Giving robots self-awareness as they move through space - and maybe even providing them with gene-like methods for storing rules of behavior - could be important steps toward creating more intelligent machines.
Now roboticists are going a step further, training their creations to do even better: understand their own image in space and interact with humans like humans do. Today, there are already robot-teachers like KeeKo, robot-pets like Moffin, robot-babysitters like iPal, and robotic companions for the elderly like Pepper.
But even these reasonably intelligent creations still have huge limitations, some scientists think. “There are niche applications for the current generations of robots,” says professor Anthony Zador at Cold Spring Harbor Laboratory—but they are not “generalists” who can do varied tasks all on their own, as they mostly lack the abilities to improvise, make decisions based on a multitude of facts or emotions, and adjust to rapidly changing circumstances. “We don’t have general purpose robots that can interact with the world. We’re ages away from that.”
Robotic spatial self-awareness – the achievement by the team at Columbia – is an important step toward creating more intelligent machines. Hod Lipson, professor of mechanical engineering who runs the Columbia lab, says that future robots will need this ability to assist humans better. Knowing how you look and where in space your parts are, decreases the need for human oversight. It also helps the robot to detect and compensate for damage and keep up with its own wear-and-tear. And it allows robots to realize when something is wrong with them or their parts. “We want our robots to learn and continue to grow their minds and bodies on their own,” Chen says. That’s what Zador wants too—and on a much grander level. “I want a robot who can drive my car, take my dog for a walk and have a conversation with me.”
Columbia scientists have trained a robot to become aware of its own "body," so it can map the right path to touch a ball without running into an obstacle, in this case a square.
Jane Nisselson and Yinuo Qin/ Columbia Engineering
Today’s technological advances are making some of these leaps of progress possible. One of them is the so-called Deep Learning—a method that trains artificial intelligence systems to learn and use information similar to how humans do it. Described as a machine learning method based on neural network architectures with multiple layers of processing units, Deep Learning has been used to successfully teach machines to recognize images, understand speech and even write text.
Trained by Google, one of these language machine learning geniuses, BERT, can finish sentences. Another one called GPT3, designed by San Francisco-based company OpenAI, can write little stories. Yet, both of them still make funny mistakes in their linguistic exercises that even a child wouldn’t. According to a paper published by Stanford’s Center for Research on Foundational Models, BERT seems to not understand the word “not.” When asked to fill in the word after “A robin is a __” it correctly answers “bird.” But try inserting the word “not” into that sentence (“A robin is not a __”) and BERT still completes it the same way. Similarly, in one of its stories, GPT3 wrote that if you mix a spoonful of grape juice into your cranberry juice and drink the concoction, you die. It seems that robots, and artificial intelligence systems in general, are still missing some rudimentary facts of life that humans and animals grasp naturally and effortlessly.
How does one give robots a genome? Zador has an idea. We can’t really equip machines with real biological nucleotide-based genes, but we can mimic the neuronal blueprint those genes create.
It's not exactly the robots’ fault. Compared to humans, and all other organisms that have been around for thousands or millions of years, robots are very new. They are missing out on eons of evolutionary data-building. Animals and humans are born with the ability to do certain things because they are pre-wired in them. Flies know how to fly, fish knows how to swim, cats know how to meow, and babies know how to cry. Yet, flies don’t really learn to fly, fish doesn’t learn to swim, cats don’t learn to meow, and babies don’t learn to cry—they are born able to execute such behaviors because they’re preprogrammed to do so. All that happens thanks to the millions of years of evolutions wired into their respective genomes, which give rise to the brain’s neural networks responsible for these behaviors. Robots are the newbies, missing out on that trove of information, Zador argues.
A neuroscience professor who studies how brain circuitry generates various behaviors, Zador has a different approach to developing the robotic mind. Until their creators figure out a way to imbue the bots with that information, robots will remain quite limited in their abilities. Each model will only be able to do certain things it was programmed to do, but it will never go above and beyond its original code. So Zador argues that we have to start giving robots a genome.
How does one do that? Zador has an idea. We can’t really equip machines with real biological nucleotide-based genes, but we can mimic the neuronal blueprint those genes create. Genomes lay out rules for brain development. Specifically, the genome encodes blueprints for wiring up our nervous system—the details of which neurons are connected, the strength of those connections and other specs that will later hold the information learned throughout life. “Our genomes serve as blueprints for building our nervous system and these blueprints give rise to a human brain, which contains about 100 billion neurons,” Zador says.
If you think what a genome is, he explains, it is essentially a very compact and compressed form of information storage. Conceptually, genomes are similar to CliffsNotes and other study guides. When students read these short summaries, they know about what happened in a book, without actually reading that book. And that’s how we should be designing the next generation of robots if we ever want them to act like humans, Zador says. “We should give them a set of behavioral CliffsNotes, which they can then unwrap into brain-like structures.” Robots that have such brain-like structures will acquire a set of basic rules to generate basic behaviors and use them to learn more complex ones.
Currently Zador is in the process of developing algorithms that function like simple rules that generate such behaviors. “My algorithms would write these CliffsNotes, outlining how to solve a particular problem,” he explains. “And then, the neural networks will use these CliffsNotes to figure out which ones are useful and use them in their behaviors.” That’s how all living beings operate. They use the pre-programmed info from their genetics to adapt to their changing environments and learn what’s necessary to survive and thrive in these settings.
For example, a robot’s neural network could draw from CliffsNotes with “genetic” instructions for how to be aware of its own body or learn to adjust its movements. And other, different sets of CliffsNotes may imbue it with the basics of physical safety or the fundamentals of speech.
At the moment, Zador is working on algorithms that are trying to mimic neuronal blueprints for very simple organisms—such as earthworms, which have only 302 neurons and about 7000 synapses compared to the millions we have. That’s how evolution worked, too—expanding the brains from simple creatures to more complex to the Homo Sapiens. But if it took millions of years to arrive at modern humans, how long would it take scientists to forge a robot with human intelligence? That’s a billion-dollar question. Yet, Zador is optimistic. “My hypotheses is that if you can build simple organisms that can interact with the world, then the higher level functions will not be nearly as challenging as they currently are.”
Artificial Intelligence is already helping to grow some of the food we eat.
The chef-farmers choose from among 45 types of herb and leafy-greens seeds, plop them into grow trays, and a few weeks later they pick and serve. While success is predicated on at least a small amount of these humans’ care, the systems are autonomously surveilled round-the-clock from Babylon’s base of operations. And artificial intelligence is helping to run the show.
Babylon piloted the use of specialized cameras that take pictures in different spectrums to gather some less-obvious visual data about plants’ wellbeing and alert people if something seems off.
Imagine consistently perfect greens and tomatoes and strawberries, grown hyper-locally, using less water, without chemicals or environmental contaminants. This is the hefty promise of controlled environment agriculture (CEA) — basically, indoor farms that can be hydroponic, aeroponic (plant roots are suspended and fed through misting), or aquaponic (where fish play a role in fertilizing vegetables). But whether they grow 4,160 leafy-green servings per year, like one Babylon farm, or millions of servings, like some of the large, centralized facilities starting to supply supermarkets across the U.S., they seek to minimize failure as much as possible.
Babylon’s soilless hydroponic growing system
Courtesy Babylon Micro-Farms
Here, AI is starting to play a pivotal role. CEA growers use it to help “make sense of what’s happening” to the plants in their care, says Scott Lowman, vice president of applied research at the Institute for Advanced Learning and Research (IALR) in Virginia, a state that’s investing heavily in CEA companies. And although these companies say they’re not aiming for a future with zero human employees, AI is certainly poised to take a lot of human farming intervention out of the equation — for better and worse.
Most of these companies are compiling their own data sets to identify anything that might block the success of their systems. Babylon had already integrated sensor data into its farms to measure heat and humidity, the nutrient content of water, and the amount of light plants receive. Last year, they got a National Science Foundation grant that allowed them to pilot the use of specialized cameras that take pictures in different spectrums to gather some less-obvious visual data about plants’ wellbeing and alert people if something seems off. “Will this plant be healthy tomorrow? Are there things…that the human eye can't see that the plant starts expressing?” says Amandeep Ratte, the company’s head of data science. “If our system can say, Hey, this plant is unhealthy, we can reach out to [users] preemptively about what they’re doing wrong, or is there a disease at the farm?” Ratte says. The earlier the better, to avoid crop failures.
Natural light accounts for 70 percent of Greenswell Growers’ energy use on a sunny day.
Courtesy Greenswell Growers
IALR’s Lowman says that other CEA companies are developing their AI systems to account for the different crops they grow — lettuces come in all shapes and sizes, after all, and each has different growing needs than, for example, tomatoes. The ways they run their operations differs also. Babylon is unusual in its decentralized structure. But centralized growing systems with one main location have variabilities, too. AeroFarms, which recently declared bankruptcy but will continue to run its 140,000-square foot vertical operation in Danville, Virginia, is entirely enclosed and reliant on the intense violet glow of grow lights to produce microgreens.
Different companies have different data needs. What data is essential to AeroFarms isn’t quite the same as for Greenswell Growers located in Goochland County, Virginia. Raising four kinds of lettuce in a 77,000-square-foot automated hydroponic greenhouse, the vagaries of naturally available light, which accounts for 70 percent of Greenswell’s energy use on a sunny day, affect operations. Their tech needs to account for “outside weather impacts,” says president Carl Gupton. “What adjustments do we have to make inside of the greenhouse to offset what's going on outside environmentally, to give that plant optimal conditions? When it's 85 percent humidity outside, the system needs to do X, Y and Z to get the conditions that we want inside.”
AI will help identify diseases, as well as when a plant is thirsty or overly hydrated, when it needs more or less calcium, phosphorous, nitrogen.
Nevertheless, every CEA system has the same core needs — consistent yield of high quality crops to keep up year-round supply to customers. Additionally, “Everybody’s got the same set of problems,” Gupton says. Pests may come into a facility with seeds. A disease called pythium, one of the most common in CEA, can damage plant roots. “Then you have root disease pressures that can also come internally — a change in [growing] substrate can change the way the plant performs,” Gupton says.
AI will help identify diseases, as well as when a plant is thirsty or overly hydrated, when it needs more or less calcium, phosphorous, nitrogen. So, while companies amass their own hyper-specific data sets, Lowman foresees a time within the next decade “when there will be some type of [open-source] database that has the most common types of plant stress identified” that growers will be able to tap into. Such databases will “create a community and move the science forward,” says Lowman.
In fact, IALR is working on assembling images for just such a database now. On so-called “smart tables” inside an Institute lab, a team is growing greens and subjects them to various stressors. Then, they’re administering treatments while taking images of every plant every 15 minutes, says Lowman. Some experiments generate 80,000 images; the challenge lies in analyzing and annotating the vast trove of them, marking each one to reflect outcome—for example increasing the phosphate delivery and the plant’s response to it. Eventually, they’ll be fed into AI systems to help them learn.
For all the enthusiasm surrounding this technology, it’s not without downsides. Training just one AI system can emit over 250,000 pounds of carbon dioxide, according to MIT Technology Review. AI could also be used “to enhance environmental benefit for CEA and optimize [its] energy consumption,” says Rozita Dara, a computer science professor at the University of Guelph in Canada, specializing in AI and data governance, “but we first need to collect data to measure [it].”
The chef-farmers can choose from 45 types of herb and leafy-greens seeds.
Courtesy Babylon Micro-Farms
Any system connected to the Internet of Things is also vulnerable to hacking; if CEA grows to the point where “there are many of these similar farms, and you're depending on feeding a population based on those, it would be quite scary,” Dara says. And there are privacy concerns, too, in systems where imaging is happening constantly. It’s partly for this reason, says Babylon’s Ratte, that the company’s in-farm cameras all “face down into the trays, so the only thing [visible] is pictures of plants.”
Tweaks to improve AI for CEA are happening all the time. Greenswell made its first harvest in 2022 and now has annual data points they can use to start making more intelligent choices about how to feed, water, and supply light to plants, says Gupton. Ratte says he’s confident Babylon’s system can already “get our customers reliable harvests. But in terms of how far we have to go, it's a different problem,” he says. For example, if AI could detect whether the farm is mostly empty—meaning the farm’s user hasn’t planted a new crop of greens—it can alert Babylon to check “what's going on with engagement with this user?” Ratte says. “Do they need more training? Did the main person responsible for the farm quit?”
Lowman says more automation is coming, offering greater ability for systems to identify problems and mitigate them on the spot. “We still have to develop datasets that are specific, so you can have a very clear control plan, [because] artificial intelligence is only as smart as what we tell it, and in plant science, there's so much variation,” he says. He believes AI’s next level will be “looking at those first early days of plant growth: when the seed germinates, how fast it germinates, what it looks like when it germinates.” Imaging all that and pairing it with AI, “can be a really powerful tool, for sure.”
Researchers from the University of Cambridge have laid the foundations for growing synthetic embryos that could develop a beating heart, gut and brain.
The organoid revolution
In 1981, scientists discovered how to keep stem cells alive. This was a significant breakthrough, as stem cells have notoriously rigorous demands. Nevertheless, stem cells remained a relatively niche research area, mainly because scientists didn’t know how to convince the cells to turn into other cells.
Then, in 1987, scientists embedded isolated stem cells in a gelatinous protein mixture called Matrigel, which simulated the three-dimensional environment of animal tissue. The cells thrived, but they also did something remarkable: they created breast tissue capable of producing milk proteins. This was the first organoid — a clump of cells that behave and function like a real organ. The organoid revolution had begun, and it all started with a boob in Jello.
For the next 20 years, it was rare to find a scientist who identified as an “organoid researcher,” but there were many “stem cell researchers” who wanted to figure out how to turn stem cells into other cells. Eventually, they discovered the signals (called growth factors) that stem cells require to differentiate into other types of cells.
For a human embryo (and its organs) to develop successfully, there needs to be a “dialogue” between these three types of stem cells.
By the end of the 2000s, researchers began combining stem cells, Matrigel, and the newly characterized growth factors to create dozens of organoids, from liver organoids capable of producing the bile salts necessary for digesting fat to brain organoids with components that resemble eyes, the spinal cord, and arguably, the beginnings of sentience.
Synthetic embryos
Organoids possess an intrinsic flaw: they are organ-like. They share some characteristics with real organs, making them powerful tools for research. However, no one has found a way to create an organoid with all the characteristics and functions of a real organ. But Magdalena Żernicka-Goetz, a developmental biologist, might have set the foundation for that discovery.
Żernicka-Goetz hypothesized that organoids fail to develop into fully functional organs because organs develop as a collective. Organoid research often uses embryonic stem cells, which are the cells from which the developing organism is created. However, there are two other types of stem cells in an early embryo: stem cells that become the placenta and those that become the yolk sac (where the embryo grows and gets its nutrients in early development). For a human embryo (and its organs) to develop successfully, there needs to be a “dialogue” between these three types of stem cells. In other words, Żernicka-Goetz suspected the best way to grow a functional organoid was to produce a synthetic embryoid.
As described in the aforementioned Nature paper, Żernicka-Goetz and her team mimicked the embryonic environment by mixing these three types of stem cells from mice. Amazingly, the stem cells self-organized into structures and progressed through the successive developmental stages until they had beating hearts and the foundations of the brain.
“Our mouse embryo model not only develops a brain, but also a beating heart [and] all the components that go on to make up the body,” said Żernicka-Goetz. “It’s just unbelievable that we’ve got this far. This has been the dream of our community for years and major focus of our work for a decade and finally we’ve done it.”
If the methods developed by Żernicka-Goetz’s team are successful with human stem cells, scientists someday could use them to guide the development of synthetic organs for patients awaiting transplants. It also opens the door to studying how embryos develop during pregnancy.
This article originally appeared on Big Think, home of the brightest minds and biggest ideas of all time.
