ tool chain for wasm). In addition to compiling C ++ to wasm, Emscript also translates POSIX system calls to web API calls and converts OpenGL to WebGL.
- JQL (JSON Query Language) is a rust-based language to query on JSON (what else?).
- Microsoft has launched an effort to train 250,000 cybersecurity workers in the United States by 2025. This effort will work with community colleges. They estimate that it will only account for 50% of the lack of security talent.
- Integrating zero-confidence security into the software development lifecycle is truly the only way forward for companies that rely on systems that are secure and accessible.
- A supply chain attack on a Node.js library (UA-Parser-JS) installs cryptocurrencies and Trojan horses to steal passwords on Linux and Windows systems. The normal function of the library is to parse user agent strings, identify the browser, operating system and other parameters.
- A cybercrime group has set up consulting companies for penetration testing, the purpose of which is to acquire customers and then gather information and launch ransomware attacks against these customers.
- A federated cryptographic system will allow sharing of medical data without compromising patient privacy. This is an essential element of “predictive, preventive, personal and participatory” medicine (aka P4).
- The European Parliament has taken steps to ban surveillance based on biometric data, private face recognition databases and predictive policing.
- Is it possible to reverse-engineer the data on which a model was trained? An attack on a fake face generator was able to identify the original faces in the training data. This has important implications for privacy and security as it seems to generalize to other types of data.
- Conflicting attacks on machine learning systems pose another set of cybersecurity challenges. Models are not code and have their own vulnerabilities and attack vectors. Atlas is a project to define the threat landscape of machine learning. Tools for hardening machine learning models against attack include IBM’s Adversarial Robustness Toolbox and Microsoft’s Counterfit.
- Researchers have discovered that you can code malware into DNA that attacks sequencing software and gives the attacker control over your computer. This attack has not (yet) been found in the wild.
- Masscan is a next generation, extremely fast port scanner. It’s similar to nmap, but much faster; it claims to be able to scan the entire internet in 6 minutes.
- ethr is an open source tool for measuring network performance across platforms developed by Microsoft in Go. Right now, it looks like the best network performance tool out there.
- Self-conscious systems constantly monitor themselves and are able to detect (and even repair) attacks.
Infrastructure and operations
Devices and things
- Amazon is working on an internet-enabled refrigerator that keeps track of what’s in it and notifies you when you are short of supplies. (And there are already similar products on the market.) Do you remember when this was a joke?
- Consumer-oriented AI: On the one hand, “smart gadgets” offer a lot of challenges and opportunities. On the other hand, it needs better deliveries than “smart” doorbells. Smart hearing aids that can be upgraded in the field as a subscription service?
- A drone has been used to deliver a lung for organ transplantation. This is only the second time a drone has been used to carry organs for transplant.
- Intel has released its next generation of neuromorphic processor, Loihi. Neuromorphic processors are based on the structure of the brain, where neurons send signals to each other asynchronously. Although still a research project, they seem to require much less power than traditional CPUs.
- ipleak and dnsleaktest are pages that tell you what information your browser is leaking. They are useful tools if you are interested in maintaining privacy. The results can be daunting.
- Dark design is the practice of designing interfaces that manipulate users into doing things they may not want to do, whether it’s accepting to provide information about their Internet use or clicking to purchase a product. Dark patterns are already common and are becoming more prevalent.
- Black Twitter has become the new “Green Book”, a virtual place for tips on dealing with a racist society. The original Green Paper was a Jim Crow-era publication that told blacks where they could travel safely, which hotels would accept them, and where they were likely to fall victim to racist violence.
- A group at Duke University has made significant progress with quantum error correction. They have made a “logical qubit” that can be read with a 99.4% probability of being correct. (Still far below what is necessary for practical quantum calculation.)
- There are now two claims of quantum domination from Chinese quantum computer projects.
- Would our response to the COVID pandemic be better if it were treated as an engineering problem rather than scientific research?