Category Archives: Uncategorized

Nena: ” create your future in the present moment”

Source: Live Nation TV, Sep 2016

On your new album Oldschool there’s a song called “Genau Jetzt” that is similarly moving.
I wrote the song with [Samy Deluxe], a famous German rap guy. We are both people who are very conscious about the present moment. Life is not about the past or the future; you create your future in the present moment. That’s what the song is all about. Even if you don’t understand the lyrics, somehow you get the feeling of the song.

Unexplainable AI

Source: MIT Technology Review, Apr 2017

The mysterious mind of this vehicle points to a looming issue with artificial intelligence.

Deep learning, the most common of these approaches, represents a fundamentally different way to program computers.

This raises mind-boggling questions. As the technology advances, we might soon cross some threshold beyond which using AI requires a leap of faith. Sure, we humans can’t always truly explain our thought processes either—but we find ways to intuitively trust and gauge people. Will that also be possible with machines that think and make decisions differently from the way a human would? We’ve never before built machines that operate in ways their creators don’t understand. How well can we expect to communicate—and get along with—intelligent machines that could be unpredictable and inscrutable?

From the outset, there were two schools of thought regarding how understandable, or explainable, AI ought to be. Many thought it made the most sense to build machines that reasoned according to rules and logic, making their inner workings transparent to anyone who cared to examine some code. Others felt that intelligence would more easily emerge if machines took inspiration from biology, and learned by observing and experiencing. This meant turning computer programming on its head. Instead of a programmer writing the commands to solve a problem, the program generates its own algorithm based on example data and a desired output. The machine-learning techniques that would later evolve into today’s most powerful AI systems followed the latter path: the machine essentially programs itself.

At first this approach was of limited practical use, and in the 1960s and ’70s it remained largely confined to the fringes of the field. Then the computerization of many industries and the emergence of large data sets renewed interest. That inspired the development of more powerful machine-learning techniques, especially new versions of one known as the artificial neural network. By the 1990s, neural networks could automatically digitize handwritten characters.

The workings of any machine-learning technology are inherently more opaque, even to computer scientists, than a hand-coded system. This is not to say that all future AI techniques will be equally unknowable. But by its nature, deep learning is a particularly dark black box.

You can’t just look inside a deep neural network to see how it works. A network’s reasoning is embedded in the behavior of thousands of simulated neurons, arranged into dozens or even hundreds of intricately interconnected layers. The neurons in the first layer each receive an input, like the intensity of a pixel in an image, and then perform a calculation before outputting a new signal. These outputs are fed, in a complex web, to the neurons in the next layer, and so on, until an overall output is produced. Plus, there is a process known as back-propagation that tweaks the calculations of individual neurons in a way that lets the network learn to produce a desired output.

The many layers in a deep network enable it to recognize things at different levels of abstraction. In a system designed to recognize dogs, for instance, the lower layers recognize simple things like outlines or color; higher layers recognize more complex stuff like fur or eyes; and the topmost layer identifies it all as a dog. The same approach can be applied, roughly speaking, to other inputs that lead a machine to teach itself: the sounds that make up words in speech, the letters and words that create sentences in text, or the steering-wheel movements required for driving.

We need more than a glimpse of AI’s thinking, however, and there is no easy solution. It is the interplay of calculations inside a deep neural network that is crucial to higher-level pattern recognition and complex decision-making, but those calculations are a quagmire of mathematical functions and variables. “If you had a very small neural network, you might be able to understand it,” Jaakkola says. “But once it becomes very large, and it has thousands of units per layer and maybe hundreds of layers, then it becomes quite un-understandable.”

David Gunning, a program manager at the Defense Advanced Research Projects Agency, is overseeing the aptly named Explainable Artificial Intelligence program.

“It’s often the nature of these machine-learning systems that they produce a lot of false alarms, so an intel analyst really needs extra help to understand why a recommendation was made,” Gunning says.

Just as many aspects of human behavior are impossible to explain in detail, perhaps it won’t be possible for AI to explain everything it does. “Even if somebody can give you a reasonable-sounding explanation [for his or her actions], it probably is incomplete, and the same could very well be true for AI,” says Clune, of the University of Wyoming. “It might just be part of the nature of intelligence that only part of it is exposed to rational explanation. Some of it is just instinctual, or subconscious, or inscrutable.”

A chapter of Dennett’s latest book, From Bacteria to Bach and Back, an encyclopedic treatise on consciousness, suggests that a natural part of the evolution of intelligence itself is the creation of systems capable of performing tasks their creators do not know how to do.

“If it can’t do better than us at explaining what it’s doing,” he says, “then don’t trust it.”

Focus upon Tasks, Not Jobs

Source: WorkMarket, Jan 2017

A new study released last week from the McKinsey Global Institute (MGI)

1. The study found that less than 5% of all occupations could be fully replaced by automation.

2. At least 60% of occupations have about 30% of tasks that could be automated, ranging from manufacturing to data processing to data collection.

3. 51% of job activities could be automated with today’s “currently demonstrated” technology.

By allowing technology to automate some of those mundane tasks, workers would then be freed to focus on tasks that are not easily replicated by automation such as decision making, creative tasks, and managing others. Automating these more routine activities will help businesses improve their performance by reducing manual error, as well as improving the speed and quality of how these tasks are completed.


Rogue One Leading into A New Hope

Math Model of Innovation

Source: MIT Technology Review, Jan 2017

the first mathematical model that accurately reproduces the patterns that innovations follow. The work opens the way to a new approach to the study of innovation, of what is possible and how this follows from what already exists.

The adjacent possible is all those things—ideas, words, songs, molecules, genomes, technologies and so on—that are one step away from what actually exists. It connects the actual realization of a particular phenomenon and the space of unexplored possibilities.

But this idea is hard to model for an important reason. The space of unexplored possibilities includes all kinds of things that are easily imagined and expected but it also includes things that are entirely unexpected and hard to imagine. And while the former is tricky to model, the latter has appeared close to impossible.

each innovation changes the landscape of future possibilities. So at every instant, the space of unexplored possibilities—the adjacent possible—is changing.

“Though the creative power of the adjacent possible is widely appreciated at an anecdotal level, its importance in the scientific literature is, in our opinion, underestimated,” say Loreto and co.

even with all this complexity, innovation seems to follow predictable and easily measured patterns that have become known as “laws” because of their ubiquity. One of these is Heaps’ law, which states that the number of new things increases at a rate that is sublinear. In other words, it is governed by a power law of the form V(n) = knβ where β is between 0 and 1.

Words are often thought of as a kind of innovation, and language is constantly evolving as new words appear and old words die out.

Given a corpus of words of size n, the number of distinct words V(n) is proportional to n raised to the β power. In collections of real words, β turns out to be between 0.4 and 0.6.

Another well-known statistical pattern in innovation is Zipf’s law, which describes how the frequency of an innovation is related to its popularity. For example, in a corpus of words, the most frequent word occurs about twice as often as the second most frequent word, three times as frequently as the third most frequent word, and so on. In English, the most frequent word is “the” which accounts for about 7 percent of all words, followed by “of” which accounts for about 3.5 percent of all words, followed by “and,” and so on.

This frequency distribution is Zipf’s law and it crops up in a wide range of circumstances, such as the way edits appear on Wikipedia, how we listen to new songs online, and so on.

They begin with a well-known mathematical sand box called Polya’s Urn. It starts with an urn filled with balls of different colors. A ball is withdrawn at random, inspected and placed back in the urn with a number of other balls of the same color, thereby increasing the likelihood that this color will be selected in future.

This is a model that mathematicians use to explore rich-get-richer effects and the emergence of power laws. So it is a good starting point for a model of innovation. However, it does not naturally produce the sublinear growth that Heaps’ law predicts.

That’s because the Polya urn model allows for all the expected consequences of innovation (of discovering a certain color) but does not account for all the unexpected consequences of how an innovation influences the adjacent possible.

So Loreto, Strogatz, and co have modified Polya’s urn model to account for the possibility that discovering a new color in the urn can trigger entirely unexpected consequences. They call this model “Polya’s urn with innovation triggering.”

The exercise starts with an urn filled with colored balls. A ball is withdrawn at random, examined, and replaced in the urn.

If this color has been seen before, a number of other balls of the same color are also placed in the urn. But if the color is new—it has never been seen before in this exercise—then a number of balls of entirely new colors are added to the urn.

Loreto and co then calculate how the number of new colors picked from the urn, and their frequency distribution, changes over time. The result is that the model reproduces Heaps’ and Zipf’s Laws as they appear in the real world—a mathematical first. “The model of Polya’s urn with innovation triggering, presents for the first time a satisfactory first-principle based way of reproducing empirical observations,” say Loreto and co.

The team has also shown that its model predicts how innovations appear in the real world. The model accurately predicts how edit events occur on Wikipedia pages, the emergence of tags in social annotation systems, the sequence of words in texts, and how humans discover new songs in online music catalogues.

Interestingly, these systems involve two different forms of discovery. On the one hand, there are things that already exist but are new to the individual who finds them, such as online songs; and on the other are things that never existed before and are entirely new to the world, such as edits on Wikipedia.

Loreto and co call the former novelties—they are new to an individual—and the latter innovations—they are new to the world.

Curiously, the same model accounts for both phenomenon. It seems that the pattern behind the way we discover novelties—new songs, books, etc.—is the same as the pattern behind the way innovations emerge from the adjacent possible.

That raises some interesting questions, not least of which is why this should be. But it also opens an entirely new way to think about innovation and the triggering events that lead to new things. “These results provide a starting point for a deeper understanding of the adjacent possible and the different nature of triggering events that are likely to be important in the investigation of biological, linguistic, cultural, and technological evolution,” say Loreto and co.

The MIT mascot (Beaver) Visits LA

… Universal Studios





Collaboration Leads to Incremental Innovation

Source: New England Complex Systems Institute, date indeterminate

Current collaborative design approaches are as a result typically characterized by heavy reliance on expensive and time-consuming processes, poor incorporation of some important design concerns (typically later life-cycle issues such as environmental impact), as well as reduced creativity due to the tendency to incrementally modify known successful designs rather than explore radically different and potentially superior ones.