Tag Archives: model

#435127 Teaching AI the Concept of ‘Similar, ...

Posted on May 22, 2019 by Android

As a human you instinctively know that a leopard is closer to a cat than a motorbike, but the way we train most AI makes them oblivious to these kinds of relations. Building the concept of similarity into our algorithms could make them far more capable, writes the author of a new paper in Science Robotics.

Convolutional neural networks have revolutionized the field of computer vision to the point that machines are now outperforming humans on some of the most challenging visual tasks. But the way we train them to analyze images is very different from the way humans learn, says Atsuto Maki, an associate professor at KTH Royal Institute of Technology.

“Imagine that you are two years old and being quizzed on what you see in a photo of a leopard,” he writes. “You might answer ‘a cat’ and your parents might say, ‘yeah, not quite but similar’.”

In contrast, the way we train neural networks rarely gives that kind of partial credit. They are typically trained to have very high confidence in the correct label and consider all incorrect labels, whether ”cat” or “motorbike,” equally wrong. That’s a mistake, says Maki, because ignoring the fact that something can be “less wrong” means you’re not exploiting all of the information in the training data.

Even when models are trained this way, there will be small differences in the probabilities assigned to incorrect labels that can tell you a lot about how well the model can generalize what it has learned to unseen data.

If you show a model a picture of a leopard and it gives “cat” a probability of five percent and “motorbike” one percent, that suggests it picked up on the fact that a cat is closer to a leopard than a motorbike. In contrast, if the figures are the other way around it means the model hasn’t learned the broad features that make cats and leopards similar, something that could potentially be helpful when analyzing new data.

If we could boost this ability to identify similarities between classes we should be able to create more flexible models better able to generalize, says Maki. And recent research has demonstrated how variations of an approach called regularization might help us achieve that goal.

Neural networks are prone to a problem called “overfitting,” which refers to a tendency to pay too much attention to tiny details and noise specific to their training set. When that happens, models will perform excellently on their training data but poorly when applied to unseen test data without these particular quirks.

Regularization is used to circumvent this problem, typically by reducing the network’s capacity to learn all this unnecessary information and therefore boost its ability to generalize to new data. Techniques are varied, but generally involve modifying the network’s structure or the strength of the weights between artificial neurons.

More recently, though, researchers have suggested new regularization approaches that work by encouraging a broader spread of probabilities across all classes. This essentially helps them capture more of the class similarities, says Maki, and therefore boosts their ability to generalize.

One such approach was devised in 2017 by Google Brain researchers, led by deep learning pioneer Geoffrey Hinton. They introduced a penalty to their training process that directly punished overconfident predictions in the model’s outputs, and a technique called label smoothing that prevents the largest probability becoming much larger than all others. This meant the probabilities were lower for correct labels and higher for incorrect ones, which was found to boost performance of models on varied tasks from image classification to speech recognition.

Another came from Maki himself in 2017 and achieves the same goal, but by suppressing high values in the model’s feature vector—the mathematical construct that describes all of an object’s important characteristics. This has a knock-on effect on the spread of output probabilities and also helped boost performance on various image classification tasks.

While it’s still early days for the approach, the fact that humans are able to exploit these kinds of similarities to learn more efficiently suggests that models that incorporate them hold promise. Maki points out that it could be particularly useful in applications such as robotic grasping, where distinguishing various similar objects is important.

Image Credit: Marianna Kalashnyk / Shutterstock.com Continue reading →

Posted in Human Robots

#435080 12 Ways Big Tech Can Take Big Action on ...

Posted on May 14, 2019 by Android

Bill Gates and Mark Zuckerberg have invested $1 billion in Breakthrough Energy to fund next-generation solutions to tackle climate. But there is a huge risk that any successful innovation will only reach the market as the world approaches 2030 at the earliest.

We now know that reducing the risk of dangerous climate change means halving global greenhouse gas emissions by that date—in just 11 years. Perhaps Gates, Zuckerberg, and all the tech giants should invest equally in innovations to do with how their own platforms —search, social media, eCommerce—can support societal behavior changes to drive down emissions.

After all, the tech giants influence the decisions of four billion consumers every day. It is time for a social contract between tech and society.

Recently myself and collaborator Johan Falk published a report during the World Economic Forum in Davos outlining 12 ways the tech sector can contribute to supporting societal goals to stabilize Earth’s climate.

Become genuine climate guardians

Tech giants go to great lengths to show how serious they are about reducing their emissions. But I smell cognitive dissonance. Google and Microsoft are working in partnership with oil companies to develop AI tools to help maximize oil recovery. This is not the behavior of companies working flat-out to stabilize Earth’s climate. Indeed, few major tech firms have visions that indicate a stable and resilient planet might be a good goal, yet AI alone has the potential to slash greenhouse gas emissions by four percent by 2030—equivalent to the emissions of Australia, Canada, and Japan combined.

We are now developing a playbook, which we plan to publish later this year at the UN climate summit, about making it as simple as possible for a CEO to become a climate guardian.

Hey Alexa, do you care about the stability of Earth’s climate?

Increasingly, consumers are delegating their decisions to narrow artificial intelligence like Alexa and Siri. Welcome to a world of zero-click purchases.

Should algorithms and information architecture be designed to nudge consumer behavior towards low-carbon choices, for example by making these options the default? We think so. People don’t mind being nudged; in fact, they welcome efforts to make their lives better. For instance, if I want to lose weight, I know I will need all the help I can get. Let’s ‘nudge for good’ and experiment with supporting societal goals.

Use social media for good

Facebook’s goal is to bring the world closer together. With 2.2 billion users on the platform, CEO Mark Zuckerberg can reasonably claim this goal is possible. But social media has changed the flow of information in the world, creating a lucrative industry around a toxic brown-cloud of confusion and anger, with frankly terrifying implications for democracy. This has been linked to the rise of nationalism and populism, and to the election of leaders who shun international cooperation, dismiss scientific knowledge, and reverse climate action at a moment when we need it more than ever.

Social media tools need re-engineering to help people make sense of the world, support democratic processes, and build communities around societal goals. Make this your mission.

Design for a future on Earth

Almost everything is designed with computer software, from buildings to mobile phones to consumer packaging. It is time to make zero-carbon design the new default and design products for sharing, re-use and disassembly.

The future is circular

Halving emissions in a decade will require all companies to adopt circular business models to reduce material use. Some tech companies are leading the charge. Apple has committed to becoming 100 percent circular as soon as possible. Great.

While big tech companies strive to be market leaders here, many other companies lack essential knowledge. Tech companies can support rapid adoption in different economic sectors, not least because they have the know-how to scale innovations exponentially. It makes business sense. If economies of scale drive the price of recycled steel and aluminium down, everyone wins.

Reward low-carbon consumption

eCommerce platforms can create incentives for low-carbon consumption. The world’s largest experiment in greening consumer behavior is Ant Forest, set up by Chinese fintech giant Ant Financial.

An estimated 300 million customers—similar to the population of the United States—gain points for making low-carbon choices such as walking to work, using public transport, or paying bills online. Virtual points are eventually converted into real trees. Sure, big questions remain about its true influence on emissions, but this is a space for rapid experimentation for big impact.

Make information more useful

Science is our tool for defining reality. Scientific consensus is how we attain reliable knowledge. Even after the information revolution, reliable knowledge about the world remains fragmented and unstructured. Build the next generation of search engines to genuinely make the world’s knowledge useful for supporting societal goals.

We need to put these tools towards supporting shared world views of the state of the planet based on the best science. New AI tools being developed by startups like Iris.ai can help see through the fog. From Alexa to Google Home and Siri, the future is “Voice”, but who chooses the information source? The highest bidder? Again, the implications for climate are huge.

Create new standards for digital advertising and marketing

Half of global ad revenue will soon be online, and largely going to a small handful of companies. How about creating a novel ethical standard on what is advertised and where? Companies could consider promoting sustainable choices and healthy lifestyles and limiting advertising of high-emissions products such as cheap flights.

We are what we eat

It is no secret that tech is about to disrupt grocery. The supermarkets of the future will be built on personal consumer data. With about two billion people either obese or overweight, revolutions in choice architecture could support positive diet choices, reduce meat consumption, halve food waste and, into the bargain, slash greenhouse gas emissions.

The future of transport is not cars, it’s data

The 2020s look set to be the biggest disruption of the automobile industry since Henry Ford unveiled the Model T. Two seismic shifts are on their way.

First, electric cars now compete favorably with petrol engines on range. Growth will reach an inflection point within a year or two once prices reach parity. The death of the internal combustion engine in Europe and Asia is assured with end dates announced by China, India, France, the UK, and most of Scandinavia. Dates range from 2025 (Norway) to 2040 (UK and China).

Tech giants can accelerate the demise. Uber recently announced a passenger surcharge to help London drivers save around $1,500 a year towards the cost of an electric car.

Second, driverless cars can shift the transport economic model from ownership to service and ride sharing. A complete shift away from privately-owned vehicles is around the corner, with large implications for emissions.

Clean-energy living and working

Most buildings are barely used and inefficiently heated and cooled. Digitization can slash this waste and its corresponding emissions through measurement, monitoring, and new business models to use office space. While, just a few unicorns are currently in this space, the potential is enormous. Buildings are one of the five biggest sources of emissions, yet have the potential to become clean energy producers in a distributed energy network.

Creating liveable cities

More cities are setting ambitious climate targets to halve emissions in a decade or even less. Tech companies can support this transition by driving demand for low-carbon services for their workforces and offices, but also by providing tools to help monitor emissions and act to reduce them. Google, for example, is collecting travel and other data from across cities to estimate emissions in real time. This is possible through technologies like artificial intelligence and the internet of things. But beware of smart cities that turn out to be not so smart. Efficiencies can reduce resilience when cities face crises.

It’s a Start
Of course, it will take more than tech to solve the climate crisis. But tech is a wildcard. The actions of the current tech giants and their acolytes could serve to destabilize the climate further or bring it under control.

We need a new social contract between tech companies and society to achieve societal goals. The alternative is unthinkable. Without drastic action now, climate chaos threatens to engulf us all. As this future approaches, regulators will be forced to take ever more draconian action to rein in the problem. Acting now will reduce that risk.

Note: A version of this article was originally published on World Economic Forum

Image Credit: Bruce Rolff / Shutterstock.com Continue reading →

Posted in Human Robots

#435056 How Researchers Used AI to Better ...

Posted on May 9, 2019 by Android

A few years back, DeepMind’s Demis Hassabis famously prophesized that AI and neuroscience will positively feed into each other in a “virtuous circle.” If realized, this would fundamentally expand our insight into intelligence, both machine and human.

We’ve already seen some proofs of concept, at least in the brain-to-AI direction. For example, memory replay, a biological mechanism that fortifies our memories during sleep, also boosted AI learning when abstractly appropriated into deep learning models. Reinforcement learning, loosely based on our motivation circuits, is now behind some of AI’s most powerful tools.

Hassabis is about to be proven right again.

Last week, two studies independently tapped into the power of ANNs to solve a 70-year-old neuroscience mystery: how does our visual system perceive reality?

The first, published in Cell, used generative networks to evolve DeepDream-like images that hyper-activate complex visual neurons in monkeys. These machine artworks are pure nightmare fuel to the human eye; but together, they revealed a fundamental “visual hieroglyph” that may form a basic rule for how we piece together visual stimuli to process sight into perception.

In the second study, a team used a deep ANN model—one thought to mimic biological vision—to synthesize new patterns tailored to control certain networks of visual neurons in the monkey brain. When directly shown to monkeys, the team found that the machine-generated artworks could reliably activate predicted populations of neurons. Future improved ANN models could allow even better control, giving neuroscientists a powerful noninvasive tool to study the brain. The work was published in Science.

The individual results, though fascinating, aren’t necessarily the point. Rather, they illustrate how scientists are now striving to complete the virtuous circle: tapping AI to probe natural intelligence. Vision is only the beginning—the tools can potentially be expanded into other sensory domains. And the more we understand about natural brains, the better we can engineer artificial ones.

It’s a “great example of leveraging artificial intelligence to study organic intelligence,” commented Dr. Roman Sandler at Kernel.co on Twitter.

Why Vision?
ANNs and biological vision have quite the history.

In the late 1950s, the legendary neuroscientist duo David Hubel and Torsten Wiesel became some of the first to use mathematical equations to understand how neurons in the brain work together.

In a series of experiments—many using cats—the team carefully dissected the structure and function of the visual cortex. Using myriads of images, they revealed that vision is processed in a hierarchy: neurons in “earlier” brain regions, those closer to the eyes, tend to activate when they “see” simple patterns such as lines. As we move deeper into the brain, from the early V1 to a nub located slightly behind our ears, the IT cortex, neurons increasingly respond to more complex or abstract patterns, including faces, animals, and objects. The discovery led some scientists to call certain IT neurons “Jennifer Aniston cells,” which fire in response to pictures of the actress regardless of lighting, angle, or haircut. That is, IT neurons somehow extract visual information into the “gist” of things.

That’s not trivial. The complex neural connections that lead to increasing abstraction of what we see into what we think we see—what we perceive—is a central question in machine vision: how can we teach machines to transform numbers encoding stimuli into dots, lines, and angles that eventually form “perceptions” and “gists”? The answer could transform self-driving cars, facial recognition, and other computer vision applications as they learn to better generalize.

Hubel and Wiesel’s Nobel-prize-winning studies heavily influenced the birth of ANNs and deep learning. Much of earlier ANN “feed-forward” model structures are based on our visual system; even today, the idea of increasing layers of abstraction—for perception or reasoning—guide computer scientists to build AI that can better generalize. The early romance between vision and deep learning is perhaps the bond that kicked off our current AI revolution.

It only seems fair that AI would feed back into vision neuroscience.

Hieroglyphs and Controllers
In the Cell study, a team led by Dr. Margaret Livingstone at Harvard Medical School tapped into generative networks to unravel IT neurons’ complex visual alphabet.

Scientists have long known that neurons in earlier visual regions (V1) tend to fire in response to “grating patches” oriented in certain ways. Using a limited set of these patches like letters, V1 neurons can “express a visual sentence” and represent any image, said Dr. Arash Afraz at the National Institute of Health, who was not involved in the study.

But how IT neurons operate remained a mystery. Here, the team used a combination of genetic algorithms and deep generative networks to “evolve” computer art for every studied neuron. In seven monkeys, the team implanted electrodes into various parts of the visual IT region so that they could monitor the activity of a single neuron.

The team showed each monkey an initial set of 40 images. They then picked the top 10 images that stimulated the highest neural activity, and married them to 30 new images to “evolve” the next generation of images. After 250 generations, the technique, XDREAM, generated a slew of images that mashed up contorted face-like shapes with lines, gratings, and abstract shapes.

This image shows the evolution of an optimum image for stimulating a visual neuron in a monkey. Image Credit: Ponce, Xiao, and Schade et al. – Cell.
“The evolved images look quite counter-intuitive,” explained Afraz. Some clearly show detailed structures that resemble natural images, while others show complex structures that can’t be characterized by our puny human brains.

This figure shows natural images (right) and images evolved by neurons in the inferotemporal cortex of a monkey (left). Image Credit: Ponce, Xiao, and Schade et al. – Cell.
“What started to emerge during each experiment were pictures that were reminiscent of shapes in the world but were not actual objects in the world,” said study author Carlos Ponce. “We were seeing something that was more like the language cells use with each other.”

This image was evolved by a neuron in the inferotemporal cortex of a monkey using AI. Image Credit: Ponce, Xiao, and Schade et al. – Cell.
Although IT neurons don’t seem to use a simple letter alphabet, it does rely on a vast array of characters like hieroglyphs or Chinese characters, “each loaded with more information,” said Afraz.

The adaptive nature of XDREAM turns it into a powerful tool to probe the inner workings of our brains—particularly for revealing discrepancies between biology and models.

The Science study, led by Dr. James DiCarlo at MIT, takes a similar approach. Using ANNs to generate new patterns and images, the team was able to selectively predict and independently control neuron populations in a high-level visual region called V4.

“So far, what has been done with these models is predicting what the neural responses would be to other stimuli that they have not seen before,” said study author Dr. Pouya Bashivan. “The main difference here is that we are going one step further and using the models to drive the neurons into desired states.”

It suggests that our current ANN models for visual computation “implicitly capture a great deal of visual knowledge” which we can’t really describe, but which the brain uses to turn vision information into perception, the authors said. By testing AI-generated images on biological vision, however, the team concluded that today’s ANNs have a degree of understanding and generalization. The results could potentially help engineer even more accurate ANN models of biological vision, which in turn could feed back into machine vision.

“One thing is clear already: Improved ANN models … have led to control of a high-level neural population that was previously out of reach,” the authors said. “The results presented here have likely only scratched the surface of what is possible with such implemented characterizations of the brain’s neural networks.”

To Afraz, the power of AI here is to find cracks in human perception—both our computational models of sensory processes, as well as our evolved biological software itself. AI can be used “as a perfect adversarial tool to discover design cracks” of IT, said Afraz, such as finding computer art that “fools” a neuron into thinking the object is something else.

“As artificial intelligence researchers develop models that work as well as the brain does—or even better—we will still need to understand which networks are more likely to behave safely and further human goals,” said Ponce. “More efficient AI can be grounded by knowledge of how the brain works.”

Image Credit: Sangoiri / Shutterstock.com Continue reading →

Posted in Human Robots

#435023 Inflatable Robot Astronauts and How to ...

Posted on May 1, 2019 by Android

The typical cultural image of a robot—as a steel, chrome, humanoid bucket of bolts—is often far from the reality of cutting-edge robotics research. There are difficulties, both social and technological, in realizing the image of a robot from science fiction—let alone one that can actually help around the house. Often, it’s simply the case that great expense in producing a humanoid robot that can perform dozens of tasks quite badly is less appropriate than producing some other design that’s optimized to a specific situation.

A team of scientists from Brigham Young University has received funding from NASA to investigate an inflatable robot called, improbably, King Louie. The robot was developed by Pneubotics, who have a long track record in the world of soft robotics.

In space, weight is at a premium. The world watched in awe and amusement when Commander Chris Hadfield sang “Space Oddity” from the International Space Station—but launching that guitar into space likely cost around $100,000. A good price for launching payload into outer space is on the order of $10,000 per pound ($22,000/kg).

For that price, it would cost a cool $1.7 million to launch Boston Dynamics’ famous ATLAS robot to the International Space Station, and its bulk would be inconvenient in the cramped living quarters available. By contrast, an inflatable robot like King Louie is substantially lighter and can simply be deflated and folded away when not in use. The robot can be manufactured from cheap, lightweight, and flexible materials, and minor damage is easy to repair.

Inflatable Robots Under Pressure
The concept of inflatable robots is not new: indeed, earlier prototypes of King Louie were exhibited back in 2013 at Google I/O’s After Hours, flailing away at each other in a boxing ring. Sparks might fly in fights between traditional robots, but the aim here was to demonstrate that the robots are passively safe: the soft, inflatable figures won’t accidentally smash delicate items when moving around.

Health and safety regulations form part of the reason why robots don’t work alongside humans more often, but soft robots would be far safer to use in healthcare or around children (whose first instinct, according to BYU’s promotional video, is either to hug or punch King Louie.) It’s also much harder to have nightmarish fantasies about robotic domination with these friendlier softbots: Terminator would’ve been a much shorter franchise if Skynet’s droids were inflatable.

Robotic exoskeletons are increasingly used for physical rehabilitation therapies, as well as for industrial purposes. As countries like Japan seek to care for their aging populations with robots and alleviate the burden on nurses, who suffer from some of the highest rates of back injuries of any profession, soft robots will become increasingly attractive for use in healthcare.

Precision and Proprioception
The main issue is one of control. Rigid, metallic robots may be more expensive and more dangerous, but the simple fact of their rigidity makes it easier to map out and control the precise motions of each of the robot’s limbs, digits, and actuators. Individual motors attached to these rigid robots can allow for a great many degrees of freedom—individual directions in which parts of the robot can move—and precision control.

For example, ATLAS has 28 degrees of freedom, while Shadow’s dexterous robot hand alone has 20. This is much harder to do with an inflatable robot, for precisely the same reasons that make it safer. Without hard and rigid bones, other methods of control must be used.

In the case of King Louie, the robot is made up of many expandable air chambers. An air-compressor changes the pressure levels in these air chambers, allowing them to expand and contract. This harks back to some of the earliest pneumatic automata. Pairs of chambers act antagonistically, like muscles, such that when one chamber “tenses,” another relaxes—allowing King Louie to have, for example, four degrees of freedom in each of its arms.

The robot is also surprisingly strong. Professor Killpack, who works at BYU on the project, estimates that its payload is comparable to other humanoid robots on the market, like Rethink Robotics’ Baxter (RIP).

Proprioception, that sixth sense that allows us to map out and control our own bodies and muscles in fine detail, is being enhanced for a wider range of soft, flexible robots with the use of machine learning algorithms connected to input from a whole host of sensors on the robot’s body.

Part of the reason this is so complicated with soft, flexible robots is that the shape and “map” of the robot’s body can change; that’s the whole point. But this means that every time King Louie is inflated, its body is a slightly different shape; when it becomes deformed, for example due to picking up objects, the shape changes again, and the complex ways in which the fabric can twist and bend are far more difficult to model and sense than the behavior of the rigid metal of King Louie’s hard counterparts. When you’re looking for precision, seemingly-small changes can be the difference between successfully holding an object or dropping it.

Learning to Move
Researchers at BYU are therefore spending a great deal of time on how to control the soft-bot enough to make it comparably useful. One method involves the commercial tracking technology used in the Vive VR system: by moving the game controller, which provides a constant feedback to the robot’s arm, you can control its position. Since the tracking software provides an estimate of the robot’s joint angles and continues to provide feedback until the arm is correctly aligned, this type of feedback method is likely to work regardless of small changes to the robot’s shape.

The other technologies the researchers are looking into for their softbot include arrays of flexible, tactile sensors to place on the softbot’s skin, and minimizing the complex cross-talk between these arrays to get coherent information about the robot’s environment. As with some of the new proprioception research, the project is looking into neural networks as a means of modeling the complicated dynamics—the motion and response to forces—of the softbot. This method relies on large amounts of observational data, mapping how the robot is inflated and how it moves, rather than explicitly understanding and solving the equations that govern its motion—which hopefully means the methods can work even as the robot changes.

There’s still a long way to go before soft and inflatable robots can be controlled sufficiently well to perform all the tasks they might be used for. Ultimately, no one robotic design is likely to be perfect for any situation.

Nevertheless, research like this gives us hope that one day, inflatable robots could be useful tools, or even companions, at which point the advertising slogans write themselves: Don’t let them down, and they won’t let you down!

Image Credit: Brigham Young University. Continue reading →

Posted in Human Robots

#434837 In Defense of Black Box AI

Posted on April 17, 2019 by Android

Deep learning is powering some amazing new capabilities, but we find it hard to scrutinize the workings of these algorithms. Lack of interpretability in AI is a common concern and many are trying to fix it, but is it really always necessary to know what’s going on inside these “black boxes”?

In a recent perspective piece for Science, Elizabeth Holm, a professor of materials science and engineering at Carnegie Mellon University, argued in defense of the black box algorithm. I caught up with her last week to find out more.

Edd Gent: What’s your experience with black box algorithms?

Elizabeth Holm: I got a dual PhD in materials science and engineering and scientific computing. I came to academia about six years ago and part of what I wanted to do in making this career change was to refresh and revitalize my computer science side.

I realized that computer science had changed completely. It used to be about algorithms and making codes run fast, but now it’s about data and artificial intelligence. There are the interpretable methods like random forest algorithms, where we can tell how the machine is making its decisions. And then there are the black box methods, like convolutional neural networks.

Once in a while we can find some information about their inner workings, but most of the time we have to accept their answers and kind of probe around the edges to figure out the space in which we can use them and how reliable and accurate they are.

EG: What made you feel like you had to mount a defense of these black box algorithms?

EH: When I started talking with my colleagues, I found that the black box nature of many of these algorithms was a real problem for them. I could understand that because we’re scientists, we always want to know why and how.

It got me thinking as a bit of a contrarian, “Are black boxes all bad? Must we reject them?” Surely not, because human thought processes are fairly black box. We often rely on human thought processes that the thinker can’t necessarily explain.

It’s looking like we’re going to be stuck with these methods for a while, because they’re really helpful. They do amazing things. And so there’s a very pragmatic realization that these are the best methods we’ve got to do some really important problems, and we’re not right now seeing alternatives that are interpretable. We’re going to have to use them, so we better figure out how.

EG: In what situations do you think we should be using black box algorithms?

EH: I came up with three rules. The simplest rule is: when the cost of a bad decision is small and the value of a good decision is high, it’s worth it. The example I gave in the paper is targeted advertising. If you send an ad no one wants it doesn’t cost a lot. If you’re the receiver it doesn’t cost a lot to get rid of it.

There are cases where the cost is high, and that’s then we choose the black box if it’s the best option to do the job. Things get a little trickier here because we have to ask “what are the costs of bad decisions, and do we really have them fully characterized?” We also have to be very careful knowing that our systems may have biases, they may have limitations in where you can apply them, they may be breakable.

But at the same time, there are certainly domains where we’re going to test these systems so extensively that we know their performance in virtually every situation. And if their performance is better than the other methods, we need to do it. Self driving vehicles are a significant example—it’s almost certain they’re going to have to use black box methods, and that they’re going to end up being better drivers than humans.

The third rule is the more fun one for me as a scientist, and that’s the case where the black box really enlightens us as to a new way to look at something. We have trained a black box to recognize the fracture energy of breaking a piece of metal from a picture of the broken surface. It did a really good job, and humans can’t do this and we don’t know why.

What the computer seems to be seeing is noise. There’s a signal in that noise, and finding it is very difficult, but if we do we may find something significant to the fracture process, and that would be an awesome scientific discovery.

EG: Do you think there’s been too much emphasis on interpretability?

EH: I think the interpretability problem is a fundamental, fascinating computer science grand challenge and there are significant issues where we need to have an interpretable model. But how I would frame it is not that there’s too much emphasis on interpretability, but rather that there’s too much dismissiveness of uninterpretable models.

I think that some of the current social and political issues surrounding some very bad black box outcomes have convinced people that all machine learning and AI should be interpretable because that will somehow solve those problems.

Asking humans to explain their rationale has not eliminated bias, or stereotyping, or bad decision-making in humans. Relying too much on interpreted ability perhaps puts the responsibility in the wrong place for getting better results. I can make a better black box without knowing exactly in what way the first one was bad.

EG: Looking further into the future, do you think there will be situations where humans will have to rely on black box algorithms to solve problems we can’t get our heads around?

EH: I do think so, and it’s not as much of a stretch as we think it is. For example, humans don’t design the circuit map of computer chips anymore. We haven’t for years. It’s not a black box algorithm that designs those circuit boards, but we’ve long since given up trying to understand a particular computer chip’s design.

With the billions of circuits in every computer chip, the human mind can’t encompass it, either in scope or just the pure time that it would take to trace every circuit. There are going to be cases where we want a system so complex that only the patience that computers have and their ability to work in very high-dimensional spaces is going to be able to do it.

So we can continue to argue about interpretability, but we need to acknowledge that we’re going to need to use black boxes. And this is our opportunity to do our due diligence to understand how to use them responsibly, ethically, and with benefits rather than harm. And that’s going to be a social conversation as well as as a scientific one.

*Responses have been edited for length and style

Image Credit: Chingraph / Shutterstock.com Continue reading →

Posted in Human Robots