Download news summary (44) - Quest Group`s Blog

https://www.bloomberg.com/news/articles/2017-03-02/ai-scientists-gather-to-plot-doomsdayscenarios-and-solutions
AI Scientists Gather to Plot Doomsday
Scenarios (and Solutions)
Researchers, cyber-security experts and policy wonks ask themselves: What could possibly go wrong?
by
Dina Bass
@dinabassMore stories by Dina BassMarch 02, 2017 6:00 AM
Artificial intelligence boosters predict a brave new world of flying cars and cancer cures.
Detractors worry about a future where humans are enslaved to an evil race of robot
overlords. Veteran AI scientist Eric Horvitz and Doomsday Clock guru Lawrence
Krauss, seeking a middle ground, gathered a group of experts in the Arizona desert to
discuss the worst that could possibly happen -- and how to stop it.
Their workshop took place last weekend at Arizona State University with funding from
Tesla Inc. co-founder Elon Musk and Skype co-founder Jaan Tallinn. Officially dubbed
"Envisioning and Addressing Adverse AI Outcomes," it was a kind of AI doomsday
games that organized some 40 scientists, cyber-security experts and policy wonks into
groups of attackers -- the red team -- and defenders -- blue team -- playing out AI-gonevery-wrong scenarios, ranging from stock-market manipulation to global warfare.
Horvitz is optimistic -- a good thing because machine intelligence is his life's work -- but
some other, more dystopian-minded backers of the project seemed to find his outlook
too positive when plans for this event started about two years ago, said Krauss, a
theoretical physicist who directs ASU's Origins Project, the program running the
workshop. Yet Horvitz said that for these technologies to move forward successfully and
to earn broad public confidence, all concerns must be fully aired and addressed.
"There is huge potential for AI to transform so many aspects of our society in so many
ways. At the same time, there are rough edges and potential downsides, like any
technology," said Horvitz, managing director of Microsoft's Research Lab in Redmond,
Washington. ``To maximally gain from the upside we also have to think through
possible outcomes in more detail than we have before and think about how we’d
deal with them."
Participants were given "homework" to submit entries for worst-case scenarios. They
had to be realistic -- based on current technologies or those that appear possible -- and
five to 25 years in the future. The entrants with the "winning" nightmares were
chosen to lead the panels, which featured about four experts on each of the two
teams to discuss the attack and how to prevent it.
Blue team, including Launchbury, Fisher and Krauss, in the War and Peace scenario
Tessa Eztioni, Origins Project at ASU
Turns out many of these researchers can match science-fiction writers Arthur C. Clarke
and Philip K. Dick for dystopian visions. In many cases, little imagination was required -scenarios like technology being used to sway elections or new cyber attacks using
AI are being seen in the real world, or are at least technically possible. Horvitz
cited research that shows how to alter the way a self-driving car sees traffic signs so
that the vehicle misreads a "stop" sign as "yield.''
The possibility of intelligent, automated cyber attacks is the one that most
worries John Launchbury, who directs one of the offices at the U.S.'s Defense
Advanced Research Projects Agency, and Kathleen Fisher, chairwoman of the
computer science department at Tufts University, who led that session. What
happens if someone constructs a cyber weapon designed to hide itself and evade all
attempts to dismantle it? Now imagine it spreads beyond its intended target to the
broader internet. Think Stuxnet, the computer virus created to attack the Iranian nuclear
program that got out in the wild, but stealthier and more autonomous.
"We're talking about malware on steroids that is AI-enabled," said Fisher, who is an
expert in programming languages. Fisher presented her scenario under a slide bearing
the words "What could possibly go wrong?" which could have also served as a tagline
for the whole event.
How did the defending blue team fare on that one? Not well, said Launchbury. They
argued that advanced AI needed for an attack would require a lot of computing power
and communication, so it would be easier to detect. But the red team felt that it would
be easy to hide behind innocuous activities, Fisher said. For example, attackers could
get innocent users to play an addictive video game to cover up their work.
To prevent a stock-market manipulation scenario dreamed up by University of Michigan
computer science professor Michael Wellman, blue team members suggested treating
attackers like malware by trying to recognize them via a database on known types of
hacks. Wellman, who has been in AI for more than 30 years and calls himself an oldtimer on the subject, said that approach could be useful in finance.
…
http://www.infoworld.com/article/3165267/advanced-persistent-threats/ai-isnt
-for-the-good-guys-alone-anymore.html
AI isn't for the good guys alone anymore
Criminals are beginning to use artificial intelligence
and machine learning to get around cyberdefenses
By Maria Korolov
| Follow
Contributing Writer, CSO | Feb 4, 2017
Last summer at the Black Hat cybersecurity conference, the DARPA Cyber Grand
Challenge pitted automated systems against one another, trying to find
weaknesses in the others’ code and exploit them.
“This is a great example of how easily machines can find and exploit new vulnerabilities,
something we’ll likely see increase and become more sophisticated over time,” said
David Gibson, vice president of strategy and market development at Varonis Systems.
His company hasn’t seen any examples of hackers leveraging artificial intelligence
technology or machine learning, but nobody adopts new technologies faster than the sin
and hacking industries, he said.
“So it’s safe to assume that hackers are already using AI for their evil purposes,” he
said.
The genie is out of the bottle
“It has never been easier for white hats and black hats to obtain and learn the tools of
the machine learning trade,” said Don Maclean, chief cybersecurity technologist at DLT
Solutions. “Software is readily available at little or no cost, and machine learning
tutorials are just as easy to obtain.”
Take, for example, image recognition.
It was once considered a key focus of artificial intelligence research. Today, tools such
as optical character recognition are so widely available and commonly used that they’re
not even considered to be artificial intelligence anymore, said Shuman Ghosemajumder,
CTO at Shape Security.
“People don’t see them as having the same type of magic as it has before,” he said.
“Artificial intelligence is always what’s coming in the future, as opposed to what we have
right now.”
Today, for example, computer vision is good enough to allow self-driving cars to
navigate busy streets.
And image recognition is also good enough to solve the puzzles routinely presented to
website users to prove that they are human, he added.
For example, last spring, Vinay Shet, the product manager for Google’s Captcha
team, told Google I/O conference attendees that in 2014, they had a distorted text
Captcha that only 33 percent of humans could solve. By comparison, the state-ofthe-art OCR systems at the time could already solve it with 99.8 percent accuracy.
The criminals are already using image recognition technology, in combination with
“Captcha farms,” to bypass this security measure, said Ghosemajumder. The popular
Sentry MBA credential stuffing tool has it built right in, he added.
So far, he said, he hasn’t seen any publicly available tool kits based on machine
learning that are designed to bypass other security mechanisms.
But there are indirect indicators that criminals are starting to use this technology, he
added.
For example, companies already know that if there’s an unnaturally large amount of
traffic from one IP address, that there’s a high chance it’s malicious, so criminals use
botnets to bypass those filters, and the defenders look for more subtle indications that
the traffic is automated and not human, he said.
They can’t just add in more randomness since human behavior is not actually random,
he said. Spotting subtle patterns in large amount of data is exactly what machine
learning is good at—and what the criminals need to do in order to effectively mimic
human behavior.
Smarter email scams
According to the McAfee Labs 2017 Threats Predictions report, cyber-criminals are
already using machine learning to target victims for Business Email Compromise
scams, which have been escalating since early 2015.
…
http://metro.co.uk/2017/03/02/computer-beat-11-pro-poker-players-using-intuition-6482908/
Computer beat 11 pro poker players using
‘intuition’
Ashitha Nagesh for Metro.co.ukThursday 2 Mar 2017 7:00 pm
It used intuition (Picture: Shutterstock)
Don’t be alarmed, but a computer just beat 11 professional poker players in a 3,000 hand
match of the incredibly complex Texas Hold’em, using ‘intuition’.
DeepStack is the first computer programme to outplay humans at the game, which it played over
a period of four weeks.
Card players tend to rely on microexpressions – tiny subconscious flashes of emotion across
the face that last no longer than a quarter of a second.
The computer, which was developed by scientists at the University of Alberta, Charles
University in Prague and Czech Technical University, honed its ‘intuition’ through deep
learning – allowing it to reassess its strategy with each move.
DeepStack has developed intuition (Picture: YouTube)
Professor Michael Bowling, from Alberta, said: ‘Poker has been a longstanding challenge
problem in artificial intelligence.
‘It is the quintessential game of imperfect information in the sense that the players don’t have the
same information or share the same perspective while they’re playing.’
He added that DeepStack’s winning streak would have major implications for the world outside
poker.
‘Think of any real world problem,’ Prof Bowling said. ‘We all have a slightly different
perspective of what’s going on – much like each player only knowing their own cards in a game
of poker.’
Heads-up no-limit hold’em apparently has more unique situations than there are atoms in
the universe – largely because players are able to wager different amounts.
DeepStack acts at a human speed – with an average of just three seconds of ‘thinking’ time – and
runs on a simple gaming laptop with an Nvidia graphics processing unit.
The researchers’ full results have been published in the journal Science
…
https://www.washingtonpost.com/news/animalia/wp/2016/04/13/octopus-slips-out-of-aquariumtank-crawls-across-floor-escapes-down-pipe-to-ocean/?utm_term=.e75efec070ea
Octopus slips out of aquarium tank, crawls
across floor, escapes down pipe to ocean
By Karin Brulliard By Karin Brulliard
Animalia
April 13, 2016
Inky the octopus has won international fame after his stealthy escape from the National Aquarium of
New Zealand. Here's how Inky spent his time at the aquarium before breaking free. (Facebook/National
Aquarium of New Zealand)
Inky the octopus didn’t even try to cover his tracks.
By the time the staff at New Zealand’s National Aquarium noticed that he
was missing, telltale suction cup prints were the main clue to an easily solved mystery.
Inky had said see ya to his tank-mate, slipped through a gap left by maintenance workers at
the top of his enclosure and, as evidenced by the tracks, made his way across the floor to a
six-inch-wide drain. He squeezed his football-sized body in — octopuses are very malleable,
aquarium manager Rob Yarrall told the New Zealand website Stuff — and made a break for the
Pacific.
“He managed to make his way to one of the drain holes that go back to the ocean. And off he
went,” Yarrall told Radio New Zealand. “And he didn’t even leave us a message.”
The cephalopod version of “Shawshank Redemption” took place three months ago, but it only
became public Tuesday. Inky, who already had some local renown in the coastal city of Napier,
quickly became a global celebrity cheered on by strangers.
Inky had resided at the aquarium since 2014, when he was taken in after being caught in a
crayfish pot, his body scarred and his arms injured. The octopus’s name was chosen from
nominations submitted to a contest run by the Napier City Council.
Kerry Hewitt, the aquarium’s curator of exhibits, said at the time that Inky was “getting used to
being at the aquarium” but added that staff would “have to keep Inky amused or he will get
bored.”
Guess that happened.
This isn’t the first time a captive octopus decided to take matters into its own hands — er,
tentacles. In 2009, after a two-spotted octopus at the Santa Monica Pier Aquarium in
California took apart a water recycling valve, directed a tube to shoot water out of the tank for 10
hours and caused a massive flood, Scientific American asked octopus expert Jennifer
Mather about the animals’ intelligence and previous such hijinks at aquariums
…
http://www.mirror.co.uk/tech/googles-artificial-intelligence-can-diagnose-9975987
Google's artificial intelligence can
diagnose cancer faster than human
doctors
The system is able to scan samples to determine whether or not tissues are
cancerous
BYJEFF PARSONS


16:00, 6 MAR 2017
UPDATED10:57, 7 MAR 2017
Top view of a female breast with a tumor (Photo: Getty Images)
Making the decision on whether or not a patient has cancer usually involves trained professionals
meticulously scanning tissue samples over weeks and months.
But an artificial intelligence (AI) program owned by Alphabet, Google's parent company, may be able to
do it much, much faster.
Google is working hard to tell the difference between healthy and cancerous tissue as well as discover if
metastasis has occured.
"Metastasis detection is currently performed by pathologists reviewing large expanses of biological
tissues. This process is labour intensive and error-prone," explained Google in a white paper outlining
the study.
Prostate cancer (Photo: Getty)
"We present a framework to automatically detect and localise tumours as small as 100 ×100 pixels in
gigapixel microscopy images sized 100,000×100,000 pixels.
"Our method leverages a convolutional neural network (CNN) architecture and obtains state-of-theart results on the Camelyon16 dataset in the challenging lesion-level tumour detection task."
Such high-level image recognition was first developed for Google's driverless car program, in order to
help the vehicles scan for road obstructions.
Now the company has adapted it for the medical field and says it's more accurate than regular human
doctors:
(Photo: Getty Images)
"At 8 false positives per image, we detect 92.4% of the tumours, relative to 82.7% by the previous
best automated approach. For comparison, a human pathologist attempting exhaustive search
achieved 73.2% sensitivity." ** but how many FPs by the human? **
…
http://languagelog.ldc.upenn.edu/nll/?p=17970
It's not easy seeing green
March 2, 2015 @ 3:00 pm · Filed by Mark Liberman
The whole dress that melted the internet thing has brought back a curious example of semidemi-science about a Namibian tribe that can't distinguish green and blue, but does
differentiate kinds of green that look just the same to us Westerners. This story has been
floating around the internets for several years, in places like the BBC and the New York Times
and BoingBoing and RadioLab, and it presents an impressive-seeming demonstration of the
power of language to shape our perception of the world. But on closer inspection, the
evidence seems to melt away, and the impressive experience seems to be wildly overinterpreted or even completely invented.
I caught the resurrection of this idea in Kevin Loria's article "No one could see the color blue until
modern times", Business Insider 2/27/2015, which references a RadioLab episode on Colors that
featured those remarkable Namibians. Loria uses them to focus on that always-popular question
"do you really see something if you don't have a word for it?" *** stupid statement –
philosophers – clearly I can change the hue / saturation / brightness so there is a JND but not a
different label – is the fact that I can discern the difference a word for it? **
Here's the relevant segment of Loria's piece:
A researcher named Jules Davidoff traveled to Namibia to investigate this, where he conducted
an experiment with the Himba tribe, which speaks a language that has no word for blue
or distinction between blue and green.
When shown a circle with 11 green squares and one blue, they could not pick out which one was
different from the others — or those who could see a difference took much longer and made
more mistakes than would make sense to us, who can clearly spot the blue square.
But the Himba have more words for types of green than we do in English. When looking at a
circle of green squares with only one slightly different shade, they could immediately spot the
different one.
Can you?
For most of us, that's harder.
…
https://www.technologyreview.com/s/603745/how-a-human-machine-mind-meld-could-make-robotssmarter/?set=603773
Intelligent Machines
How a Human-Machine Mind-Meld Could Make
Robots Smarter
Kindred AI is teaching robots new tasks using human virtual-reality “pilots.” The
ultimate goal is to create a new kind of artificial intelligence.


by Will Knight
March 2, 2017
A secretive Canadian startup called Kindred AI is teaching robots how to
perform difficult dexterous tasks at superhuman speeds by pairing them
with human “pilots” wearing virtual-reality headsets and holding motiontracking controllers.
The technology offers a fascinating glimpse of how humans might work in
synchronization with machines in the future, and it shows how tapping into
human capabilities might amplify the capabilities of automated systems. For all
the worry over robots and artificial intelligence eliminating jobs, there are
plenty of things that machines still cannot do. The company demonstrated the
hardware to MIT Technology Review last week, and says it plans to launch a
product aimed at retailers in the coming months. The long-term ambitions are
far grander. Kindred hopes that this human-assisted learning will foster a
fundamentally new and more powerful kind of artificial intelligence.
Kindred was created by several people from D-Wave, a quantum computing
company based in Burnaby, Canada. Kindred is currently testing conventional
industrial robot arms capable of grasping and placing objects that can be
awkward to handle, like small items of clothing, more quickly and reliably than
would normally be possible. The arms do this by occasionally asking for help
from a team of humans, who use virtual-reality hardware to view the
challenge and temporarily take control of an arm.
“A pilot can see, hear, and feel what the robot is seeing, hearing, and feeling.
When the pilot acts, those actions move the robot,” says Geordie Rose, who is a
cofounder and the CEO of Kindred, and who previously cofounded D-Wave.
“This allows us to show robots how to act like people. Humans aren't the fastest
or best at all aspects of robot control, like putting things in specific locations,
but humans are still best at making sense of tricky or unforeseen situations.”
** hole is the unexpected Query **
Kindred’s system uses several machine-learning algorithms, and tries to
predict whether one of these would provide the desired outcome, such as
grasping an item. *** note the cognitive flexibility ** If none seems to offer a
high probability of success, it calls for human assistance. Most importantly, the
algorithms learn from the actions of a human controller. To achieve this, the
company uses a form of reinforcement learning, an approach that involves
experimentation and strengthening behavior that leads to a particular goal
(see “10 Breakthrough Technologies 2017: Reinforcement Learning”).
Rose says the system can grasp small items of clothing about twice as fast as a
person working on his own can, while a robot working independently would be
too unreliable to deploy. One person can also operate several robots at once.
Rose adds that Kindred is exploring all sorts of human-in-the loop systems,
from ones where a person simply clicks on an image to show a robot where to
grasp something, to full-body exoskeletons that provide control over a
humanoid robot. He says that pilots usually learn how to control a remote
robotic system effectively. “When you’re using the control apparatus, at first it’s
very frustrating, but people’s minds are very plastic, and you adjust,” says Rose.
The technical inspiration for the technology comes from Suzanne Gildert, who
was previously a senior researcher at D-Wave, and who is Kindred’s chief
scientific officer. The company has been operating in stealth mode for several
years, but drew attention when details of a patent filed by Gildert surfaced
online. The patent describes a scheme for combining different tele-operation
systems with machine learning. Indeed, Kindred’s vision for its technology
seems to extend well beyond building robots more skilled at sorting.
“The idea was if you could do that for long enough, and if you had some sort of
AI system in the background learning, that maybe you could try out many
different AI models and see which ones trained better,” *** how to get to
cognitive flexibility ** Gildert says. “Eventually, my thought was, if you can
have a human demonstrating anything via a robot, then there’s no reason that
robot couldn’t learn to be very humanlike.”
…
https://www.nytimes.com/2017/03/04/world/asia/north-korea-missile-program-sabotage.html?_r=0
Trump Inherits a Secret Cyberwar Against
North Korean Missiles
By DAVID E. SANGER and WILLIAM J. BROADMARCH 4, 2017
An image distributed by the North Korean government showing the country’s leader, Kim Jong-un,
visiting a missile test center in North Pyongan Province. Analysts say the pair of engines he is standing in
front of could power an intercontinental ballistic missile. Credit Korean Central News Agency, via
Reuters
WASHINGTON — Three years ago, President Barack Obama ordered Pentagon officials to step
up their cyber and electronic strikes against North Korea’s missile program in hopes of
sabotaging test launches in their opening seconds.
Soon a large number of the North’s military rockets began to explode, veer off course,
disintegrate in midair and plunge into the sea. Advocates of such efforts say they believe that
targeted attacks have given American antimissile defenses a new edge and delayed by several
years the day when North Korea will be able to threaten American cities with nuclear weapons
launched atop intercontinental ballistic missiles.
But other experts have grown increasingly skeptical of the new approach, arguing that
manufacturing errors, disgruntled insiders and sheer incompetence can also send missiles awry.
Over the past eight months, they note, the North has managed to successfully launch three
medium-range rockets. And Kim Jong-un, the North Korean leader, now claims his country is in
“the final stage in preparations” for the inaugural test of his intercontinental missiles — perhaps
a bluff, perhaps not.
An examination of the Pentagon’s disruption effort, based on interviews with officials of the
Obama and Trump administrations as well as a review of extensive but obscure public records,
found that the United States still does not have the ability to effectively counter the North Korean
nuclear and missile programs. Those threats are far more resilient than many experts thought,
The New York Times’s reporting found, and pose such a danger that Mr. Obama, as he left
office, warned President Trump they were likely to be the most urgent problem he would
confront.
Pakistan
…
http://www.sciencemag.org/news/2017/03/brainlike-computers-are-black-box-scientists-are-finallypeering-inside?utm_campaign=news_daily_2017-03-07&et_rid=54802259&et_cid=1203472
A representation of a neural network.
Akritasa/Wikimedia Commons
Brainlike computers are a black box.
Scientists are finally peering inside
By Jackie SnowMar. 7, 2017 , 3:15 PM
Last month, Facebook announced software that could simply look at a photo and tell,
for example, whether it was a picture of a cat or a dog. A related program
identifies cancerous skin lesions as well as trained dermatologists can. Both
technologies are based on neural networks, sophisticated computer algorithms at the
cutting edge of artificial intelligence (AI)—but even their developers aren’t sure
exactly how they work. Now, researchers have found a way to "look" at neural
networks in action and see how they draw conclusions.
Neural networks, also called neural nets, are loosely based on the brain’s use of layers
of neurons working together. Like the human brain, they aren't hard-wired to produce a
specific result—they “learn” on training sets of data, making and reinforcing connections
between multiple inputs. A neural net might have a layer of neurons that look at pixels
and a layer that looks at edges, like the outline of a person against a background. After
being trained on thousands or millions of data points, a neural network algorithm
will come up with its own rules on how to process new data. But it's unclear what
the algorithm is using from those data to come to its conclusions.
“Neural nets are fascinating mathematical models,” says Wojciech Samek, a researcher
at Fraunhofer Institute for Telecommunications at the Heinrich Hertz Institute in Berlin.
“They outperform classical methods in many fields, but are often used in a black box
manner.”
In an attempt to unlock this black box, Samek and his colleagues created software that
can go through such networks backward in order to see where a certain decision was
made, and how strongly this decision influenced the results. Their method, which they
will describe this month at the Centre of Office Automation and Information Technology
and Telecommunication conference in Hanover, Germany, enables researchers to
measure how much individual inputs, like pixels of an image, contribute to the
overall conclusion. Pixels and areas are then given a numerical score for their
importance. With that information, researchers can create visualizations that impose a
mask over the image. The mask is most bright where the pixels are important and
darkest in regions that have little or no effect on the neural net’s output.
For example, the software was used on two neural nets trained to recognize horses.
One neural net was using the body shape to determine whether it was horse. The other,
however, was looking at copyright symbols on the images that were associated with
horse association websites.
This work could improve neural networks, Samek suggests. That includes helping
reduce the amount of data needed, one of the biggest problems in
AI development, by focusing in on what the neural nets need. It could also help
investigate errors when they occur in results, like misclassifying objects in an image.
…
http://www.recode.net/2017/3/1/14771330/boston-dynamics-new-nightmare-robot
Roboticists say Boston Dynamic’s new
nightmare robot is unlike anything ever
seen before
Handle can move with confidence — and carry 100
pounds.
BY APRIL GLASER@APRILASER MAR 1, 2017, 4:19PM EST

TWEET

SHARE

LINKEDIN
Boston Dynamics, the Alphabet-owned robotics company, unveiled a new robot this
week that robotics experts say is unlike anything they’ve ever seen before.
The massive legged, wheeled machine is called Handle. Marc Raibert, the CEO of
Boston Dynamics, called it “nightmare inducing.” (Video of Handle was first seen in
January, when venture capitalist Steve Jurvetson posted a YouTube video of Raibert
introducing the new creation at a conference.)
“It’s very impressive,” said Vikash Kumar, a robotics researcher at the University of
Washington. “Nothing like this has been shown before.”
What sets Handle apart is its ability to move with confidence and an understanding of
what its body is capable of — it’s remarkably lifelike. Robots are usually stiff, slow
and careful. They’re programmed to be cautious in their movements.
“This robot is using the momentum of its body. These dynamic movements are not
what we know robots to be able to do,” said Kumar. “There are other robots that have
the hardware capabilities of doing something similar, but on the algorithmic side,
we’re not at the point where we can really leverage those capabilities.”
But Handle appears to be getting close.
“What's most impressive is how dynamic and powerful it is, while still being stable
and, mostly, in control,” said Siddhartha Srinivasa, a professor of computer
engineering at Carnegie Mellon, who specializes in robotic manipulation. Srinivasa is
moving to teach at the University of Washington next year.
The robot may one day be used in warehouses, as illustrated by the video released by
Boston Dynamics. It can jump over hurdles and land on its wheeled feet, lift a single
leg while moving, stroll in the snow and go down stairs.
Handle can also carry a payload of 100 pounds, which roboticists say is impressive for
its size and shape.
“The field at the moment is trying build robots to take care of really small, low-weight
objects,” said Kumar. “To deal with an object that is comparable with its own weight
changes the whole ballgame altogether.”
Robots in warehouses now often act like mechanical shelving, carrying items from
one part of the facility to another, saving humans time and future back problems.
Kiva’s robots — from the warehouse robotics company that Amazon bought in 2014,
for example — followed barcoded stickers on the floor.
To be sure, Handle probably has a long way to go before taking a step on a job site.
“It is important to remember that there are many perception and control challenges in
getting such a machine to operate reliably in a real warehouse,” said Ken Goldberg, a
robotics professor at Berkeley.
It’s not that Handle is necessarily any more dangerous than other industrial robots,
which have killed people who got in their way — like what happened when an
engineer at a Volkswagen plant in Germanywas crushed to death in 2015 by a
stationary factory robot.
Still, industrial robots have stringent safety standards and often have kill zones that
humans are supposed to avoid when the machine is operating. And Handle, with all its
confidence and dynamic movements, will have to be explicit about its intent as it
operates around humans.
“It’s like a car with a blinker sign,” said Kumar. “We need to create ways to
communicate so that humans can anticipate what a robot is trying to do and not be
surprised by their action.”
It’s also not clear how Boston Dynamics plans to prepare or market its robot for the
real world.
Google bought the robotics lab in 2013 under the direction of Android founder Andy
Rubin, who had hired around 300 robotics engineers with Google.
But Rubin left Google in 2014 to start his own hardware incubator. The search giant
put Boston Dynamics up for sale last year after public relations at Alphabet expressed
concerns that the nightmarish robots they made — like the two-legged
humanoid Atlas and its massive robotic dog named Spot — were “terrifying” and
“ready to take human jobs.”
Though Google has — puzzlingly — yet to find a buyer for Boston Dynamics, the
team clearly hasn’t stopped moving.
“This is one of the most remarkable robots I have seen in a long time,” said Srinivasa.
“Boston Dynamics is truly the Apple of robotics, when it comes to tightly integrating
hardware design with software and artificial intelligence.
The future is exciting.
…
http://www.huffingtonpost.com/mark-changizi-phd/perceiving-colors-differently_b_988244.html
Understanding Color Perception: Is Your
‘Red’ the Same as My ‘Red?’
10/02/2011 05:21 am ET | Updated Dec 02, 2011
Mark Changizi, Ph.D. Neuroscientist, Author of ‘Harnessed’ & ‘Vision Revolution’
How do we know that your “red” looks ** this is a quale statement – what you
perceive in your consciousness is what I perceive ** the same as my “red”? For
all we know, your “red” looks like my “blue.” In fact, for all we know your “red”
looks nothing like any of my colors at all! If colors are just internal labels ** labels
here is not meant to imply the word – it is meant to describe the representation
internal in a vocabulary of conscious thought ** , then as long as everything gets
labeled, why should your brain and my brain use the same labels? ** as long as we
can align – why would they have to be the same – keep in mind the Kaku view –
philosophers waste our time with these thought problems – he speaks to the ‘what is
life concern that has disappeared **
Richard Dawkins wrote a nice little piece on color, and along the way he asked these
questions.
He also noted that not only can color labels differ in your and my brain, but perhaps
the same color labels could be used in non-visual modalities of other animals.
Bats, he notes, use audition for their spatial sense, and perhaps furry moths are
heard as red, and leathery locusts as blue. Similarly, rhinoceroses may use
olfaction for their spatial sense, and could perceive water as orange and rival
male markings as gray. ** we’ve discussed these issues many times – but I think
our recent discussions on representations can lock down what we can say about
the characteristics of our respective representations **
However, I would suggest that most discussions of rearrangements of color qualia (a
quality or property as perceived or experienced by a person) severely underestimate
how much structure comes along with our color perceptions. *** this is exactly what
we are suggesting in QuEST – to comprehend qualia – to capture how they are related
and how they can interact is equivalent to what he calls structure ** Once one more
fully appreciates the degree to which color qualia are linked to one another and
to non-color qualia, it becomes much less plausible to single color qualia out as
especially permutable. ** we of course in QuEST make all qualia and even ‘self’
just different aspects of the working memory – color is just one quale – as is time
and self and … every word was developed to communicate to another agent some
aspect of your qualia representation **
Few of us, for example, would find it plausible to imagine that others might perceive
music differently, e.g., with pitch and loudness swapped, so that melody to them
sounds like loudness modulations to me, and vice versa. Few of us would find it
plausible to imagine that some other brain might perceive “up” (in one’s visual field)
and “down” as reversed. *** buy yet we have inverting googles that allow us to do
this experiment ** And it is not quite so compelling to imagine that one might
perceive the depth of something as the timbre of an instrument, and vice versa. And so
on. ** he makes a big deal about inverted color – I’m not sure why – maybe because
it is what is easy to describe to folks – I’ve never taken this argument to be unique to
color **
Unlike color qualia, most alternative possible qualia rearrangements do not seem
plausible. Why is that? Why is color the butt of nearly all the “inverted-spectra”
arguments?
The difference is that these other qualia seem to be more than just mere labels
that can be permuted willy nilly. Instead, these other qualia are deeply
interconnected with hosts of other aspects of our perceptions. They are part of a
complex structured network of qualia, and permuting just one small part of the
network destroys the original shape and structure of the network — and when the
network’s shape and structure is radically changed, the original meanings of the
perceptions (and the qualia) within it are obliterated. ** everything in this paragraph
is exactly where we are focusing our representation / objective function argument –
…
https://www.technologyreview.com/s/603797/toyota-tests-backseat-driver-software-that-could-takecontrol-in-dangerous-moments/?set=603799
Intelligent Machines
Toyota Tests Backseat Driver Software That Could
Take Control in Dangerous Moments
Cameras that watch where a driver is looking allow a car to judge when they are likely
to miss a dangerous situation.


by Tom Simonite
March 7, 2017
This
modified Lexus is used by Toyota to test autonomous driving software.
Making a turn across oncoming traffic is one of the most dangerous
maneuvers drivers undertake every day. Researchers at Toyota think it’s
one of the situations in which a software guardian angel built into your car
could save lives.
In trials at private testing grounds in the U.S., left turns are one of the first
scenarios Toyota has used to test the concept of a system it has dubbed
“Guardian,” which judges whether a human is about to make a dangerous
mistake.
Radar and other sensors on the outside of the car monitor what’s happening
around the vehicle, while cameras inside track the driver’s head movements and
gaze. Software uses the sensor data to estimate when a person needs help
spotting or avoiding a hazardous situation, or can be left alone.
So far Toyota is just testing the ability of software to understand the hazards
around a car and whether a person has spotted them, but the company plans
to eventually make Guardian capable of taking action if a person doesn’t look
ready to do so themselves.
“Imagine going through an intersection and you’re going to get T-boned—the
right thing for the car to do is accelerate you out of it,” says Ryan Eustice, VP of
autonomous driving at Toyota Research Institute, established in 2015 to work
on robotics and automated driving (see “Toyota’s Billion Dollar Bet”). The
group first said it would start developing Guardian last year (see “Toyota
Joins the Race for Self-Driving Cars with an Invisible Co-Pilot”).
Eustice argues that the Guardian effort could have a widespread impact on
public safety earlier than cars that fully remove driving duties from humans.
Toyota is working on such technology, along with competitors like
Alphabet, Ford, and Uber. But despite high-profile testing programs on public
roads, Eustice and his counterparts at other companies say that truly driverless
vehicles are still some years from serving the public, and will initially be limited
to certain routes or locales.
“We see an opportunity to deploy it sooner and more widely,” says Eustice of
the backseat driver approach. That’s because unlike full autonomy it won’t
be reliant on hyper-detailed maps and could be easily packaged into a
conventional vehicle sold to consumers, he says. However, he declines to
predict how soon the Guardian might be ready for commercialization.
Steven Shladover, a researcher at the University of California, Berkeley, says
the claim that Guardian could save lives sooner than fully autonomous vehicles
makes sense. “If the driver has a 99 percent chance of detecting hazards and the
automation system also has a 99 percent chance of detecting hazards, that gives
the combination of the driver and system a 99.99 percent chance,” he says. “But
this is much simpler and easier than designing a fully automated system that
could reach that 99.99 percent level by itself.”
Getting the relationship between Guardian and humans right will be
critical, though. Any mistakes it makes such as intervening or sending a
warning when a person has correctly interpreted a situation would
undermine a person’s trust in the system, and could even lead to new kinds
of accidents, says Shladover.
…
https://www.nytimes.com/2017/03/07/world/europe/wikileaks-cia-hacking.html
WikiLeaks Releases Trove of Alleged C.I.A.
Hacking Documents
Leer en español
By SCOTT SHANE, MARK MAZZETTI and MATTHEW ROSENBERGMARCH 7, 2017
The C.I.A. headquarters in Langley, Va. If the WikiLeaks documents are authentic, the release would be a
serious blow to the C.I.A. Credit Jason Reed/Reuters
WASHINGTON — WikiLeaks on Tuesday released thousands of documents that it said
described sophisticated software tools used by the Central Intelligence Agency to break into
smartphones, computers and even Internet-connected televisions.
If the documents are authentic, as appeared likely at first review, the release would be the latest
coup for the anti-secrecy organization and a serious blow to the C.I.A., which maintains its own
hacking capabilities to be used for espionage.
The initial release, which WikiLeaks said was only the first part of the document collection,
included 7,818 web pages with 943 attachments, the group said. The entire archive of C.I.A.
material consists of several hundred million lines of computer code, it said.
Among other disclosures that, if confirmed, would rock the technology world, the WikiLeaks
release said that the C.I.A. and allied intelligence services had managed to bypass encryption on
popular phone and messaging services such as Signal, WhatsApp and Telegram. According to
the statement from WikiLeaks, government hackers can penetrate Android phones and collect
“audio and message traffic before encryption is applied.”
The source of the documents was not named. WikiLeaks said the documents, which it called
Vault 7, had been “circulated among former U.S. government hackers and contractors in an
unauthorized manner, one of whom has provided WikiLeaks with portions of the archive.”
WikiLeaks said the source, in a statement, set out policy questions that “urgently need to be
debated in public, including whether the C.I.A.’s hacking capabilities exceed its mandated
powers and the problem of public oversight of the agency.” The source, the group said, “wishes
to initiate a public debate about the security, creation, use, proliferation and democratic control
of cyberweapons.”
…