Close Ad

The Spoon

Daily news and analysis about the food tech revolution

AI

Can You Use AI to Create a Wi-Fi Powered Rolling Pin? Join Today’s Interactive Workshop to Find Out!

Last June at the Smart Kitchen Summit, Scott Heimendinger talked about today’s temptation among product builders to follow a similar path with AI as consumer product builders forged a decade ago when they added Wi-Fi connectivity to almost everything.

“There’s a temptation that’s especially prevalent in our industry, but in others as well, that when new technologies become available to the world, we want to kind of slap those on what we’re doing,” he said. “We slapped WiFi on a bunch of things, but the world does not need a WiFi-enabled rolling pin.”

This doesn’t mean Scott doesn’t embrace AI as an inventor and product builder to help make him more productive in doing his job. It’s the opposite, and the one-time technical lead for Modernist Cuisine and founder of one of the earliest consumer sous vide hardware startups in Sansaire leverages AI tools to assist across the product development process. From ideation to market research to writing code, Heimendinger tells me that AI is a crucial tool to help him do his job faster and more efficiently.

If you’re looking for ways to figure out how to use AI in your job, join us this morning for this month’s Food AI Co-Lab as Scott shows us how to leverage these tools as we, yes, build an imaginary new product in a Wi-Fi rolling pin and look to bring it to market.

Join us for today’s interactive workshop at 8:30 Pacific Time to learn how a successful product developer and inventor like Scott leverages AI to up his game as he shares practical tips for you to do the same. Scott will work through his process and show ways he leverages a variety of tools and offer tips for you to do just the same. And who knows, maybe you’ll come away with an idea for the next big product (just not an AI-rolling pin, ok?). Register now!

Anova Serves Up a Generous Helping of AI With Launch of Anova Precision Oven 2.0

Last week, Anova announced the second generation of its Precision Oven, just over four years after it began to ship the first generation Precision Oven. The Anova Precision Oven 2.0 is packed with a number of new features, including an in-oven camera and Anova Intelligence, a suite of new AI features designed to power new ways for users to assist in the cooking process.

In fact, the company listed a bunch of features currently offered by their AI-powered cooking and a number of features that they are working on. The features the Anova Precision Oven 2.0 currently has include:

  • Ingredient Recognition: The AI system automatically identifies what’s inside the oven with the internal camera.
  • Suggested Cooking Methods: The oven’s AI will suggest cooking methods tailored to the ingredients, ranging from basic roasting to more complex sous vide style.
  • Packaged Food Conversion: The oven will scan the packaging, and the AI will choose the right settings.
  • Recipe Conversion: The company says the Anova Precision Oven 2.0’s AI can convert nearly any recipe to make it work with its settings, with the caveat that this feature will improve over time as it gains more data.

According to Anova, upcoming features for Anova Intelligence include:

  • Assistant Mode: Anova’s AI-powered co-pilot will simplify complex cooking techniques and offer personalized cooking guidance.
  • Complex Meal Creation: When preparing multi-component dishes, the oven will suggest optimal settings for each ingredient, “streamlining” the cooking process for recipes that typically require juggling multiple cooking techniques.
  • Cook Recall: The oven will recognize repeat recipes for dishes you prepare frequently and return to your last-used settings.
  • Doneness Detection: Powered by the internal camera, the oven monitors the cook’s desired crispness level and alerts them when it reaches the preferred doneness level.
  • Auto Shutdown: The oven will detect when a cook cycle has finished and whether food has been removed, then notify you before automatically shutting off.
  • “Clean Me” Reminders: Equipped with an internal camera that monitors for dirt buildup, the oven will remind you when it’s time for a cleaning.

With the addition of the camera and new AI features, it looks like Anova hopes to fill the void left after Weber sunsetted the June oven about a year ago. While some features (like auto-shutdown) don’t seem all that interesting, I am intrigued by features like the coming co-pilot mode.

In addition to the new AI features, the second-gen oven includes even tighter temperature management (powered by three internal temperature sensors and a more powerful on-board processor) and better steam management. The new oven also includes a new app and an additional recipe subscription service for $1.99 a month or $9.99 annually.

All of these new features come with a hefty price tag increase at $1199, double that of the launch price of the first-gen oven. While some may pass on the 2.0 due to the price increase, given the void left by June and the cult following Anova has, I expect the new Anova oven to sell fairly well when it ships.

The Idea of Food ‘Teleportation’ Isn’t New—But AI Is Finally Making Distributed Digital Food Replication a Reality

Over the past decade, there’s been no shortage of attempts to better understand, map, and recreate food and its properties with the click of a mouse.

From new data ontologies to a proliferation of digital noses, we’ve seen incremental steps toward an envisioned world where the fundamental building blocks of food can be better understood. However, in the past year, there has been a rapid acceleration in our collective ability to digitize various properties of food, largely driven by advances in AI.

The latest example of this comes from Osmo, which recently announced its development of the ability to digitally “teleport” a scent by using AI to digitize and re-materialize it.

Scent Teleportation

As the company’s CEO, Alex Wiltschko, explained:

“We select a scent to teleport and introduce it to a machine called the GCMS (Gas Chromatography-Mass Spectrometry). If it’s a liquid, we inject it directly; if it’s a physical sample, like a plum, we use headspace analysis, trapping the scent in the air around the object and absorbing it through a tube. The GCMS identifies the raw data, which can be interpreted as molecules, and uploads it to a cloud. There, it becomes a coordinate on our Principal Odor Map — a novel, advanced AI-driven tool that can predict what a particular combination of molecules smells like. This formula is sent to one of our Formulation Robots, which treats it as a recipe and mixes different scents to replicate our sample.”

In other words, Osmo breaks down the building blocks (molecules), creates a map, and then sends this digital map to an essence “printer” that re-creates it.

This announcement comes just weeks after leaders of NotCo’s scent and flavor AI team shared research on their new generative AI that creates scent and flavor formulations. Here’s how Aadit Patel, NotCo’s head of product, described the model:

“The system takes your prompt—such as ‘an ocean scent on a breezy summer day on a tropical island’—and creates a novel chemical formulation of that scent in one shot.” The model then generates a corresponding fragrance formula. According to Patel, the model is built on a “natural language to chemical composition” framework, tokenizing molecules to create a system capable of understanding and generating novel combinations.

With years of work focused on digitally understanding, quantifying, mapping, and reproducing scents, flavors, and food building blocks, what is allowing these latest efforts to make such significant leaps?

In a word (or two): AI. In the past, creating a facsimile of a flavor or scent took thousands of hours, relying on trained experts to work in a lab or kitchen, drawing on years of expertise. Now, AI is expediting that process with orders of magnitude more efficiency, often leaving the expert to provide a final sign-off to ensure the AI-created formula meets standards for proximity to the desired result, as well as checks for safety, cost feasibility, and more.

For the record, Osmo isn’t the first company to discuss “teleporting” a formula for digital recreation. In 2018, Japanese startup Open Meals made headlines with its “sushi teleportation” demo, essentially sending 3D printing instructions to create a sushi-like meal. We also saw Cana’s ambitious attempt to make a Star Trek replicator (though, as it turns out, investors weren’t quite ready to enter the food teleportation age).

All of this follows years of efforts to quantify and understand food digitally, including the creation of ontologies for the Internet of Food and early attempts to use AI to analyze food. But over the past couple of years, there’s no doubt that parallel advances in AI (especially in large language models) and breakthroughs in food, olfactory, and chemical science are ushering in a world where true food “teleportation”—or, more accurately, the ability to understand and synthetically recreate food, flavors, and scents—has arrived.

I’m excited to see where this all goes. To manifest the vision laid out in science fiction over the years and imaginative product concepts like that of Open Meal required a true digital understanding of the molecular building blocks of food. With AI, we are closer than ever to that understanding, and the products we’ll see built in the coming decade will not only create some mind-blowing consumer experiences but also possibly fundamentally change how food and beverage products are made and distributed.

A First Look at Roku Shoku, Sony’s Culinary Recording System to Capture and Replicate Chefs’ Recipes

This past week in Japan, Sony unveiled a project they’ve been developing in secret called Roku Shoku, a culinary recording system designed to capture exactly how a chef prepares a meal. The system also serves as a guidance tool, helping casual or inexperienced cooks create dishes with the precision of an expert with years of training.

Sony has been working on this project, which stands for “Record” (Roku) and “Cooking” (Shoku), for the past five years. Last week, the entertainment and consumer electronics giant held the first-ever press demonstration of the recording studio for The Spoon team.

“We have a recording studio here in Tokyo,” said Tomoko Nomoto, Project Leader for Roku Shoku. “We invite Michelin-starred chefs, or even grandmothers, to the studio and ask them to cook with our system. We then record their culinary data, including temperature, steam levels, and the entire cooking process.”

The project is led by a Sony R&D team out of Tokyo and is separate from research in the area of gastronomy that has taken place at Sony’s AI Research division Sony AI (the formal Gastronomy program announced in 2020 has been sunsetted, but Sony continues to work on gastronomy-related projects). Since launching the Tokyo recording studio in 2021, the team has captured thousands of recipes across a range of cuisines, including Japanese, Chinese, French, Italian, and Thai.

The Roku Shoku system features induction cooktops with temperature sensors, scales to monitor and weigh ingredients, cameras to capture a chef’s movements, and an off-the-shelf game controller made by Steam to control the setup.

Nomoto shared that users can replicate meals precisely as chefs cook them, a claim I tested myself. You can watch me trying it in the video below.

First Ever Look at Sony's Roku Shoku Culinary Recording System

According to Nomoto, the goal is to use Roku Shoku both to document recipes for restaurants and food service locations and to preserve culinary creations for future use.

“The first step will be to work with restaurants that want to share a consistent experience worldwide or recreate dishes that are no longer available, like when a chef passes away or retires,” said Nomoto.

Spoon readers might recall Cloudchef, another system that records chef creations. Nomoto explained a key difference: Sony plans for Roku Shoku to enable only human chefs to recreate these meals, while Cloudchef eventually aims to use robots for meal replication. Currently, both systems are focused solely on human use (see Spoon’s Tiffany McClurg using the Cloudchef system here).

The company has launched a website where you can find out more and request a demo.

Spoonful of AI Podcast: Raj Singh Talks GPT Food Cam and Building Websites With AI

We caught up with Raj Singh, head of Solo for Mozilla (and maker of the GPT Food Cam app) for an episode of The Spoon’s newest podcast, Spoon Full of AI.

Singh, who we wrote about last month after he launched his calorie-counting app, talked with me about the importance of understanding how AI can be used to focus on more meaningful work.

Below are some of the key themes from this week’s episode:

A key theme is the role of AI as a “co-pilot,” enhancing productivity by automating low-leverage tasks. Singh emphasizes the “95% rule,” highlighting that AI’s current strength lies in augmenting human capabilities rather than achieving perfect automation.

The Evolving Landscape of AI Integration: Singh talks about the rapid evolution of AI integration, from browser-based features like Firefox’s “Orbit” to Adobe’s Firefly, which promises AI-generated video editing enhancements.

AI as a Productivity Tool: Singh illustrates how AI tools like ChatGPT are revolutionizing workflows, citing examples such as generating SQL queries for data analysis and providing UX insights, tasks traditionally requiring specialized roles.

AI’s Disruptive Potential: The discussion touches upon AI’s potential to disrupt existing business models, pointing to his skunkworks project, GPT Food Cam, which is a free, AI-powered food logging app that rivals paid services.

Spoon Full of AI is The Spoon’s new podcast about all things AI. It will include a healthy dose of food-related AI, but it also goes beyond that.

Subscribe to Apple Podcasts, Spotify, or wherever you get your podcasts!

And oh yeah, you can listen below by hitting play!

NotCo Has Created A Generative AI for Flavor and Fragrance That Can Create Unique Formulations With Text Prompts

Food-tech company NotCo has developed a novel generative AI model, the Generative Aroma Transformer (GAT), capable of creating new flavor and fragrance formulations. The model, which the company presented on at the Food AI Summit last month, could be a potentially disruptive new tool that could impact a variety of consumer goods markets such as food, personal care, home care, and beauty industries.

The company’s Senior VP of Product, Aadit Patel, talked about the new model in a post on Linkedin today, highlighting how GAT can translate textual prompts into unique chemical formulations. “The system intakes your prompt, such as ‘an ocean scent on a breezy summer day on a tropical island’ to create a novel chemical formulation of that scent in one-shot.” From there, the model generates a corresponding fragrance formula. According to Patel, the model is built on a “natural language to chemical composition” framework, tokenizing molecules to create a system capable of understanding and generating novel combinations.

NotCo says early tests have been extremely positive. Their research indicates that GAT’s abilities rival those of human perfumers. In blind smell tests, fragrances created by GAT proved indistinguishable from those crafted by human experts. This finding signals a potentially significant shift in the industry, where only 600 certified perfumers exist globally.

How it Works

According to research presented at the Food AI Summit, the core of GAT’s functionality lies in its ability to understand and model complex interactions between volatile molecules. The model is trained on an extensive dataset of historical fragrance formulations and the molecular structures of volatile compounds. This training enables GAT to decipher the subtle relationships between different molecules and predict how they will interact to create specific aroma profiles.

The model employs a dual-system transformer network comprising an encoder and a decoder. The encoder processes the user’s prompt (with inputs such as top note (cherry candy), middle note (vanilla) and bottom note (cherry)), capturing the desired aroma profile. This is passed to the decoder, which generates a corresponding sequence of tokens representing the fragrance’s molecular structure.

GAT leverages the atomic structure of volatiles to generate novel formulations. Each molecule is represented as a graph, with atoms described by valence, degree, hydrogen count, hybridization, formal charge, and atomic number. These details are then translated into numerical representations and fed into a Graph Neural Network (GNN) model, which creates a unique vector representing each molecule. Similar vectors indicate similar molecules, allowing GAT to identify and utilize molecular structures with desired aromatic properties.

The potential impact of NotCo’s GAT is substantial. Developing new formulations for flavors and fragrances has traditionally been a time-consuming and resource-intensive process, often requiring weeks or months of expert work. If GAT can achieve the same outcome in mere seconds, it could significantly reduce flavor and fragrance development costs.

If you’d like to learn more about NotCo’s new generative AI tool for developing flavors and fragrances, NotCo’s head of machine learning, Francisco Clavero, and one of their key flavor and fragrance scientists, Cindy Sigler, will be our guests at the next Food AI Co-Lab on October 17th. You can register for this virtual event here.

When it Comes to Using AI To Shape New Culinary Creations, Ali Bouzari Thinks Food is Mostly ‘All Hands’

In the most recent episode of the Spoon Podcast, I caught up with food scientist Ali Bouzari to discuss his work and get his thoughts on new technologies that are helping to shape the future of food.

I first met Bouzari when he spoke at the Culinary Institute of America a few years ago about how robotics could impact food service and other sectors. At the time, he talked about Creator—a burger restaurant powered by robots—and suggested that food robots could sometimes do things that most food service employees could not replicate. He specifically referred to how Creator’s burger bot could create more intricate structures in the burger patty than possible to enhance mouthfeel.

When I asked him about this on the podcast, he suggested that while yes, there are things technology can do, he was worried about the recent obsession with AI and using it to craft recipes and new culinary creations. He drew a parallel between AI’s notorious difficulty in rendering realistic-looking human hands in artwork and the challenge of using AI in food production.

“You know that recurring motif where somebody will put a seemingly impressive piece of AI-generated imagery up and be like, ‘My God, look at Darth Vader doing this thing in Saturday Night Fever or something.’ And everybody always says, ‘Look at the hands, look at the fingers.’ And there’s always something wrong with the hands. There’s something that is difficult for AI to crack. What I would say it is most of food is hands. Food is basically all hands.”

Bouzari also shared how multiple clients had approached him after playing with generative AI tools to experiment with developing food products. “We have clients being like, ‘Hey, ChatGPT said we should put arrowroot flour in this cookie.’ I think that somebody is feeding all of the AI brains a lot of great information about arrowroot. Because three different people on three different projects have said that AI said, ‘Have you tried arrowroot?’ which is, in a lot of instances, kind of a useless ingredient.

But thinking about things like AI have caught his attention, Bouzari told me the biggest challenge that has his attention nowadays is the impact of climate change and how food brands are facing a reality that their products may not have a future if they continue to do things – and create food products – in the same way as they have in the past.

One example he gave is the global cacao shortage. “Chocolate is in trouble,” Bouzari said. He pointed to how disruptions in cacao production are driving up costs and threatening the availability of what is a beloved staple. This isn’t some distant, theoretical issue Bouzari told me. “It’s already happening.”

And it’s not just chocolate.

“Coffee’s next,” said Bouzari. “Coffee might do a thing where, like grapes, it just creeps higher and higher latitudes as things change.”

And because of this urgency food brands are now faced with that Bouzari gets a little annoyed with how food makers are sometimes distracted with shiny new toys while missing the big picture.

“My thinking with food is it’s a little bit extra irksome, the conversation around AI sometimes, where people say, ‘I’ve spent six months trying to get this generative AI to make me a new pasta recipe,’ when I don’t think we need that. And the water and energy cost of all of that computation is directly contributing to, I think, the actual biggest existential problem we have, which is climate change.”

We also talk about Bouzari’s experience on the Netflix Show Snack vs. Chef, his thoughts on alternative proteins and what gets him excited about the future.

You can listen to the full episode on Apple Podcasts, Spotify or by clicking play below.

I’ve Seen The Future of Food Logging Apps, And It’s GPT Food Cam

If you’ve ever tried to log your food intake with an app, you probably have realized the following:

  1. Manual food logging with an app is a pain.
  2. Food logging apps are often inaccurate because they require users to estimate portion size, ingredients, etc.
  3. Food logging apps have cumbersome onboarding processes, ask for a lot of personal info, and usually try to upsell users into premium subscriptions.

For these reasons, I have not used a food-logging app for more than a few weeks, but that may soon change with GPT Food Cam. What is GPT Food Cam? In a few words, it’s a free food-logging app that lets you snap each meal or snack with your smartphone camera and uses AI to estimate calories. The app, which can be downloaded from the iOS App Store, doesn’t ask you to take a survey or require a subscription, and, from what I can tell within a day of use, it is really pretty darn good.

That’s my description, but what does Raj Singh, the longtime entrepreneur who is the visionary behind the app, have to say about it? According to Singh, who posted recently about the app on LinkedIn, GPT Food Cam is different from other food logging apps in three primary ways:

Instant Camera Access: The app opens straight to the camera, allowing users to quickly capture their meals without navigating menus. “I wanted it to be fast and low friction,” Singh said. “In social settings, it’s less intrusive to quickly snap a photo.”

Calorie Ranges Instead of Exact Figures: Singh said that because AI has its own limitations and portion sizes vary, the app provides a calorie range. “By presenting a range, it’s mostly right,” Singh said. “The goal is to build the habit of food logging and become a more mindful eater.”

Free and Unobtrusive: Unlike many apps that require subscriptions or bombard users with ads, GPT Food Cam is entirely free and supported by occasional, non-intrusive advertisements. “Right now, ads are making four times the revenue of the AI costs,” Singh said in a phone interview with The Spoon. “This allows us to keep the app free and potentially expand its features and availability to more countries.”

After working with a food coach who encouraged him to send photos of his meals for feedback, Singh sought a convenient digital solution to continue the practice. However, he found existing apps lacking—either too complex, costly, or both.

“They were designed for the 5% who need precision, but I wanted something simple, free, and for the other 95%,” Singh said.

According to Singh, GPT Food Cam leverages Gemini Flash, a fast and cost-effective AI model, to analyze images and estimate calorie content. Users simply snap a photo of their meal, and the app processes the image to provide an approximate calorie range.

“A lot of this is prompt engineering,” Singh explained. “We use ‘chain-of-thought’ prompting, where we break down the AI’s task into specific steps. The prompt instructs the AI to look at what’s in the picture, consider each ingredient independently, estimate serving sizes based on context—like whether it’s in a bowl or on a plate—and then estimate the calories of each item before adding them up.”

Singh emphasized that while AI isn’t perfect—with about 95% accuracy—it’s sufficient for promoting mindful eating. “AI has consistently been 95% accurate,” he said. “It’s great for recommendations and suggestions, but when it comes to critical workflows, it might get things wrong 5% of the time. For food logging, where precision isn’t as critical, this level of accuracy is acceptable.”

The creation of GPT Food Cam came after a serendipitous conversation with a friend. Singh’s friend, Zvika Ashkenazi, mentioned that his son, Ben Ashkenazi, was seeking an unpaid summer internship and wondered if Singh could mentor him. Singh soon began working with Ben, and six weeks later, GPT FoodCam was born.

“Ben is graduating from ASU in Computer Science in December,” Singh said. “He taught himself React, iOS development, and more this summer with minimal help from my network. He built this end-to-end.”

While GPT Food Cam emerged just in the last couple of months after Singh’s epiphany and Ben Ashkenazi’s coding work, Singh has been toying with the idea of a low-friction app to track food intake for a decade. In 2009, he tried to develop a similar application but soon realized the technology wasn’t mature enough.

“In 2009, I tried to create this exact app,” said Singh, who is currently the head of product for Mozilla’s Solo after the browser company acquired his startup Pulse in 2022. “It wasn’t good enough, and so we pivoted into a recipe company, which became Allthecooks.” Allthecooks would go on to become the number one recipe community on Android in 2010, with 30 million users, and would later be acquired by Cookpad.

Unlike then, “the tech is now here, making GPT Food Cam a reality,” Singh said. “Advancements in AI and image recognition have finally caught up with the vision I had over a decade ago.”

With the technology to make friction-free food logging a reality, Singh told me he wants to disrupt the food logging industry by offering a free, low-friction app, but he thinks it can do so with little involvement from him going forward.

“I build some things for fun. At the onset of a new project, I’m like, ‘This is not gonna make money, but the world needs it,’ or, ‘This is gonna be my next business, and I’m leaving where I’m at.'”

Singh made it clear he is happy at Mozilla and, in fact, used the product he conceived of building for Mozilla (Solo, an AI website builder) to create the website for GPT Food Cam. From here, he will let Ashkenazi run with the product, even if he periodically suggests some ideas to make it successful.

“I think it can be very, very disruptive. People are paying $10 a month for apps they don’t need to. This app can encourage better habits without the cost and complexity.”

Singh said he is also considering expanding the app’s capabilities and reach. With the ad revenue already exceeding the AI costs by a four-to-one margin, there’s potential to increase daily usage limits (currently, users are limited to six snaps a day) and make the app available in more countries.

Selfishly, I hope he and Ashkenazi succeed because, from what I’ve seen so far, I think the app is, in fact, potentially disruptive, and I hope to keep on using it. Who knows, maybe Ashkenazi (with a little help from Singh) can put their app on a similar journey we saw with Marco Arment’s Overcast app, which originally was a passion project that emerged from Arment’s annoyance with the current state of podcast apps to become the most user-friendly podcast app (and most popular, outside of Apple’s podcast app) in the world.

IFA Smart Kitchen Roundup: Appliance Brands Try to Tap Into AI Zeitgeist With AI-Powered Food Recognition

This weekend at IFA, several big appliance brands used the show to tell the world that they are all in on AI, mainly through the integration of cameras into their ovens paired with software to enable personalized recipes and customized shopping lists.

Siemens showed off the iQ 700 oven has a built-in camera that recognizes over 80 different dishes and automatically adjusts to the ideal cooking settings. This feature allows users to place food, like a frozen pizza, in the oven and hit start for optimized cooking. The updated model offers more food recognition capabilities than previous versions and includes an optional steam function to achieve a crispy crust on baked goods.

Hisense debuted the Hi9 Series Oven, equipped with AI-powered InCamera technology for intelligent baking with over 140 pre-programmed recipes. The company also introduced a smart fridge in the Hisense Refrigerator PureFlat Smart Series, and its description sounds like they’ve been taking cues from Samsung and the Family Hub. The company described the fridge as “a home appliance control center” that “allows you to adjust temperature settings remotely through the ConnectLife app.”. The fridge also has AI-powered inventory tracking, though the company was light on details about how the tracking feature works.

Beko also let everyone know that they are trying to jam AI into as many things as possible, including their ovens. Like with HiSense and Siemens, they pointed to camera-assisted cooking in their ovens. From the release: “Beko brings AI-assisted camera technology to its Smart Home ovens, delivering a self-improving cooking experience for optimal results in the kitchen whatever the dish. With food recognition and cooking suggestions across more than 30 different food types, the new Beko Autonomous Cooking technology uses AI to finish cooking according to personalized browning levels.”

Ovens with cameras and food recognition aren’t exactly new, as we’ve been seeing this feature for the better part of a decade since June (RIP) debuted the technology. The appliance industry often displays a herd mentality, and clearly, the herd feels they’ve got to show off their AI chops, even if the technology is somewhat pedestrian at this point.

Electrolux Debuts Taste Assist AI on AEG Line

Not every new AI-feature introduction at IFA was tied to integrated cameras and image recognition. Electrolux introduced its AI Taste Assist feature on its AEG line of kitchen appliances. According to the announcement, AI Taste Assist will take recipes from the Internet, import them, and send cooking instructions to the oven, but not before it recommends ways to enhance and optimize the cook. In a demo on-stage by Electrolux at IFA, the company emphasized how the new feature was meant to overcome what they called the “cooking gap”, which they described as the limitations of existing recipes and the enhanced capabilities of modern cooking equipment. The feature that Electrolux primarily promoted to bridge this gap was steam cooking, a feature that was injected into a lasagna recipe in an on-stage demo of the Taste Assist feature by Christopher Duncan, Electrolux’s SVP of Taste for Europe.

One notable absence at Electrolux’s IFA new conference was GRO, the next-generation modular kitchen concept the company announced in June of 2022. All indications are that the Swedish appliance brand has not made any progress in commercializing GRO, probably partly due to the company’s struggles over the past couple of years. The company laid off approximately three thousand employees last year, and earlier this year, it saw the departure of its longtime CEO, Jonas Samuelson, as the company continued to struggle post-pandemic and in the fast of increased competition from Asian appliance brands.

SideChef Unveils AI Feature in App That Creates Step-by-Step Recipes From Photos of Food

SideChef recently introduced RecipeGen AI, a new beta feature that generates step-by-step recipes from a photo of any dish. Users can upload pictures of meals from restaurants or social media, and the app will provide a shoppable recipe based on the image.

From the release: “We are living in exciting times, where every inspiration can become a person’s reality,” says SideChef Founder & CEO, Kevin Yu. “At SideChef we’re excited to be the first to use AI to allow any home cook to make their food inspiration a reality for themselves and loved ones, with a single photo!

CNET writer Amanda Smith gave the app a test drive and came away with mixed feelings. While the app successfully identified many ingredients, it missed key components in some cases, such as sourdough focaccia and strawberry butter. It also occasionally added ingredients that weren’t in the dish, like bell peppers, leaving Smith feeling the accuracy was somewhat hit or miss.

Smith’s takeaway: Succes “depends on the recipe. It has a hard time with nuance and, like other AI tools, tends to make it up if it’s unsure. It’s a handy little app that could be used to inspire new ideas and ingredient concoctions or if you’re in a restaurant and don’t want to bother the waiter with dish details.”

Samsung Food Also Debuts AI-Powered Shopping Lists From Photos

SideChef isn’t the only smart kitchen company debuting photo-to-recipes/shopping lists powered by AI in their apps. At IFA last week, Samsung announced new AI-powered meal planning and food management features. The Vision AI feature now allows users to add ingredients to their Food List by simply taking a photo with their smartphone, expanding beyond the previous limitations of Samsung’s Family Hub smart fridge. This list can be used to suggest recipes, prioritize items nearing expiration, and automatically update after meals are cooked or ingredients are purchased.

Additionally, the company announced a new premium tier called Samsung Food+, a $7/month subscription service offers personalized weekly meal plans, tailored to users’ nutritional goals and dietary preferences, and tracks macronutrients and caloric intake. This premium tier also integrates more advanced AI functionality, allowing users to customize recipes and receive a full week of meal recommendations, helping reduce food waste and simplify grocery shopping by making the app a central hub for food management and meal preparation.

From Data-Scraping to Discernment Layer: How NotCo’s Giuseppe AI Has Evolved Over the Past Decade

Almost a decade ago, while others experimenting with AI focused on algorithms for trading, diagnostics, or digital advertising, a company called NotCo was experimenting with AI by the name of Giuseppe to create plant-based foods that could match the taste and texture of their animal-based counterparts.

According to Aadit Patel, SVP of AI Product and Engineering at NotCo, the company’s founders (Patel would join a couple of years after the company was founded in 2015) realized early on that, in order to build an AI model that could help create plant-based products mimicking the taste, texture, and functionality of their animal-based counterparts, they would need a whole lot of data.

The problem was, as a startup, they didn’t have any.

When I asked Patel in a recent interview how the company overcame the infamous “cold start” problem—the challenge many embryonic AI models face before they have built large datasets on which to train—he told me they found the solution in a very public place: the U.S. government’s website.

“In the early days, when we had no money, we literally scraped the USDA website,” said Patel. “If you go to the USDA website, there’s a bunch of free data materials for you to use. And I guess no one had actually joined it together to create a comprehensive dataset… So the first versions of Giuseppe were built on that.”

This cobbled-together dataset formed the foundation for Giuseppe’s recommendations, leading to the creation of products like NotMilk, which uses unexpected combinations like pineapple and cabbage to replicate the taste of dairy milk.

As NotCo grew, so did Giuseppe’s capabilities. New analytical labs in San Francisco and Santiago, Chile, gave the company a wealth of new data on which to train its AI. Over time, the model’s ability to create innovative food products also improved.

One of the biggest hurdles in food development is the fragmented nature of the supply chain. Data is scattered across various entities—ingredient suppliers, flavor houses, manufacturers, and research institutions—each holding critical information that contributes to the success of a product. Over time, the company realized that to create an AI capable of building innovative products, it couldn’t rely solely on NotCo’s datasets. Instead, Giuseppe would need to integrate and analyze data from across this complex web of partners.

“What we’ve done with Giuseppe is figure out a way to incentivize this very fragmented ecosystem,” Patel said.

According to Patel, pulling together these disparate datasets from across the product development and supply chain would result in a more holistic understanding of what is needed for a successful product that is better aligned with market realities.

“We realized that if we just made an AI system that’s specific to CPG, we’d be losing out,” said Patel.

Generative AI and Flavor and Fragrance Development

One recent expansion of Giuseppe’s capabilities has been the exploration of new flavors and fragrances using generative AI. While GenAI models like ChatGPT have become infamous for creating sometimes strange and off-putting combinations when designing recipes and new food product formulations, Patel explained that the company has been able to overcome issues with general LLMs by creating what he calls a discernment layer. This layer filters and evaluates the multitude of generated possibilities, narrowing them down to the most promising candidates.

“Discernment is key because it’s not just about generating ideas; it’s about identifying the ones that are likely to succeed in the real world,” Patel said. “With generative AI, you can prompt it however you want and get an infinite amount of answers. The question is, how do we discern which of these 10,000 ideas are the ones most likely to work in a lab setting, a pilot setting, or beyond?”

The discernment layer works by incorporating additional data points and contextual knowledge into the model. For instance, it might consider a formulation’s scalability, cost-effectiveness, or alignment with consumer preferences. This layer also allows human experts to provide feedback and fine-tune the AI’s outputs, creating a process that combines AI’s creativity with the expertise of flavor and fragrance professionals.

Early tests have shown positive results. When tasked with creating a new flavor, both the AI and the human perfumers receive the same brief. When the results are compared in A/B tests, Patel says the outputs of Giuseppe’s generative AI were indistinguishable from those created by human experts.

“What we’ve built is a system where AI and human expertise complement each other,” said Patel. “This gives us the flexibility to create products that are not just theoretically possible but also market-ready.”

CPG Brands Still Have a Long Way to Go With AI-Enhanced Food Creation

Nearly a decade after building an AI model with scraped data from the USDA website, NotCo has evolved its AI to create new products through a collaborative approach that results in a modern generative AI model incorporating inputs from its partners up and down the food value chain. This collaborative approach is being used for internal product development and third-party CPG partners, many of whom Patel said approached the company after they announced their joint venture with Kraft Heinz.

“Ever since our announcement with Kraft Heinz and signing a joint venture, there’s been a lot of inbound interest from a lot of other large CPGs asking ‘What can you do for us?’ and ‘What is Giuseppe?’ They want to see it.”

When I told Patel I thought that big CPG brands have come a long way over the past twelve months in their embrace and planning for using AI, he slightly disagreed. He said that while there’s a lot of interest, most big brands haven’t actually transformed their business to fully create products with the help of AI.

“I would say there’s strong intent to adopt it, but I think there hasn’t been put forth like a concrete action plan to actually develop the first AI-enabled R&D workforce,” said Patel. “There is room, I think, for new AI tech for formulators, and room for best practices and lessons learned of adopting AI.”

You can watch my full interview with Aadit below.

The NotCo team will be at the Food AI Summit talking about their new efforts using generative AI to develop flavor and fragrance, so make sure to get your tickets here.

NotCo's Aadit Patel Talks About the Evolution The Company's Food AI Giuseppe

The Idea of Smell-O-Vision Has Been Around for Over a Century. AI May Finally Make It Work

Since the early 1900s, the entertainment industry has been attempting to pair the experience of smell with video entertainment.

In 1916, the Rivoli Theater in New York City introduced scents into the theater during a movie called The Story of Flowers. In 1933, the Rialto Theater installed an in-theater smell system. Hans Laube developed a technique called Scentovision, which was introduced at the 1939 World’s Fair. A decade ago, Japanese researchers were also exploring “Smell-O-Vision” for home TVs, working on a television that used vaporizing gel pellets and emitted air streams from each corner of the screen into the living room.

However, none of these efforts took off, primarily because they didn’t work very well. These attempts at Smell-O-Vision failed because we’ve never been able to adequately recreate the world’s smells in an accurate or scalable way, largely because we’ve never been able to digitally capture them.

This doesn’t mean the fragrance and scent industry hasn’t been robust and growing, but it’s a very different task to create a singular fragrance for a consumer product than to develop something akin to a “smell printer” that emits scents on command. The latter requires a comprehensive digital understanding of scent molecules, something that has only recently become possible.

The digital understanding of the world of smells has accelerated in recent years, and one company leading the way is Osmo, a startup that has raised $60 million in funding. Osmo is led by Alex Wiltschko, a Harvard-trained, ex-Googler who received his PhD in olfactory neuroscience from Harvard in 2016. Wiltschko, who led a group at Google that spent five years using machine learning to predict how different molecules will smell, founded Osmo in early 2023 with the mission of “digitizing smell to improve the health and well-being of human life” by “building the foundational capabilities to enable computers to do everything our noses can do.”

Osmo employed AI to explore the connection between molecular structure and the perception of smell, demonstrating that a machine can predict scents with remarkable accuracy. They developed a machine-learning model using graph neural networks (GNNs), trained on a dataset of 5,000 known compounds, each labeled with descriptive smells like “fruity” or “floral.” This model was then tested on 400 novel compounds, selected to be structurally distinct from anything previously studied or used in the fragrance industry, to see how well it could predict their scents compared to human panelists.

The model’s capabilities were further challenged in an “adversarial” test, where it had to predict scents for molecules that were structurally similar but smelled different. Osmo’s model correctly predicted scents 50% of the time in this difficult scenario. Additionally, the model was able to generalize well beyond the original training data, assessing other olfactory properties like odor strength across a massive dataset of 500,000 potential scent molecules.

The Principal Odor Map (POM) created by Osmo’s model outperformed human panelists in predicting the consensus scent of molecules, marking a significant advancement in olfactory science and demonstrating that AI can predict smells based on molecular structure better than individual human experts in many cases.

We’ve been able to digitally capture and categorize other sensory categories, such as vision, which has led to massive new industry value creation in robotics and autonomous vehicles. The biggest leaps have been a result of machine learning models, and now we’re seeing another massive leap forward in capabilities and product innovation through the application of generative AI.

One potential application Wiltschko describes is “teleporting scent,” where we’ll be able to capture a smell from one part of the world and digitally transfer it to another. To do this, he envisions a world where a local AI-guided molecular sensor could instantly identify the molecular makeup of any scent. From there, his odor map can create what is essentially a formula ready for teleportation without significant manual intervention by scent experts.

This idea, using AI to recreate scents based on a digital framework quickly, could lay the foundation for what film and TV makers have long dreamed of: creating technology that can recreate odors and smells at scale. In other words, we may finally enter a world where Smell-O-Vision becomes a reality. The potential for video entertainment, virtual reality, and other experiences in food service, travel, and more would no doubt lead to a multitude of new applications, much like we’ve seen over the past couple of decades with advances in computer and machine vision.

The Food AI Weekly Bulletin: Will AI Fakery Make Restaurant Reviews a Thing of the Past?

Welcome to the Food AI Weekly Bulletin, our new weekly wrapup that highlights important happenings at the intersection of AI and food.

Nowadays, there’s a lot of noise around how AI is changing food, so we decided to create a weekly brief to bring you what’s important, decipher through all the noise, and deliver actionable insights. If you’d like to sign up for our weekly Food AI Weekly, you can do so here.

Highlights

Is AI Ruining Restaurant Reviews? A new study shows people cannot distinguish between real and AI-generated reviews.

AI Food Art Is Everywhere (And It’s Not Great for Freelancers) Generative AI tools like Midjourney and DALL-E are revolutionizing food imagery, but what does this mean for freelancers and creatives who traditionally provided these services?

First, Al Michaels. Next, How About an AI-powered Anthony Bourdain? The news of Al Michaels allowing AI to replicate his voice has almost everyone freaking out, but what does it mean for the future of AI-generated avatars of famous food personalities?

Swallowing A Robot. Endiatx has developed the Pillbot, a tiny robot that can be swallowed to explore the gastrointestinal tract, potentially revolutionizing diagnostics and personalized nutrition.

Food & Nutrition Centric LLMs Could Be an Investible Opportunity. VCs see potential in industry-specific AI models, particularly in the domains of biology, chemistry, and materials, as these specialized LLMs could offer unique investment opportunities.

Brightseed’s Forager AI Finds Novel Bioactives. Cranberry giant Ocean Spray teams up with Brightseed to uncover new bioactive compounds in cranberries.

Our Favorite AI Food Art of the Week. We’ll be making this a regular feature. If you’d like your art featured, submit it on our Spoon slack. 

We’re going to be exploring all of this at our Food AI Summit in September. Join us won’t you? Super Early Bird pricing expires at the end of this month.

Is AI Ruining Restaurant Reviews?

What happens when humans can’t tell real restaurant reviews from fake ones? The restaurant industry has begun asking itself this question as a tidal wave of fake AI reviews floods online sites.

According to Yale professor Balazs Kovacs, humans are already losing their ability to discern the real from the fake. Kovacs recently unveiled the results of a study demonstrating AI’s ability to mimic human-written restaurant reviews convincingly. For his test, Kovacs fed Yelp reviews into GPT-4 and then asked a panel of human test subjects whether they could tell the difference. At first, the results generated by GPT-4 were too perfect, so Kovacs then prompted GPT-4 to insert “

While this raises obvious concerns about the authenticity of online reviews and the trustworthiness of consumer-generated content, it shouldn’t be surprising. Figure 01’s human-like speech tics were creepy, but mostly because of how human its awkward conversation seemed. With typos and sub-par grammar—in other words, what we see every day on social media—it makes sense that AI-generated reviews seemed more human.

One potential workaround to this problem of AI-generated reviews is using AI to detect and notify us what fake content is, but early tests show that even AI can’t tell what is real and what is fake. Another suggestion is to require reviewers to have purchased a product to review it (similar to having Amazon labels whose reviews are from verified purchasers) and apply it to restaurants. My guess is that this will be the best (and potentially last) line of defense against the coming tidal wave of AI reviews.

AI Food Art Is Everywhere (And It’s Not Great for Freelancers)

One early application of generative AI, as it applies to food, is the creation of images. Midjourney, DALL-E, and other tools allow us to create instant realistic images with a few sentences. As a result, we’ve seen CPGs, food tech software companies, and restaurant tech startups jump on the generative art trend.

While that isn’t necessarily good news for actual artists (this WSJ article is a must-read about the impact of AI on freelancers and creatives), these tools have democratized professional-ish like photos and art for folks in the same way Canva made professional-style graphics and presentations available to anyone.

One company that’s benefitted significantly is Innit. The company, which focused in its early life on hiring celebrity chefs like Tyler Florence and spending tens of thousands on photo shoots for a recipe, is now whipping them instantly with generative AI for its Innit FoodLM.

While most Internet-savvy marketing types at food brands, restaurants, and other food-related businesses have at least learned to dabble in generative AI prompt engineering, that hasn’t stopped some from trying to create a business out of it. Lunchbox created an AI food image generator utilizing DALL-E as the underlying LLM over a year ago (the website has since gone dark), and just this week I got pitched on a new AI-powered food generator that wants to charge for its service (which is essentially a user interface to manage prompt engineering for an underlying LLM (which most likely is Midjourney or GPT-4). There’s likely a small lifespan for these types of services, but my guess is most marketing folks will learn to prompt engineers directly with popular image generators like Midjourney.

First, Al Michaels. Next, an AI-Powered Anthony Bourdain?

The Internet freaked out yesterday when news broke that Al Michaels has agreed to let an AI copy his voice, and rightly so. First off, it’s creepy. Second, this is the exact thing was the main reason the Hollywood writers and actors guilds kept striking for so long, so I’m guessing the Hollywood creative community isn’t exactly happy with Al. And finally, it goes to show you that if you throw enough money at us humans, the temptation to cave to the bots will be too much.

My guess is we’ll eventually see AI-generated avatars of famous chefs. All it would take is for the estate of Julia Child or Anthony Bourdain to get a good enough offer and it won’t be long before we hear (and maybe see) their avatars.

Swallowing A Robot

According to Venturebeat, Endiatx has developed a microscopic robot that can traverse your body and is equipped with cameras, sensors, and wireless communication capabilities. The robot, called Pillbot, allowing doctors to examine the gastrointestinal tract and be used both for diagnostic and therapeutic purposes.

The company’s CEO, Torrey Smith, has taken 43 of these Pillbots and swallowed one live on stage, which can be seen here. If this technology actually works (and those pills can be made smaller because, holy cow, that’s a literal big pill to swallow), it’s not hard to imagine these being used to dial in and optimize personalized nutrition regimens.

Food & Nutrition Centric LLMs Could Be an Investible Opportunity

Business Insider asked some VCs what they’re bored by when it comes to AI and what they’re excited about. Not surprisingly, they talked alot about how it will be hard for startups to break through in foundational large language models, where big players like Open AI and Google play. And like any good VC looking at an early market they talked up up picks and shovels

Even as investors shift their focus to promising AI infrastructure startups, there may still be some opportunities for new LLM startups to win, especially when they’re trained for specific industries, explained Kahini Shah, a principal at Obvious Ventures.

“We’re excited about what we call Generative Science at Obvious, i.e, large, multi-modal models trained in domains such as biology, chemistry, materials,” she said.

Brightseed’s Forager AI Finds Novel Bioactives

Brightseed, a company that uses AI to accelerate bioactive and food compound discovery, announced that it has (in partnership with Ocean Spray) used its Forager AI to uncover novel bioactive compounds in cranberries. Forager identified multiple bioactives, such as terpenes, which Brightseed believes hold significant potential for human health. These findings, based on in silico analyses, will undergo further clinical validation and will be presented at the American Society of Nutrition’s NUTRITION 2024 conference.

This acceleration effect of new health-positive compounds is another example of the AI acceleration effect I wrote about yesterday. Things are beginning to move exponentially faster at every stage of the food value chain, which over time means our basic understanding of the rules underpinning what we do (such as food product development) gives way to entirely new rules that are rewritten in large part by AI.

Our Favorite AI Food Image of the Week: Hungry Monkey

We like looking at AI-generated food art and figured we’d show you some of our favorites on a weekly basis. 

If you’d like to submit your AI-created food art (or if you’ve found one you think we should feature, drop the image and the source/attribution (preferably a link) on our Spoon Slack.