How AI can save bananas... and deliver our shopping
AI is helping farmers spot disease and food manufacturers create new recipes, as well as streamlining shopping and making our food tastier
Wilted leaves, black discolouration, weak roots and tiny bugs all subtle yet important signs that signal apocalypse for bananas. If diseases spread, we risk losing one of our favourite fruits. There are many potential solutions, but one tool could see smartphones using artificial intelligence (AI) to analyse photos of banana plants all snapped by farmers in their groves, helping spot infestations and infections earlier and more easily.
It's one way AI is slipping into our diets. And there's plenty more, from boosting the flavour of basil to machine-generated recipes. Potential improvements aren't only about tastier meals, but also addressing challenges such as sustainability and distribution, says Max Elder, research director for the Food Futures Lab at the Institute for the Future.
"Technologists, food companies, farmers and more are embedding artificial intelligence broadly defined into every corner of our food system from production to formulation to distribution and even consumption," he said. "AI is being used in really creative ways to design a future food system we want: one that is more healthy, humane and sustainable."
He added: "Our global agricultural system has gone through three major agricultural revolutions: the first involved muscles, the second machines and the third biology. We're on the cusp of a fourth agricultural revolution that involves intelligence." Here's what that looks like now and how you can use AI in your kitchen, for shopping and to get better food.
Bananas are a perfect fruit: they're tasty and healthy, delightful for breakfast or dessert, beloved by athletes and children alike. Imagine, if you can, a world without these yellow-skinned marvels.
We could face it. Thanks to Fusarium wilt, also known as Panama disease, bananas are at risk of being wiped out for good. Among the many solutions being considered to help spot and stop the spread of that disease and others is an AI-based app built by researchers at the International Center for Tropical Agriculture.
Designed to be used by farmers, the Tumaini app was tested at holdings in the Congo, India, Colombia and more, with a 90% success rate in spotting five serious diseases and one pest faced by bananas. Farmers who spot signs of distress on a banana take a photo and upload it into the app, which assesses in real-time using AI-based image processing whether the fruit is at risk from a pest or disease. It's a sophisticated tool, looking at the leaf, bunches of fruit or even the entire plant. "There is very little data on banana pests and diseases for low-income countries, but an AI tool such as this one offers an opportunity to improve crop surveillance, fast-track control and mitigation efforts, and help farmers to prevent production losses," said Michael Selvaraj, the study's lead author, in a statement released alongside the research.
The photos are GPS tagged as they are collected, which means the project can track outbreaks of disease even monitoring a crop's status via satellite imagery or drones. "This is not just an app," said Selvaraj. "But a tool that contributes to an early warning system that supports farmers directly, enabling better crop protection and development and decision making to address food security."
However, there's more to food production than avoiding disease: there's also taste. MIT researchers used AI to tweak basil, tracking the growing conditions and altering them to improve the flavour. They found that exposing the basil to light 24 hours a day led to tastier plants. And researchers at Wageningen University in the Netherlands have teamed up with Chinese tech firm Tencent for an "Autonomous Greenhouse Challenge," where participants compete to create a tastier cherry tomato crop within six months using remote controlled technologies, sensors and machine learning.
"One of the bottlenecks with innovation in our food system lies in our fields," said Elder. "Plants take a long time to grow. Sunlight, water and a lot of patience are all required. So agricultural innovation in terms of production is hindered by the natural limit of biological growth. Various forms of AI can dramatically speed up that process. AI-empowered indoor agriculture can speed up harvest times so that R&D that once took a generation of farmers now takes a few months."
Of course, farmers have long selected for specific features in crops. According to the United States Department of Agriculture, 90% of corn and soya grown in the US is genetically modified. That's helpful as it selects for useful traits -- such as resilience -- but plenty of people aren't fans of eating genetically modified organisms (GMOs). AI may have a solution. "AI is enabling a new form of crop production called computational breeding which uses predictive analytics and novel data sets to empower farmers to plant non-modified seeds that mimic many of the benefits from their modified counterparts," said Elder. He points to Hi Fidelity Genetics (HFG), which just raised $8.5 million in funding to create non-genetically modified crops through computational breeding.
AI could help make meat production more sustainable, too. "For a more humane food system, companies like Cainthus are using computer vision and AI to design digital dairies that monitor individual needs and comforts of cows to optimise for wellbeing," Elder said. "When it comes to environmental monitoring and enforcement, researchers at Stanford University have been using machine learning to teach computers to track down the number and location of all concentrated animal feeding operations, so-called CAFOs, in the US."
Building better food
Designer food could stretch further. Vegan alternatives to meat are increasingly popular, with startups such as Impossible and Beyond Meat producing animal-free burgers that some believe taste as good as if not better than the versions that require dead cows. Elder suggests that AI could help with the shift to more sustainable plant-based food, without giving up taste and texture, pointing to Chilean startup The Not Company. "They have created an AI food scientist that uses machine learning to train itself to reformulate animal-based food products from the molecular level using novel combinations of plant-based ingredients," he said.
Of course, AI doesn't have the ability to taste all that we can do is train it to know what we like. Instead, The Not Company is taking what Elder calls the "centaur approach", which has humans and machines working together, outperforming what either is capable of on their own. "Right now, machine learning still needs human taste buds to train its algorithms," Elder said. "The Institute for the Future has many forecasts around the future of human-machine symbiosis, and I think the food space is ripe for such collaboration."
That's exactly the approach taken by IBM and McCormick. The spices and seasoning maker is always working on new ways to create ready meals and unique new flavour combinations, work that's currently done by hundreds of food developers. McCormick has decades of food development data and was seeking a way to make better use of it.
Hamed Faridi, the chief science officer at McCormick, was listening to the radio when he heard a report about IBM's Chef Watson, which used the tech giant's Watson AI platform to come up with recipe ideas. "When he heard that interview, he pulled his car over because he'd finally found someone who might be able to help," said Robin Lougee, consumer and agriculture research lead at IBM Research.
Watson turned out to not be the right solution -- the version discussed in the radio programme was designed for chefs creating recipes, not food manufacturers. Plus, it depends partially on an idea known as the "flavour pairing hypothesis", which suggests that compounds with the same base chemicals will taste good together, but it only holds true in certain cultures. McCormick needs its recipes to work globally. Instead, McCormick used IBM Research's AI for Product Composition platform as the basis for the system; it's also been used to design fragrances and concrete, two very different products with recipes at their core.
The more than 500 product developers at McCormick start with a brief: for example, create Tuscan flavours to be used on a protein or vegetables. "With flavours, you have a goal, but flavours don't have a perfect mathematical model to describe it," said Lougee. "So there's a creative aspect." They'll come up with an idea for what it should taste like and start working through compounds to create that taste. "They'll bake up a batch and have product developers taste it and give feedback," she said. That feedback loop runs over and over again, until the developer gets it right.
"They're really searching the space of all possible raw material ingredient combinations, and all the different proportions, to try to find that great candidate," she said. And McCormick has tens of thousands of potential ingredients to choose from. The taste of garlic differs, for example, depending on the variety and where it's grown, but also its age and the size of the granules. Plus, product developers need to consider shelf life, compliance, and how easy it is to manufacture. That's the problem, but McCormick isn't answering it by trying to wholly replace those human developers. "We're really an aid, like a digital system or an apprentice to try to help the human do a better job," she said, "to explore faster and further and spark their creativity."
To be clear, IBM hasn't modelled human taste. Instead, the AI platform is used to look for patterns in the "treasure trove of data" that McCormick has accumulated over many decades of food production, not only including the formulas that worked but also those that didn't. "It generates candidate formulas that it then evaluates," Lougee explained. "It serves up... a number of suggestions." The human developer then picks and chooses what they'd like to use.
So far, McCormick has released three different products developed using this system: flavour packs for Tuscan chicken, Bourbon pork tenderloin and New Orleans sausage. But Lougee's favourite example of an unexpected flavour combination was pizza seasoning. "With pizza, you think basil or oregano," she said. "But the system suggested cumin. That's something that would be atypical to me, but you can imagine it would have a certain warmth. The product developer tried it and liked it."
Things can get weirder. Researchers at MIT trained a recurrent neural network on pizza recipes found on food blogs, letting it come up with its own flavours anyone for sweet potatoes, beans and Brie? The team brought a chef in to massage the recipes into an edible shape, and he reportedly added the shrimp, jam and chorizo combination to the menu at his own pizzeria.
Humans and AI working together on food is a bit of a theme. "In the midst of 'AI awakening' where machines are becoming good at many 'human' jobs, people are worried that AI will ultimately lead to mass unemployment by replacing them," the MIT researchers said in a post when they launched their project. "On the contrary, we believe that we can achieve the most creative and productive outcomes when humans and machines work together to enhance each other's complementary strengths and skills."
Get in the kitchen
Most of the time, cooking in our kitchens isn't about reinventing the pizza, but figuring out a decent dinner out of the contents of our cupboards and there's an app for that. Chefling is one of several artificial intelligence-based tools designed to scan our existing ingredients and suggest a meal, identifying substitute ingredients if you're short of something or have dietary preferences or requirements.
"Our machine learning has been based on a lot of natural-language processing techniques so that we understand the recipe better, which is the foundation of creating machine readable recipes," said co-founder and CEO Jeff Xie. That means Chefling can send instructions directly to smart appliances, but also helps inform ingredient matching and substitution and personalisation. "We are also testing some image recognition technology so that the inventory can be done in a more automatic way," he added. That should be available early in 2020.
While it's handy to have a digital inventory of your food and instant suggestions on recipes and substitute ingredients, Xie sees this going much further. "AI will be everyone's personal butler, or nutrition consultant to help people achieve their dietary goal at almost no cost," he predicted. "Significantly lower food wastage will be achieved because AI can help plan the consumption and purchase much better." And if cooking still sounds like too much effort, he says that integration with smart appliances will mean artificial intelligence can decide a cooking programme, meaning that people need only shuffle food from one appliance to another and press a start or stop button. That may not feel like cooking to some of us, but for many people it will mean avoiding burnt or tasteless dinners.
Supermarkets and distribution
Of course, we still need to get the food to our house. British tech company Ocado is perhaps best known as the delivery service behind Waitrose, but its efforts go far beyond bringing avocados and sourdough to your home. Ocado's Smart Platform is an end-to-end virtual logistics system for retailers, including a front end for consumers to shop all the way through to warehouse, supply chain and delivery. "AI is used in all of the parts of that system," explained Alex Harvey, general manager for warehouse automation at Ocado Technology.
For those shopping for groceries, that means AI-based product recommendations and recipe suggestions, but that could go further in the future. He foresees a time when customers can "take a quick picture of their shelves at the beginning of the week and at the end, and say please order anything that's missing. People talk about smart fridges, but that seems like a hard thing if anyone looks at mine, it'll be random bowls covered in plates for leftovers."
Ocado also famously has smart, robot-filled warehouses. Harvey says they already make use of AI for demand forecasting for stocking and purchasing, "We have very good models that allow us to have very low waste and very high fulfilment," he claimed. Once inside the warehouse, storage locations for various products constantly change depending on demand and frequency of access to make picking for orders as efficient as possible. Products are still picked by hand Ocado is working on developing bots to do the job, but it's not simple to do with workers packing the order into bots that Harvey describes as looking like washing machines on wheels. Those aren't independently automated, but "orchestrated" by software that uses AI to pick routes; kind of like an automated air traffic controller, but its workspace is a digital twin of the warehouse rather than the open skies. Meanwhile, the real warehouse is overseen by CCTV that's analysed by machine learning.
Linking the warehouse and our kitchens is the delivery service. "We have really powerful routing and forecasting algorithms, which are dynamically optimising the routing," Harvey said. Every item you place in your basket on the Ocado site sparks a recalculation: the system considers which van is best suited in terms of weight and location, and which route is most efficient. "We do a hugely complicated set of optimisations for all customers, as everyone is adding to their baskets we're constantly rejigging," Harvey continued. "We make something like three million computing calculations per second."
For deliveries in the future, Ocado is eyeing driverless cars but only if someone else manages to build them because Ocado isn't planning on creating its own. However, that raises another problem: getting groceries from that autonomous vehicle outside your house and into your kitchen. Some delivery services may require you to fetch them yourself, but Harvey says it's a core proposition for Ocado to deliver directly to your kitchen. "We'd need a robot to be able to deal with the infinite uncertainty of everybody's unique setup," he said. "Everybody's house is different."
Another challenge is to automate order picking, which requires a dextrous robot that can manipulate a 54,000-strong range of products, from large items to small and delicate to perishable imagine the AI required to make a bot pick up an egg. Achieving that is some time off, Harvey notes, but if Ocado can master automated order picking, he believes the technology can have wide-reaching benefits. "That's the kind of manipulation one would need for assisted living," he said.
That's some way off; it's no coincidence that the AI that supports humans' work is already in use, but that which aims to fully replace people remains an idea for the future. The use of AI in the food industry highlights that the technology isn't a replacement for farmers, chefs, food developers, or order pickers, but something to make their lives easier and ensure our fruit bowls stay full of bananas.
The ultimate law enforcement agency guide to going mobile
Best practices for implementing a mobile device programFree download
The business value of Red Hat OpenShift
Platform cost savings, ROI, and the challenges and opportunities of Red Hat OpenShiftFree download
Managing security and risk across the IT supply chain: A practical approach
Best practices for IT supply chain securityFree download
Digital remote monitoring and dispatch services’ impact on edge computing and data centres
Seven trends redefining remote monitoring and field service dispatch service requirementsFree download