Tag Archives: language
The question of whether an artificial general intelligence will be developed in the future—and, if so, when it might arrive—is controversial. One (very uncertain) estimate suggests 2070 might be the earliest we could expect to see such technology.
Some futurists point to Moore’s Law and the increasing capacity of machine learning algorithms to suggest that a more general breakthrough is just around the corner. Others suggest that extrapolating exponential improvements in hardware is unwise, and that creating narrow algorithms that can beat humans at specialized tasks brings us no closer to a “general intelligence.”
But evolution has produced minds like the human mind at least once. Surely we could create artificial intelligence simply by copying nature, either by guided evolution of simple algorithms or wholesale emulation of the human brain.
Both of these ideas are far easier to conceive of than they are to achieve. The 302 neurons of the nematode worm’s brain are still an extremely difficult engineering challenge, let alone the 86 billion in a human brain.
Leaving aside these caveats, though, many people are worried about artificial general intelligence. Nick Bostrom’s influential book on superintelligence imagines it will be an agent—an intelligence with a specific goal. Once such an agent reaches a human level of intelligence, it will improve itself—increasingly rapidly as it gets smarter—in pursuit of whatever goal it has, and this “recursive self-improvement” will lead it to become superintelligent.
This “intelligence explosion” could catch humans off guard. If the initial goal is poorly specified or malicious, or if improper safety features are in place, or if the AI decides it would prefer to do something else instead, humans may be unable to control our own creation. Bostrom gives examples of how a seemingly innocuous goal, such as “Make everyone happy,” could be misinterpreted; perhaps the AI decides to drug humanity into a happy stupor, or convert most of the world into computing infrastructure to pursue its goal.
Drexler and Comprehensive AI Services
These are increasingly familiar concerns for an AI that behaves like an agent, seeking to achieve its goal. There are dissenters to this picture of how artificial general intelligence might arise. One notable alternative point of view comes from Eric Drexler, famous for his work on molecular nanotechnology and Engines of Creation, the book that popularized it.
With respect to AI, Drexler believes our view of an artificial intelligence as a single “agent” that acts to maximize a specific goal is too narrow, almost anthropomorphizing AI, or modeling it as a more realistic route towards general intelligence. Instead, he proposes “Comprehensive AI Services” (CAIS) as an alternative route to artificial general intelligence.
What does this mean? Drexler’s argument is that we should look more closely at how machine learning and AI algorithms are actually being developed in the real world. The optimization effort is going into producing algorithms that can provide services and perform tasks like translation, music recommendations, classification, medical diagnoses, and so forth.
AI-driven improvements in technology, argues Drexler, will lead to a proliferation of different algorithms: technology and software improvement, which can automate increasingly more complicated tasks. Recursive improvement in this regime is already occurring—take the newer versions of AlphaGo, which can learn to improve themselves by playing against previous versions.
Many Smart Arms, No Smart Brain
Instead of relying on some unforeseen breakthrough, the CAIS model of AI just assumes that specialized, narrow AI will continue to improve at performing each of its tasks, and the range of tasks that machine learning algorithms will be able to perform will become wider. Ultimately, once a sufficient number of tasks have been automated, the services that an AI will provide will be so comprehensive that they will resemble a general intelligence.
One could then imagine a “general” intelligence as simply an algorithm that is extremely good at matching the task you ask it to perform to the specialized service algorithm that can perform that task. Rather than acting like a single brain that strives to achieve a particular goal, the central AI would be more like a search engine, looking through the tasks it can perform to find the closest match and calling upon a series of subroutines to achieve the goal.
For Drexler, this is inherently a safety feature. Rather than Bostrom’s single, impenetrable, conscious and superintelligent brain (which we must try to psychoanalyze in advance without really knowing what it will look like), we have a network of capabilities. If you don’t want your system to perform certain tasks, you can simply cut it off from access to those services. There is no superintelligent consciousness to outwit or “trap”: more like an extremely high-level programming language that can respond to complicated commands by calling upon one of the myriad specialized algorithms that have been developed by different groups.
This skirts the complex problem of consciousness and all of the sticky moral quandaries that arise in making minds that might be like ours. After all, if you could simulate a human mind, you could simulate it experiencing unimaginable pain. Black Mirror-esque dystopias where emulated minds have no rights and are regularly “erased” or forced to labor in dull and repetitive tasks, hove into view.
Drexler argues that, in this world, there is no need to ever build a conscious algorithm. Yet it seems likely that, at some point, humans will attempt to simulate our own brains, if only in the vain attempt to pursue immortality. This model cannot hold forever. Yet its proponents argue that any world in which we could develop general AI would probably also have developed superintelligent capabilities in a huge range of different tasks, such as computer programming, natural language understanding, and so on. In other words, CAIS arrives first.
The Future In Our Hands?
Drexler argues that his model already incorporates many of the ideas from general AI development. In the marketplace, algorithms compete all the time to perform these services: they undergo the same evolutionary pressures that lead to “higher intelligence,” but the behavior that’s considered superior is chosen by humans, and the nature of the “general intelligence” is far more shaped by human decision-making and human programmers. Development in AI services could still be rapid and disruptive.
But in Drexler’s case, the research and development capacity comes from humans and organizations driven by the desire to improve algorithms that are performing individualized and useful tasks, rather than from a conscious AI recursively reprogramming and improving itself.
In other words, this vision does not absolve us of the responsibility of making our AI safe; if anything, it gives us a greater degree of responsibility. As more and more complex “services” are automated, performing what used to be human jobs at superhuman speed, the economic disruption will be severe.
Equally, as machine learning is trusted to carry out more complex decisions, avoiding algorithmic bias becomes crucial. Shaping each of these individual decision-makers—and trying to predict the complex ways they might interact with each other—is no less daunting a task than specifying the goal for a hypothetical, superintelligent, God-like AI. Arguably, the consequences of the “misalignment” of these services algorithms are already multiplying around us.
The CAIS model bridges the gap between real-world AI, machine learning developments, and real-world safety considerations, as well as the speculative world of superintelligent agents and the safety considerations involved with controlling their behavior. We should keep our minds open as to what form AI and machine learning will take, and how it will influence our societies—and we must take care to ensure that the systems we create don’t end up forcing us all to live in a world of unintended consequences.
Image Credit: MF Production/Shutterstock.com Continue reading
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.
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