Welcome to WhatTheyThink’s Quarterly Miscellany.

Spicing IT Up

OK, this has gone too far. An “AI-powered spice dispenser”?!?! Good grief. Says The Byte:

Enter the Spicerr, a supposedly “AI-powered” “smart” spice dispenser that will automatically decide how much seasoning you should add to your barren foodstuffs. 

They remind us of the Juicero, a fruit and vegetable juicer that had been the darling of the health food industry until Bloomberg revealed that its plastic juice packs could easily be squeezed by hand rather than with a $700 machine. Oops. Anyway, back to the Spicerr:

The Spicerr is designed like a minimalist, tech-inflected pepper grinder with a revolver’s cylinder stuck on the bottom. It holds six pre-packaged spice capsules at a time, which you have to buy from the manufacturer, like so many hated inkjet printers.

What, no subscription or “Spice as a Service”?

Spicerr sells an “Essential Collection” that comes with black pepper, turmeric, crushed pepper, ginger, cinnamon, and cumin, as well as three other collections for “family cooking,” “baking with kids,” and plain ol’ “BBQ.”

We know we said this thing has the “AI” label slapped on it, but it’s unclear what exactly the “AI-powered platform” actually is, other than something that collects your data, apparently, via its accompanying app.

We think it’s a marketing ploy rather than anything technical.

“By analyzing your preferences and interactions, Spicerr quickly learns your tastes and suggests dishes and spice blends perfectly suited to your palate,” the website reads.

Anyone who’s had a palate for any length of time can probably figure out what their tastes are without data. Or, better yet, just go ahead and add some paprika willy nilly. Throw caution to the wind! Use an entire bay leaf! Live dangerously! Carpe diem—seize the fish!

Lost In Translation?

One increasingly common use for “AI”—or, more specifically, large-language models (LLMs)—is translation apps, and in our experience these apps do a fairly decent job. Obviously, where they fall down are things like idiom, region-specific terminology, and industry jargon. (We used a translation app called Deepl for all the German translations of articles for the drupa Daily last year, and we did need a human overseeing the translation as it got a lot of printing industry-specific terms wrong.) We also used Google Translate to get around Düsseldorf whilst we were in town for drupa. (We also used Google Translate in 2020 during the pandemic lockdown when we went down a rabbithole into a music genre called rock progressivo Italiano and used the app to translate song lyrics from Italian. Not the most poetic approach but at least it gave a sense of what the singer was on about. And one wonders how much of the oddball lyrics were the result of translation errors or whatever the lyricist was on. Probably a bit of both…)

Anyway, The Conversation points out that, to understand the future of AI, we should look at some of the failures of translation apps. As long ago as 2006, when it launched, Google Translate only supported three languages: English, Chinese, and Arabic. And while today it supports 249, that’s still only a fraction of the world’s estimated ~7,000 languages. (We do not know if Klingon is supported.)

Between a handful of those languages, like English and Spanish, translations are often flawless. Yet even in these languages, the translator sometimes fails on idioms, place names, legal and technical terms, and various other nuances.

Between many other languages, the service can help you to get the gist of a text, but often contains serious errors. The largest annual evaluation of machine translation systems – which now includes translations done by LLMs that rival those of purpose-built translation systems – bluntly concluded in 2024 that “MT is not solved yet”.

In 2021, the Google Translate app reached 1 billion installs, and there is some comfort in knowing that surveys have found that users generally get that they should use these apps cautiously and only for “low stakes” applications, like getting the gist of foreign language content—like, say, Italian song lyrics, wayfinding signage, or German train ticket machine instructions. Only 2% of translation app users say they use it for “high stakes” applications, like interacting with healthcare professionals or the police, and you can imagine what sorts of problems can arise.

As a result, as we did with drupa content, professionals in international law and commerce and other high stakes areas rely on human translators to oversee the machine translations.

Many human translators are freelancers in a marketplace mediated by platforms with machine-translation capabilities. It’s frustrating to be reduced to wrangling inaccurate output, not to mention the precarity and loneliness endemic to platform work. Translators also have to contend with the real or perceived threat that their machine rivals will eventually replace them – researchers refer to this as automation anxiety.

We like that term: automation anxiety. So what does this bode for the future:

the forseeable future for LLMs is one in which they are excellent at a few tasks, mediocre in others, and unreliable elsewhere.

Hmm…sounds like most humans.

We will use them where the risks are low, while they may harm unsuspecting users in high-risk settings – as has already happened to laywers who trusted ChatGPT output containing citations to non-existent case law. 

At the end of the day:

These LLMs will aid human workers in industries with a culture of quality assurance, like computer programming, while making the experience of those workers worse. Plus we will have to deal with new problems such as their threat to human artistic works and to the environment. The urgent question: is this really the future we want to build?

Brain Computing

Here’s a headline to reckon with. From Futurism: “Weird New Computer Runs AI on Captive Human Brain Cells.” (The deck head sounds like a Google Translate fail: “And you can buy compute on the cloud.” Huh?)

Australian startup Cortical Labs just launched the “world’s first code deployable biological computer.” Say what?

The shoe box-sized device, dubbed CL1, is a notable departure from a conventional computer, and uses human brain cells to run fluid neural networks.

Back in 2022, Cortical Labs made the news when it taught human brain cells in a petri dish how to play Pong.

The CL1, however, is a fundamentally different approach, as New Atlas reports. It makes use of hundreds of thousands of tiny neurons, roughly the size of an ant brain each, which are cultivated inside a “nutrient rich solution” and spread out across a silicon chip, according to the company’s website.

… “A simple way to describe it would be like a body in a box, but it has filtration for waves, it has where the media is stored, it has pumps to keep everything circulating, gas mixing, and of course temperature control,” Cortical Labs chief science officer Brett Kagan told New Atlas late last year.

Ah, and that’s the simple way to describe it.

It has yet to prove itself actually useful, but the company sees many potential applications.

“There’s so many different options,” he told Australian broadcaster ABC News, suggesting it could be used for “disease modelling, or drug testing.’

Or perhaps even correcting grammar.

Nonetheless, the approach could have some key advantages. For instance, the neurons only use a few watts of power, compared to infamously power-hungry AI chips that require orders of magnitude more than that.

One has to ask, though: where are they getting the brain cells? And whose? Abby Normal?

AI Goes to the Movies

Those of us who like foreign films and spent much of our college years discovering the works of Fellini, Bergman, Kurosawa, etc., generally prefer subtitled to dubbed films. But, on occasion, beggars can’t be choosers and we have to rely on a dubbed version. Now, via Futurism, steaming services are using AI to create dubbed foreign films. And it sounds horrible.

From the streaming service that brought you crappy AI-generated movie posters and totally nonsensical AI-generated synopses, Amazon Prime Video presents: "AI-aided" dubbing! Which will replace actors' original dialog with a translated, machine-amalgamated mess. That's movie magic, people.

Amazon announced that it will be debuting the feature in English and Latin American Spanish for 12 movies and shows, including the 2003 animated film El Cid: La Leyenda. The folks at Futurism are not fans:

Amazon has been pretty ardent on AI, and its huge streaming platform has become a petri dish for all kinds of grotesque machine-generated experimentation. Last fall, for example, it began offering AI-generated recaps for TV shows. Also included in that suite of features? A generative AI tool to recommend you movies with similar plot points and character arcs to your favorite films, just to give you an idea of how much it wants to soullessly codify all spontaneity in art.

Still, Amazon hasn’t revealed just how the AI dubbing is being executed, or what kind of quality control may be involved.

“This AI-aided pilot program is a hybrid approach to dubbing in which localization professionals collaborate with AI to ensure quality control,” the company said in the announcement. “AI-aided processes like this one, which incorporate the right amount of human expertise, can enable localization for titles that would not otherwise be accessible to customers.”

Still, we prefer subtitles…unless they’re also AI-generated.

All the AI That’s Fit to Print

Il Foglio, an Italian newspaper, recently conducted an experiment in which they had AI generate an entire issue. Actual journalists were limited to asking questions of a chatbot and checking the answers before inserting them. The idea wasn’t necessarily to eventually replace actual journalists. Says Gizmodo:

Claudio Cerasa, editor of Il Foglio, said the experiment serves as a test for how AI could work “in practice” in a newsroom and forces journalists to ask tough questions about what the technology means for the future journalism.

“It will be the first daily newspaper in the world on newsstands created entirely using artificial intelligence,” said Cerasa. “For everything. For the writing, the headlines, the quotes, the summaries. And, sometimes, even for the irony.”

The four-page “Il Fogolio AI” was part of the larger proper Tuesday edition, and can also be read online, if you know Italian. (Hmm...you could use Google Translate’s AI to read an AI-generated newspaper. How meta…)

Generative artificial intelligence is good at producing verisimilitudes of genuine writing, something that looks clear and authoritative. There have been attempts to improve the “thinking” process of chatbots, but they are ultimately glorified autocomplete systems and face the intractable problem of simply making things up. Chatbots that present their logic as they produce a response will even sometimes admit as much. Ultimately, the problem with all language models is that the user has to look closely at all the generated text and correct errors if they even spot them at all. Newsrooms in particular have to be careful to not harm their credibility amongst the public even further by publishing slop.

Note that we do not use AI to generate our Around the Web miscellanies. Make of that what you will…

Watch This

Some companies are so committed to AI that they are using it for things that do not really require it. Take, for example, a watch. Watches have been telling the time for centuries. Is AI really necessary? Apple seems to think so. Says Gizmodo:

According to Bloomberg, Apple is working on a version of its smartwatch that will be equipped with cameras that will facilitate the expansion of the company’s “visual intelligence” features that are currently only available on the latest model of iPhone.

It has some potential uses, we suppose:

Per Bloomberg, the company is working to shoehorn the lenses into both its Series and Ultra models of the Apple Watch, with the Series expected to have a camera inside its display while the Ultra would have its camera on the side of the watch. Through the extra eye, the watch would be able to tap into the company’s visual intelligence tools, which can do things like identify objects or translate text in a picture.

True, real-time translation would be a benefit, but see the caveat about AI translation above.

And because the Apple Watch has an actual purpose beyond just being an AI novelty device—something that failed rivals like the Humane Ai Pin cannot claim—it’s probably a worthy trojan horse for AI features. Also, as is always the case with these things, it’d help Apple collect way more data to train its own systems. Meta has found some success with its Ray-Ban partnership to produce smartglasses, reportedly selling more than one million pairs, so there is some path for the AI wearable when paired with other purposes.

Artificiale Italiano

Now, granted, we have been to restaurants (i.e., not expensive ones) that had badly printed menus that showed that their printer desperately needed some kind of color management solution to make the food not look like some kind of hazardous waste. But this? Surely it’s possible to acquire images for a food menu without resorting to AI. Unfortunately, Uber Eats thought otherwise, and created utterly unappealing food images. Via The Futurist:

The situation was spotted by X-formerly-Twitter personality Online Boy, who noticed that the online Uber Eats menu for a restaurant called Roma Pizza, in New York City, appeared to have used AI to generate pictures of its cuisine, resulting in monstrosities… 

Well, first of all, it confused a “pizza pie” with a dessert pie:

And what in the name of all that’s holy is this:

Skip the sauce, we’d need just the vodka to consume that.

“Impressive example of Uber Eats doing more work to deliver an inarguably worse experience,” wrote one onlooker. “Welcome to the future of AI powered experiences.”

Or they could at least hire a content developer who has ever been to a restaurant or seen, you know, food.