Tag Archives: speak
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.
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.
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
As over-hyped as artificial intelligence is—everyone’s talking about it, few fully understand it, it might leave us all unemployed but also solve all the world’s problems—its list of accomplishments is growing. AI can now write realistic-sounding text, give a debating champ a run for his money, diagnose illnesses, and generate fake human faces—among much more.
After training these systems on massive datasets, their creators essentially just let them do their thing to arrive at certain conclusions or outcomes. The problem is that more often than not, even the creators don’t know exactly why they’ve arrived at those conclusions or outcomes. There’s no easy way to trace a machine learning system’s rationale, so to speak. The further we let AI go down this opaque path, the more likely we are to end up somewhere we don’t want to be—and may not be able to come back from.
In a panel at the South by Southwest interactive festival last week titled “Ethics and AI: How to plan for the unpredictable,” experts in the field shared their thoughts on building more transparent, explainable, and accountable AI systems.
Not New, but Different
Ryan Welsh, founder and director of explainable AI startup Kyndi, pointed out that having knowledge-based systems perform advanced tasks isn’t new; he cited logistical, scheduling, and tax software as examples. What’s new is the learning component, our inability to trace how that learning occurs, and the ethical implications that could result.
“Now we have these systems that are learning from data, and we’re trying to understand why they’re arriving at certain outcomes,” Welsh said. “We’ve never actually had this broad society discussion about ethics in those scenarios.”
Rather than continuing to build AIs with opaque inner workings, engineers must start focusing on explainability, which Welsh broke down into three subcategories. Transparency and interpretability come first, and refer to being able to find the units of high influence in a machine learning network, as well as the weights of those units and how they map to specific data and outputs.
Then there’s provenance: knowing where something comes from. In an ideal scenario, for example, Open AI’s new text generator would be able to generate citations in its text that reference academic (and human-created) papers or studies.
Explainability itself is the highest and final bar and refers to a system’s ability to explain itself in natural language to the average user by being able to say, “I generated this output because x, y, z.”
“Humans are unique in our ability and our desire to ask why,” said Josh Marcuse, executive director of the Defense Innovation Board, which advises Department of Defense senior leaders on innovation. “The reason we want explanations from people is so we can understand their belief system and see if we agree with it and want to continue to work with them.”
Similarly, we need to have the ability to interrogate AIs.
Two Types of Thinking
Welsh explained that one big barrier standing in the way of explainability is the tension between the deep learning community and the symbolic AI community, which see themselves as two different paradigms and historically haven’t collaborated much.
Symbolic or classical AI focuses on concepts and rules, while deep learning is centered around perceptions. In human thought this is the difference between, for example, deciding to pass a soccer ball to a teammate who is open (you make the decision because conceptually you know that only open players can receive passes), and registering that the ball is at your feet when someone else passes it to you (you’re taking in information without making a decision about it).
“Symbolic AI has abstractions and representation based on logic that’s more humanly comprehensible,” Welsh said. To truly mimic human thinking, AI needs to be able to both perceive information and conceptualize it. An example of perception (deep learning) in an AI is recognizing numbers within an image, while conceptualization (symbolic learning) would give those numbers a hierarchical order and extract rules from the hierachy (4 is greater than 3, and 5 is greater than 4, therefore 5 is also greater than 3).
Explainability comes in when the system can say, “I saw a, b, and c, and based on that decided x, y, or z.” DeepMind and others have recently published papers emphasizing the need to fuse the two paradigms together.
Implications Across Industries
One of the most prominent fields where AI ethics will come into play, and where the transparency and accountability of AI systems will be crucial, is defense. Marcuse said, “We’re accountable beings, and we’re responsible for the choices we make. Bringing in tech or AI to a battlefield doesn’t strip away that meaning and accountability.”
In fact, he added, rather than worrying about how AI might degrade human values, people should be asking how the tech could be used to help us make better moral choices.
It’s also important not to conflate AI with autonomy—a worst-case scenario that springs to mind is an intelligent destructive machine on a rampage. But in fact, Marcuse said, in the defense space, “We have autonomous systems today that don’t rely on AI, and most of the AI systems we’re contemplating won’t be autonomous.”
The US Department of Defense released its 2018 artificial intelligence strategy last month. It includes developing a robust and transparent set of principles for defense AI, investing in research and development for AI that’s reliable and secure, continuing to fund research in explainability, advocating for a global set of military AI guidelines, and finding ways to use AI to reduce the risk of civilian casualties and other collateral damage.
Though these were designed with defense-specific aims in mind, Marcuse said, their implications extend across industries. “The defense community thinks of their problems as being unique, that no one deals with the stakes and complexity we deal with. That’s just wrong,” he said. Making high-stakes decisions with technology is widespread; safety-critical systems are key to aviation, medicine, and self-driving cars, to name a few.
Marcuse believes the Department of Defense can invest in AI safety in a way that has far-reaching benefits. “We all depend on technology to keep us alive and safe, and no one wants machines to harm us,” he said.
A Creation Superior to Its Creator
That said, we’ve come to expect technology to meet our needs in just the way we want, all the time—servers must never be down, GPS had better not take us on a longer route, Google must always produce the answer we’re looking for.
With AI, though, our expectations of perfection may be less reasonable.
“Right now we’re holding machines to superhuman standards,” Marcuse said. “We expect them to be perfect and infallible.” Take self-driving cars. They’re conceived of, built by, and programmed by people, and people as a whole generally aren’t great drivers—just look at traffic accident death rates to confirm that. But the few times self-driving cars have had fatal accidents, there’s been an ensuing uproar and backlash against the industry, as well as talk of implementing more restrictive regulations.
This can be extrapolated to ethics more generally. We as humans have the ability to explain our decisions, but many of us aren’t very good at doing so. As Marcuse put it, “People are emotional, they confabulate, they lie, they’re full of unconscious motivations. They don’t pass the explainability test.”
Why, then, should explainability be the standard for AI?
Even if humans aren’t good at explaining our choices, at least we can try, and we can answer questions that probe at our decision-making process. A deep learning system can’t do this yet, so working towards being able to identify which input data the systems are triggering on to make decisions—even if the decisions and the process aren’t perfect—is the direction we need to head.
Image Credit: a-image / Shutterstock.com Continue reading