Tag Archives: she
#434786 AI Performed Like a Human on a Gestalt ...
Dr. Been Kim wants to rip open the black box of deep learning.
A senior researcher at Google Brain, Kim specializes in a sort of AI psychology. Like cognitive psychologists before her, she develops various ways to probe the alien minds of artificial neural networks (ANNs), digging into their gory details to better understand the models and their responses to inputs.
The more interpretable ANNs are, the reasoning goes, the easier it is to reveal potential flaws in their reasoning. And if we understand when or why our systems choke, we’ll know when not to use them—a foundation for building responsible AI.
There are already several ways to tap into ANN reasoning, but Kim’s inspiration for unraveling the AI black box came from an entirely different field: cognitive psychology. The field aims to discover fundamental rules of how the human mind—essentially also a tantalizing black box—operates, Kim wrote with her colleagues.
In a new paper uploaded to the pre-publication server arXiv, the team described a way to essentially perform a human cognitive test on ANNs. The test probes how we automatically complete gaps in what we see, so that they form entire objects—for example, perceiving a circle from a bunch of loose dots arranged along a clock face. Psychologist dub this the “law of completion,” a highly influential idea that led to explanations of how our minds generalize data into concepts.
Because deep neural networks in machine vision loosely mimic the structure and connections of the visual cortex, the authors naturally asked: do ANNs also exhibit the law of completion? And what does that tell us about how an AI thinks?
Enter the Germans
The law of completion is part of a series of ideas from Gestalt psychology. Back in the 1920s, long before the advent of modern neuroscience, a group of German experimental psychologists asked: in this chaotic, flashy, unpredictable world, how do we piece together input in a way that leads to meaningful perceptions?
The result is a group of principles known together as the Gestalt effect: that the mind self-organizes to form a global whole. In the more famous words of Gestalt psychologist Kurt Koffka, our perception forms a whole that’s “something else than the sum of its parts.” Not greater than; just different.
Although the theory has its critics, subsequent studies in humans and animals suggest that the law of completion happens on both the cognitive and neuroanatomical level.
Take a look at the drawing below. You immediately “see” a shape that’s actually the negative: a triangle or a square (A and B). Or you further perceive a 3D ball (C), or a snake-like squiggle (D). Your mind fills in blank spots, so that the final perception is more than just the black shapes you’re explicitly given.
Image Credit: Wikimedia Commons contributors, the free media repository.
Neuroscientists now think that the effect comes from how our visual system processes information. Arranged in multiple layers and columns, lower-level neurons—those first to wrangle the data—tend to extract simpler features such as lines or angles. In Gestalt speak, they “see” the parts.
Then, layer by layer, perception becomes more abstract, until higher levels of the visual system directly interpret faces or objects—or things that don’t really exist. That is, the “whole” emerges.
The Experiment Setup
Inspired by these classical experiments, Kim and team developed a protocol to test the Gestalt effect on feed-forward ANNs: one simple, the other, dubbed the “Inception V3,” far more complex and widely used in the machine vision community.
The main idea is similar to the triangle drawings above. First, the team generated three datasets: one set shows complete, ordinary triangles. The second—the “Illusory” set, shows triangles with the edges removed but the corners intact. Thanks to the Gestalt effect, to us humans these generally still look like triangles. The third set also only shows incomplete triangle corners. But here, the corners are randomly rotated so that we can no longer imagine a line connecting them—hence, no more triangle.
To generate a dataset large enough to tease out small effects, the authors changed the background color, image rotation, and other aspects of the dataset. In all, they produced nearly 1,000 images to test their ANNs on.
“At a high level, we compare an ANN’s activation similarities between the three sets of stimuli,” the authors explained. The process is two steps: first, train the AI on complete triangles. Second, test them on the datasets. If the response is more similar between the illusory set and the complete triangle—rather than the randomly rotated set—it should suggest a sort of Gestalt closure effect in the network.
Machine Gestalt
Right off the bat, the team got their answer: yes, ANNs do seem to exhibit the law of closure.
When trained on natural images, the networks better classified the illusory set as triangles than those with randomized connection weights or networks trained on white noise.
When the team dug into the “why,” things got more interesting. The ability to complete an image correlated with the network’s ability to generalize.
Humans subconsciously do this constantly: anything with a handle made out of ceramic, regardless of shape, could easily be a mug. ANNs still struggle to grasp common features—clues that immediately tells us “hey, that’s a mug!” But when they do, it sometimes allows the networks to better generalize.
“What we observe here is that a network that is able to generalize exhibits…more of the closure effect [emphasis theirs], hinting that the closure effect reflects something beyond simply learning features,” the team wrote.
What’s more, remarkably similar to the visual cortex, “higher” levels of the ANNs showed more of the closure effect than lower layers, and—perhaps unsurprisingly—the more layers a network had, the more it exhibited the closure effect.
As the networks learned, their ability to map out objects from fragments also improved. When the team messed around with the brightness and contrast of the images, the AI still learned to see the forest from the trees.
“Our findings suggest that neural networks trained with natural images do exhibit closure,” the team concluded.
AI Psychology
That’s not to say that ANNs recapitulate the human brain. As Google’s Deep Dream, an effort to coax AIs into spilling what they’re perceiving, clearly demonstrates, machine vision sees some truly weird stuff.
In contrast, because they’re modeled after the human visual cortex, perhaps it’s not all that surprising that these networks also exhibit higher-level properties inherent to how we process information.
But to Kim and her colleagues, that’s exactly the point.
“The field of psychology has developed useful tools and insights to study human brains– tools that we may be able to borrow to analyze artificial neural networks,” they wrote.
By tweaking these tools to better analyze machine minds, the authors were able to gain insight on how similarly or differently they see the world from us. And that’s the crux: the point isn’t to say that ANNs perceive the world sort of, kind of, maybe similar to humans. It’s to tap into a wealth of cognitive psychology tools, established over decades using human minds, to probe that of ANNs.
“The work here is just one step along a much longer path,” the authors conclude.
“Understanding where humans and neural networks differ will be helpful for research on interpretability by enlightening the fundamental differences between the two interesting species.”
Image Credit: Popova Alena / Shutterstock.com Continue reading
Posted in Human Robots
#433807 The How, Why, and Whether of Custom ...
A digital afterlife may soon be within reach, but it might not be for your benefit.
The reams of data we’re creating could soon make it possible to create digital avatars that live on after we die, aimed at comforting our loved ones or sharing our experience with future generations.
That may seem like a disappointing downgrade from the vision promised by the more optimistic futurists, where we upload our consciousness to the cloud and live forever in machines. But it might be a realistic possibility in the not-too-distant future—and the first steps have already been taken.
After her friend died in a car crash, Eugenia Kuyda, co-founder of Russian AI startup Luka, trained a neural network-powered chatbot on their shared message history to mimic him. Journalist and amateur coder James Vlahos took a more involved approach, carrying out extensive interviews with his terminally ill father so that he could create a digital clone of him when he died.
For those of us without the time or expertise to build our own artificial intelligence-powered avatar, startup Eternime is offering to take your social media posts and interactions as well as basic personal information to build a copy of you that could then interact with relatives once you’re gone. The service is so far only running a private beta with a handful of people, but with 40,000 on its waiting list, it’s clear there’s a market.
Comforting—Or Creepy?
The whole idea may seem eerily similar to the Black Mirror episode Be Right Back, in which a woman pays a company to create a digital copy of her deceased husband and eventually a realistic robot replica. And given the show’s focus on the emotional turmoil she goes through, people might question whether the idea is a sensible one.
But it’s hard to say at this stage whether being able to interact with an approximation of a deceased loved one would be a help or a hindrance in the grieving process. The fear is that it could make it harder for people to “let go” or “move on,” but others think it could play a useful therapeutic role, reminding people that just because someone is dead it doesn’t mean they’re gone, and providing a novel way for them to express and come to terms with their feelings.
While at present most envisage these digital resurrections as a way to memorialize loved ones, there are also more ambitious plans to use the technology as a way to preserve expertise and experience. A project at MIT called Augmented Eternity is investigating whether we could use AI to trawl through someone’s digital footprints and extract both their knowledge and elements of their personality.
Project leader Hossein Rahnama says he’s already working with a CEO who wants to leave behind a digital avatar that future executives could consult with after he’s gone. And you wouldn’t necessarily have to wait until you’re dead—experts could create virtual clones of themselves that could dispense advice on demand to far more people. These clones could soon be more than simple chatbots, too. Hollywood has already started spending millions of dollars to create 3D scans of its most bankable stars so that they can keep acting beyond the grave.
It’s easy to see the appeal of the idea; imagine if we could bring back Stephen Hawking or Tim Cook to share their wisdom with us. And what if we could create a digital brain trust combining the experience and wisdom of all the world’s greatest thinkers, accessible on demand?
But there are still huge hurdles ahead before we could create truly accurate representations of people by simply trawling through their digital remains. The first problem is data. Most peoples’ digital footprints only started reaching significant proportions in the last decade or so, and cover a relatively small period of their lives. It could take many years before there’s enough data to create more than just a superficial imitation of someone.
And that’s assuming that the data we produce is truly representative of who we are. Carefully-crafted Instagram profiles and cautiously-worded work emails hardly capture the messy realities of most peoples’ lives.
Perhaps if the idea is simply to create a bank of someone’s knowledge and expertise, accurately capturing the essence of their character would be less important. But these clones would also be static. Real people continually learn and change, but a digital avatar is a snapshot of someone’s character and opinions at the point they died. An inability to adapt as the world around them changes could put a shelf life on the usefulness of these replicas.
Who’s Calling the (Digital) Shots?
It won’t stop people trying, though, and that raises a potentially more important question: Who gets to make the calls about our digital afterlife? The subjects, their families, or the companies that hold their data?
In most countries, the law is currently pretty hazy on this topic. Companies like Google and Facebook have processes to let you choose who should take control of your accounts in the event of your death. But if you’ve forgotten to do that, the fate of your virtual remains comes down to a tangle of federal law, local law, and tech company terms of service.
This lack of regulation could create incentives and opportunities for unscrupulous behavior. The voice of a deceased loved one could be a highly persuasive tool for exploitation, and digital replicas of respected experts could be powerful means of pushing a hidden agenda.
That means there’s a pressing need for clear and unambiguous rules. Researchers at Oxford University recently suggested ethical guidelines that would treat our digital remains the same way museums and archaeologists are required to treat mortal remains—with dignity and in the interest of society.
Whether those kinds of guidelines are ever enshrined in law remains to be seen, but ultimately they may decide whether the digital afterlife turns out to be heaven or hell.
Image Credit: frankie’s / Shutterstock.com Continue reading
Posted in Human Robots
#433474 How to Feed Global Demand for ...
“You really can’t justify tuna in Chicago as a source of sustenance.” That’s according to Dr. Sylvia Earle, a National Geographic Society Explorer who was the first female chief scientist at NOAA. She came to the Good Food Institute’s Good Food Conference to deliver a call to action around global food security, agriculture, environmental protection, and the future of consumer choice.
It seems like all options should be on the table to feed an exploding population threatened by climate change. But Dr. Earle, who is faculty at Singularity University, drew a sharp distinction between seafood for sustenance versus seafood as a choice. “There is this widespread claim that we must take large numbers of wildlife from the sea in order to have food security.”
A few minutes later, Dr. Earle directly addressed those of us in the audience. “We know the value of a dead fish,” she said. That’s market price. “But what is the value of a live fish in the ocean?”
That’s when my mind blew open. What is the value—or put another way, the cost—of using the ocean as a major source of protein for humans? How do you put a number on that? Are we talking about dollars and cents, or about something far larger?
Dr. Liz Specht of the Good Food Institute drew the audience’s attention to a strange imbalance. Currently, about half of the yearly global catch of seafood comes from aquaculture. That means that the other half is wild caught. It’s hard to imagine half of your meat coming directly from the forests and the plains, isn’t it? And yet half of the world’s seafood comes from direct harvesting of the oceans, by way of massive overfishing, a terrible toll from bycatch, a widespread lack of regulation and enforcement, and even human rights violations such as slavery.
The search for solutions is on, from both within the fishing industry and from external agencies such as governments and philanthropists. Could there be another way?
Makers of plant-based seafood and clean seafood think they know how to feed the global demand for seafood without harming the ocean. These companies are part of a larger movement harnessing technology to reduce our reliance on wild and domesticated animals—and all the environmental, economic, and ethical issues that come with it.
Producers of plant-based seafood (20 or so currently) are working to capture the taste, texture, and nutrition of conventional seafood without the limitations of geography or the health of a local marine population. Like with plant-based meat, makers of plant-based seafood are harnessing food science and advances in chemistry, biology, and engineering to make great food. The industry’s strategy? Start with what the consumer wants, and then figure out how to achieve that great taste through technology.
So how does plant-based seafood taste? Pretty good, as it turns out. (The biggest benefit of a food-oriented conference is that your mouth is always full!)
I sampled “tuna” salad made from Good Catch Food’s fish-free tuna, which is sourced from legumes; the texture was nearly indistinguishable from that of flaked albacore tuna, and there was no lingering fishy taste to overpower my next bite. In a blind taste test, I probably wouldn’t have known that I was eating a plant-based seafood alternative. Next I reached for Ocean Hugger Food’s Ahimi, a tomato-based alternative to raw tuna. I adore Hawaiian poke, so I was pleasantly surprised when my Ahimi-based poke captured the bite of ahi tuna. It wasn’t quite as delightfully fatty as raw tuna, but with wild tuna populations struggling to recover from a 97% decline in numbers from 40 years ago, Ahimi is a giant stride in the right direction.
These plant-based alternatives aren’t the only game in town, however.
The clean meat industry, which has also been called “cultured meat” or “cellular agriculture,” isn’t seeking to lure consumers away from animal protein. Instead, cells are sampled from live animals and grown in bioreactors—meaning that no animal is slaughtered to produce real meat.
Clean seafood is poised to piggyback off platforms developed for clean meat; growing fish cells in the lab should rely on the same processes as growing meat cells. I know of four companies currently focusing on seafood (Finless Foods, Wild Type, BlueNalu, and Seafuture Sustainable Biotech), and a few more are likely to emerge from stealth mode soon.
Importantly, there’s likely not much difference between growing clean seafood from the top or the bottom of the food chain. Tuna, for example, are top predators that must grow for at least 10 years before they’re suitable as food. Each year, a tuna consumes thousands of pounds of other fish, shellfish, and plankton. That “long tail of groceries,” said Dr. Earle, “is a pretty expensive choice.” Excitingly, clean tuna would “level the trophic playing field,” as Dr. Specht pointed out.
All this is only the beginning of what might be possible.
Combining synthetic biology with clean meat and seafood means that future products could be personalized for individual taste preferences or health needs, by reprogramming the DNA of the cells in the lab. Industries such as bioremediation and biofuels likely have a lot to teach us about sourcing new ingredients and flavors from algae and marine plants. By harnessing rapid advances in automation, robotics, sensors, machine vision, and other big-data analytics, the manufacturing and supply chains for clean seafood could be remarkably safe and robust. Clean seafood would be just that: clean, without pathogens, parasites, or the plastic threatening to fill our oceans, meaning that you could enjoy it raw.
What about price? Dr. Mark Post, a pioneer in clean meat who is also faculty at Singularity University, estimated that 80% of clean-meat production costs come from the expensive medium in which cells are grown—and some ingredients in the medium are themselves sourced from animals, which misses the point of clean meat. Plus, to grow a whole cut of food, like a fish fillet, the cells need to be coaxed into a complex 3D structure with various cell types like muscle cells and fat cells. These two technical challenges must be solved before clean meat and seafood give consumers the experience they want, at the price they want.
In this respect clean seafood has an unusual edge. Most of what we know about growing animal cells in the lab comes from the research and biomedical industries (from tissue engineering, for example)—but growing cells to replace an organ has different constraints than growing cells for food. The link between clean seafood and biomedicine is less direct, empowering innovators to throw out dogma and find novel reagents, protocols, and equipment to grow seafood that captures the tastes, textures, smells, and overall experience of dining by the ocean.
Asked to predict when we’ll be seeing clean seafood in the grocery store, Lou Cooperhouse the CEO of BlueNalu, explained that the challenges aren’t only in the lab: marketing, sales, distribution, and communication with consumers are all critical. As Niya Gupta, the founder of Fork & Goode, said, “The question isn’t ‘can we do it’, but ‘can we sell it’?”
The good news is that the clean meat and seafood industry is highly collaborative; there are at least two dozen companies in the space, and they’re all talking to each other. “This is an ecosystem,” said Dr. Uma Valeti, the co-founder of Memphis Meats. “We’re not competing with each other.” It will likely be at least a decade before science, business, and regulation enable clean meat and seafood to routinely appear on restaurant menus, let alone market shelves.
Until then, think carefully about your food choices. Meditate on Dr. Earle’s question: “What is the real cost of that piece of halibut?” Or chew on this from Dr. Ricardo San Martin, of the Sutardja Center at the University of California, Berkeley: “Food is a system of meanings, not an object.” What are you saying when you choose your food, about your priorities and your values and how you want the future to look? Do you think about animal welfare? Most ethical regulations don’t extend to marine life, and if you don’t think that ocean creatures feel pain, consider the lobster.
Seafood is largely an acquired taste, since most of us don’t live near the water. Imagine a future in which children grow up loving the taste of delicious seafood but without hurting a living animal, the ocean, or the global environment.
Do more than imagine. As Dr. Earle urged us, “Convince the public at large that this is a really cool idea.”
Widely available
Medium availability
Emerging
Gardein
Ahimi (Ocean Hugger)
New Wave Foods
Sophie’s Kitchen
Cedar Lake
To-funa Fish
Quorn
SoFine Foods
Seamore
Vegetarian Plus
Akua
Good Catch
Heritage
Hungry Planet
Odontella
Loma Linda
Heritage Health Food
Terramino Foods
The Vegetarian Butcher
May Wah
VBites
Table based on Figure 5 of the report “An Ocean of Opportunity: Plant-based and clean seafood for sustainable oceans without sacrifice,” from The Good Food Institute.
Image Credit: Tono Balaguer / Shutterstock.com Continue reading
Posted in Human Robots