AI Week: Apple makes machine learning moves

Image credits: Apple

Keeping up with a rapidly evolving industry like AI is a daunting task. So, until an AI can do it for you, here’s a handy roundup of the stories of the past few weeks in the world of machine learning, along with notable research and experiments we haven’t covered ourselves.

One could say that last week Apple very visibly and intentionally jumped into the ultra-competitive AI race. It’s not that the company hasn’t signaled its investments and AI prioritization earlier. But at its WWDC event, Apple made it abundantly clear that AI was behind many of the features in both the upcoming hardware and software.

For example, iOS 17, which is expected to arrive later this year, can suggest recipes for similar dishes from an iPhone photo using computer vision. AI also powers Journal, a new interactive diary that provides personalized suggestions based on the activities of other apps.

iOS 17 will also feature updated autocorrect powered by an AI model that can more accurately predict the next words and phrases a user might use. Over time, it will become tailored, learning the most used words by users, including swear words, in a fun way.

Artificial intelligence is central to Apple’s Vision Pro augmented reality headsets, also notably FaceTime on the Vision Pro. Using machine learning, the Vision Pro can create a virtual avatar of the wearer, interpolating a full range of facial contortions up to skin tension and muscle work.

Image credits: Apple

It might not be generative AI, which is arguably the hottest subcategory of AI today. But Apple’s intention, it seems to me, was to stage some sort of comeback to prove it’s not one to be underestimated after years of faltering machine learning projects, from the underwhelming Siri to the self-driving car in manufacturing hell.

Projecting force is not just a marketing ploy. Apple’s historic underperformance in AI has reportedly led to a major brain drain, with The Information reporting that talented machine learning scientists, including a team who had been working on the type of technology behind OpenAI ChatGPT, have left Apple for greener pastures.

Proving that he is actually serious about AIshipping AI-imbued products seem like a necessary move and a benchmark that some of Apple’s competitors have failed to achieve in the recent past. (I’m looking at you, Meta.) Apparently, Apple had a raid last week even if it didn’t particularly talk about it.

Here are the other notable AI titles from the past few days:

• Meta create a music generator: Not to be outdone of Google, Meta has released its own AI-powered music generator and, unlike Google, made it open source. Called MusicGen, Metas’ music generation tool can turn a text description into about 12 seconds of audio.
• Regulators look into AI security: Following the British government’s announcement last week With plans to host a global AI security summit this fall, OpenAI, Google DeepMind, and Anthropic have pledged to provide early or priority access to their AI models to support assessment and security research.
• AI, meets the cloud: Salesforce is launching a new suite of products aimed at strengthening its position in the ultra-competitive AI space. Called AI Cloud, the suite, which includes tools designed to deliver enterprise AI, is Salesforce’s latest interdisciplinary effort to augment its portfolio of AI-enabled products.
• Testing text-to-video AI: TechCrunch moved forward with Gen-2, Runways AI that generates short video clips from text. The verdict? There is still a long way to go before the technology comes close to generating cinematic quality footage.
• More money for enterprise AI: In a sign that there’s plenty of money to spend on generative AI startups, Coherewhich is developing an ecosystem of AI models for the company, announced last week that it had raised $270 million as part of its Series C round.
• No GPT-5 for you: OpenAI isn’t yet training GPT-5, OpenAI CEO Sam Altman said at a recent conference hosted by the Economic Times months after the Microsoft-backed startup pledged not to work on GPT-4’s successor for some time after that. many industry executives and academics have expressed concerns on the rapid rate of advancement of Altman’s great language models.
• AI Writing Assistant for WordPress: Automaticthe company behind WordPress.com and the main contributor to the WordPress open source project, launched an AI assistant for the popular content management system on Tuesday.
• Instagram gets a chatbot: Instagram may be working on an AI chatbot, according to the imagesleaked by app researcher Alessandro Paluzzi. According to the leaks, which reflect ongoing app developments that may or may not ship, these AI agents may answer questions or give advice.

Other machine learning

If you’re curious about how AI could impact science and research in the coming years, a team of six national labs wrote a report, based on workshops conducted last year, on just that. One might be tempted to say that based on last year’s trends and not this one, where things have progressed so fast, the report may already be out of date. But while ChatGPT has made huge strides in technology and consumer awareness, the truth isn’t particularly relevant to serious research. Large-scale trends are and are moving at a different pace. The 200-page report is definitely not light reading, but each section is helpfully divided into digestible parts.

Elsewhere in the National Labs ecosystem, Los Alamos researchers are hard at work advancing the field of memristors, which combine data storage and processing much like our own neurons. It’s a fundamentally different approach to computation, though it has yet to pay off outside the lab, but this new approach at least seems to be getting the ball rolling.

The structure of AI with speech analysis is on display in this report on police interactions with people they stopped. Natural language processing has been used as one of several factors in identifying linguistic patterns predicting escalation of arrests, especially with black men. Human and machine learning methods are mutually reinforcing. (Read the document here.)

Image credits: Cyrille Verdon / Renaud Defrancesco BUREAU 141 / EPFL

DeepBreath is a model trained on breathing recordings taken from patients in Switzerland and Brazil that its creators at EPFL say can help identify respiratory conditions early. The plan is to release it in a device called the Pneumoscope, under the Onescope spin-off company. Well, probably follow them for more information on how the company is doing.

Another AI health advance comes from Purdue, where researchers have built software that approximates hyperspectral images with a smartphone camera, successfully monitoring blood hemoglobin and other metrics. It’s an interesting technique: Using the phone’s super-slow-mo mode, it gets a lot of information about every pixel in the image, giving a model enough data to extrapolate from. It could be a great way to get this kind of health information without special hardware.

Image credits: MIT extension

I still wouldn’t trust an autopilot to perform evasive maneuvers, but MIT is approaching the technology with research that helps AI avoid obstacles while maintaining a desirable flight path. Any old algorithm can propose wild changes of direction in order not to crash, but doing it while maintaining stability and without breaking anything inside is more difficult. The team managed to get a simulated jet to perform some Top Gun-like maneuvers autonomously and without losing stability. It’s harder than it looks.

The latest this week is Disney Research, who can always be counted on to show something interesting that also applies to movie theater or theme park operations. At CVPR they demonstrated a powerful and versatile facial landmark detection network that can track facial movements continuously and using more arbitrary landmarks. Motion capture already works without the catch dots, but that should make it even more quality and more dignified for the actors.