IoT Security - 90-Percent of Devices are NOT Secure

The explosion in the number of devices connected to the cloud come with a price.  That price is security.  Armando Lucrecio, a Senior Technical Program Manager with Amazon Connected Products addresses the issue in the Business School of AI’s WeeklyWed, talking about how some 90% of products in IoT are not secure.  Lucrecio outlines the myriad of reasons for this.  He makes it clear: Understanding and addressing the security issues will help a technology with great promise.

“Connected product” is probably a better term to use than the Internet of Things, or IoT.  Lucrecio likes the label because it better helps us understand how we are talking about  networking:  Security cameras; Alexa, Google Home and Apple’s Siri; Nest thermostats; home appliances, and a seemingly endless number of other possibilities.  There’s even word now that Amazon has a robotic dog that patrols your home and communicates with you on “the network.”

“It’s all about the network,” Lucrecio says. This connectivity allows for communication to occur, and with it, the extraction of data that enables us to make the decisions that benefit everyone involved.  If you control the network, you can can extract all the data and analyze it, providing the answers you need to better serve your customers.

The security issues crop up because of all the players involved.  Since there is no way that one company can build its own network, players rely on telecommunications companies.  Then there’s the product itself.  If you buy the hardware, those that made it can build in a back door for someone else’s access, or they can insert a trojan horse to later take control.  Then there are hackers who work to take control of either the product or device and the network for whatever reason, most likely for monetary gain.

For more on this important topic, watch this video of Armando Lucrecio's talk;

IoT Security - The Internet of Things

Evolution of a Robot

 Evolution of a Robot 

Autonomous Delivery Service Online

A “cute little robot” has graduated from college dorm burrito deliveries to providing security to women walking across campus in the dark.  Kiwibot is now “a robot that people can accept,” says David Rodriguez, the company’s Director of Strategy and Business Operations.  Mr. Rodriguez made the comments in a guest appearance online on WeekleyWed, a feature of the Business School of AI.

The surprising revelations come in an hour-long discussion that covered everything from how Kiwibot went from an incubator to a company expected to attract “billions of dollars” in funding.  Part of the reason for the company’s success, Rodriguez says, is that “small vehicles delivering burritos make a lot more sense than three-ton cars getting in the right-of-way.”  In other words, Kiwibot is not “threatening.”  

Sustainable Technology

“Technology is not sustainable unless it’s solving a real world problem,” Rodriguez says.  Here he addresses how Artificial Intelligence and Machine Learning are used to enable these autonomous vehicles to do what they do.  “It’s not an ‘over-techy’ product,” paving the way for “robot solutions that go beyond delivery.”  The robots are now getting requests to accompany women who want them for security reasons.

As exciting as it may sound, developing a self-driving robot that people “fall in love with,” Rodriguez says everyone in the company needs to be focused on solving a problem.  In this case, the focus is on reducing the cost of making deliveries.  In the discussion, Rodriguez talks about taking delivery costs from $10 down to three.  Now the business world is starting to take note, attracting interest from companies like Shopify, Ordermark, Chick-fil-A and others.  Kiwibot deployment stretches from Stanford to Berkley, Los Angeles to Miami, Detroit to Pittsburg and beyond.

The Actual Robot:

 

“We thought we were done.”  Rodriguez talks about surviving a disaster on the U.C. Berkeley campus.  One of the Kiwibots exploded, overheating because of a problem involving the use of four different lightbulbs in the small, autonomous vehicles.  To the shock of the company’s founders, distressed students held a candle light vigil on campus for the robot. People are falling in love with the little guy.  After four years of development and then deployment, Kiwibots now have enough momentum that the company believes they are on the way to overcoming some major hurdles.

While competition to be at the forefront of the multi-billion-dollar delivery business is fierce, Kiwibot’s founders maintain focus on solutions.  This has ranged from overcoming things like surviving cold weather to building a navigation system that avoids dangerous streets.  “If we find a neighborhood where people have a really bad attitude, we don’t go there.”  Kiwibots may face hurdles and even some roadblocks, but they look to be heading down the road of widespread acceptance.

First-ever Mobile Broadband Coverage Map

Finally! Now we can compare! Cellular providers' coverage map laid bare. The nation's first-ever mobile broadband coverage map is put out by the Federal Communications Commission. The FCC used information from AT&T, Verizon, and the others, along with information from their customers, to make the map.

The map is a tool, allowing you to compare coverage of both data and voice plans of the four largest carriers in the United States.  This is a dramatic improvement over maps offered by the carriers:  AT&T, Verizon, T-Mobile, Verizon and US Cellular.

A huge step forward

Consumers have long complained about not being able to make calls or access their mobile carrier's network in remote areas like that pictured here, Stafford, Humboldt Redwoods State Park, California.

The tool includes options to track wireless download and upload speeds of at least 5 Mbps and 1 Mbps for each of the largest carriers.  This all was set in place by a law passed in 2020 which requires the FCC to collect and release data comparing different wired, fixed-wireless, satellite and mobile broadband service providers.

OSU's robot Cassie ran a 5K

Milestone in Artificial Intelligence and Machine Learning

A milestone in Artificial Intelligence and Machine Learning is reached here with this robot. "Cassie" successfully ran a 5K marathon on one charge with no tether. The applications for this technology are are phenomenal! Consider what impact this will have on the delivery business and being able to help the disabled and elderly. The robot learns how to stay upright by itself and is able to make subtle adjustments. Cassie crashed twice in the 53 minute run around the Oregon State University campus, but the improvements to the deep reinforcement learning algorithm make it possible to dream unimaginable benefits for mankind over the next decade.

Donald Trump Lost

The giant has spoken.  Donald Trump lost.  YouTube says so.  The San Bruno-based company announced today that from here on out it will begin removing certain videos from its platform.  This comes after the U.S. Supreme Court refused to hear a Republican effort to overturn the November 3rd, 2020 General Election.  This was the first such case to make it all the way to the high court.  The unanimous decision met the "safe harbor" deadline to overturn the results showing Joe Biden as the winner.

Photo © Henry Mulak

In its blog, YouTube says, "...enough states have certified their election results to determine a President-elect.  Given that, we will start removing any piece of content uploaded today (or anytime after) that misleads people by alleging that widespread fraud or errors changed the outcome of the 2020 U.S. Presidential election, ..."

The decision strikes a blow against conspiracy theorists, led by Donald Trump, who claim widespread fraud.  Without YouTube, such people and organizations will be forced onto platforms which hopefully can be more easily targeted by the rule of law and dealt with accordingly.

A Big Part of the Story

While this is a story about the election, it's also about the power of new media and the transformation in how Americans get information.  Much is being said about misinformation and its impact on American politics, but a kernel of hope can be derived from today's announcement.

For one, we see how giants like Alphabet-owned YouTube and Facebook are, at least, making what appear to be an attempt at culling the truth from what is now a constant flood of information.  But there's also a hopeful message here about where Americans are getting their news.  And that is, as the YouTube blog says: "Authoritative news organizations" are "the most popular videos about the election" on the platform.   As someone who works for one of those "organizations," and knows how much sweat and tears go into honest reporting, I'm am heartened by today's announcement.

Hungary No Longer a Democracy

Hungary is no longer a democracy.  So says Freedom House, a non-governmental organization based in the United States that monitors governments around the world.  If true, Hungary has become the first country within the European Union to abandon its democratic institutions in favor of another form of government.  The title given to this new form of government is "Transitional / Hybrid Regime."  Poland is clearly on the same route, according to Freedom House which put together this graphic to help visualize the transition:

Hungary is led by Prime Minister Victor Orbán who rose to power in 2010, just five years after the central European nation was heralded as a model in the post-Soviet era.  The new report also cites the erosion of democratic institutions, not only in Poland, but the Czech Republic, Slovakia, Latvia and Montenegro.

The report says, "Prime Minister Victor Orbán's government in Hungry has ... dropped any pretense of respecting democratic institutions," becoming "the first country to descend by two regime categories and leave the group of democracies entirely."

Getting up to Speed on Artificial Intelligence

Not a day goes by anymore without a mention of Artificial Intelligence in the mainstream press, and for good reason.  Take, for example, the ongoing coronavirus pandemic.  A company specializing in AI was first to spot the outbreak.  AI and its smaller cousin, Machine Learning, are now being offered up as helping with solutions to the pandemic.  If you’re foggy on what AI is, let’s get you up to speed.  The science has been around since the 1940s.  Now it looks to be maturing enough to really change our world in significant ways, and in dramatic fashion.  More than ever, it needs to be understood so everyone can have a hand in shaping the future.

(Pictured: A neuron modeled after one in our brain) 

The concept of Artificial Intelligence has been around since the development of a mathematical model of a biological neuron.  This was done some 80 years ago.  We’re not talking about just computing.  This is also about learning.  Cognition.  That’s what we refer to when we say “learning.”  Trying to get a computer to do it has proven extremely difficult.  But now, it’s finally taking off.  There are a number of reasons for this, not the least of which is how much digital data we now have.  This is due to the fact that computers have now been in widespread use by the average consumer for some 30 years.

(Pictured: An artificial neuron)

Data is everything.  It used to all be in analog form. It took the human mind to process it.  i.e. Read a book by turning the pages, process what you read and somehow put it to use.  We do that through the 100-billion neurons in our brain, though just how, is something we are still trying to figure out.  Whatever the case, we know that these single-celled neurons work together in amazing ways, making us the most intelligent creatures to walk the planet.  The big issue for computer scientists has been trying to replicate that intelligence for use in machines.

Don’t be misled into thinking that somehow scientists one day just dreamed up the concept of patterning Artificial Intelligence after the neurons in our brain.  Quite the opposite.  We are talking about decades of work by some of the brightest minds in science and engineering.  The work on the development of artificial neurons, an example of one seen above, has its origins in the 1940s.

Developing these artificial neurons into networks that could actually do computations started by the 1980s: “Artificial neural networks,” as we now say.  This is a huge development.  Now we go beyond just putting hardware and programming software together to solve a problem.  We branch into actual learning, Machine Learning.  Training a neural network to recognize faces comes to mind.  There are lots and lots of other applications. More on that in a moment.  And we also need to add that mathematical neural networks are gross oversimplifications of the brain neural networks.  Whatever the case, we first have to explain how AI is NOT a computer program.

Some may get stuck on the now old concept of writing a software program to solve a problem, like how to email someone.  This type of program doesn’t require learning, other than you figuring out how to use it.  Actually getting the program to learn on its own is a whole new ballgame.  One might think, for example, that a programmer just sits down and writes something to recognize what someone is saying.  Speech recognition!  No, no!  What is really happening now is that there are a multitude of already written clever modules prepared and on the shelf for use in making computers think intelligently.

Python is well known when it comes programming a new AI project.  It’s relatively simple and can be easily learned.  R is also at the forefront.  It’s popular because open source programs giving users a lot of latitude in building an intelligent machine.  Lisp, Prolog, Java, Tensorflow and Torch are just some of the others.

(Pictured: An unsolvable mathematical equation) 

Now for the catch. Knowing a computer program and how to code is one thing, but the real barrier you come up against in the field of Artificial Intelligence is the math.  Let’s face it, if you understand the artificial neuron pictured above you would not be reading this.  Note the picture here of the problem no one can solve. You have to know the math to work in AI and that stops many people from even taking up the subject.  AI has become so pervasive that we all need a better understanding of it going forward.  For the record though, the specific math used in AI that we’re talking about is linear algebra, probability, multivariate calculus and optimization.  To demonstrate some of the complexity involved in math for neural networks, in this case “back propagation,” try to get your mind around what an expert offered to explain the problem you see in the picture on the previous page:

“The problem is complex and complicated to untangle the solution and as the number of variables increase (as in our case of 13,002 weights & biases in our neural network of 4 layers with 784 neurons, 16 neurons, 16 neurons & 10 neurons). The complexity of the problem increases astronomically, including the computational effort and the time it takes to solve it. Also, the brute force method would take years to solve this problem. Hence, we use gradient descent to solve this problem by finding the minimum of the cost function and the appropriate weights and biases.”  

The human brain, with more neurons in its pre-frontal cortex than any other, is a neural network.  Advances in AI recently have focused on the development of artificial neural networks, a computing system of highly interconnected elements or notes.  These neural networks are organized in layers and produce responses akin to what might be expected if a human is involved.  Deep learning involves neural networks with many middle layers.  This idea, for the purposes of this paper, is hugely oversimplified.  That’s because “consciousness” has not been achieved by machines, so it cannot be easily explained.  In fact, the big debate in the field right now is if AI will ever match or surpass human intelligence, or consciousness.

There are so many artificial neural networks, and more are being constructed all the time.  The image below cuts to the issue of having more than just an input and an output.  That would simply be linear, something easily accomplished with anything nowadays having to do with input and output.  For example, type on the keyboard in a computer texting program and a word appears.  That’s linear input to output with no computation involved.  But here with NNs we have layers of neurons; what makes them clever is having at least one middle layer to solve complex problems that are more than linear.  The issue with artificial neural networks is that they learn, going from input to an output that is generally unknown. In the following image you see three inputs and one result, or output.  Imagine a computer sensing three inputs, like an alarm at 7 a.m., the sun coming up and a dog barking.  Your brain would go through a series of responses to determine what to do next.  Now imagine training the artificial neural  network to respond and you get an idea about how far we’ve come and what more needs to be accomplished.

(Graphic: an artificial neural network)

The reason we are now hearing so much about AI is that the engineers and scientists have really begun to get their arms around how to make it work.  Evidence of it is everywhere as Artificial Intelligence has hit the mainstream with Apple’s personal assistant Siri, Amazon’s Alexa, Tesla’s self-driving capabilities, Netflix, Pandora, Nest and so on.  Take a deep dive into each of these services and you delve into what AI can really do.  But these businesses only give us limited view of the big picture.

Choose any other segment of society, including government, transportation, law and order, or healthcare and you begin to see just how promising AI can be.  Future prospects for AI in healthcare, for example, include drug creation and helping people make healthier choices and wiser decisions.  That’s just to start. One of the reasons the healthcare aspect of AI is taking longer in its development is privacy issues, but there are an increasing number of companies jumping into its development.  For example, there’s a concept to mine called “precision medicine.” It employs numerous technologies to guide individually tailored diagnosis and treatment for patients.  The technology will learn about you with the assistance of your healthcare provider and then tell the doctor how best to treat you.

The government part of AI freaks people out the most.  The surveillance state is talked about a lot, including the use of cameras in public places to track your movements and alert authorities when you’ve done something wrong.  Now there are even cell phone data maps already at work finding out who are ignoring quarantines in hopes of  stopping the spread of the coronavirus.  You see, the data is already out there since your phone tracks where you are and AI can be employed to determine behaviors.  

In conclusion, we could say that Artificial Intelligence is finally coming of age.  AI has matured enough to become mainstream.  It’s been nearly a century in the making and will likely touch every aspect of our existence in the not-to-distant future.  This means new and wonderful things, but also brings great peril.  The power of AI in the wrong hands is something we need to contemplate, just as we also imagine the dawn of a new era.  This new era will rival the introduction of electricity as well as the Industrial Revolution.  Now we can contemplate not having to think too hard because getting a machine to do a lot of the difficult work is here.  Another way to think about it is that we are freeing our minds up to contemplate an even greater existence for humankind.  What we all do need to do though is become aware of what’s happening with AI.  Think hard now, because Artificial Intelligence offers us something beyond the scope of our imaginations.