AWS DeepLens: Can You See The Future?

“In farming, we plant the seeds, make sure they have enough water and nutrients, and reap the grown crops. Why can’t technology be more like this? It can, and that’s the promise of machine learning. Learning algorithms are the seeds, data is the soil, and the learned programs are the grown plants.” — The Master Algorithm by Pedro Domingos

“For at least the last half-century, important measures of computing, such as processing speed, transistor density, and memory, have been approximately doubling every eighteen to twenty-four months, which is an exponential pace (power of 2).”

“You’ve probably experienced this remarkable achievement yourself without realizing it.”

“Your first smartphone may have had a spacious eight gigabytes of memory, a small miracle for its time. Two years later, if you bothered to upgrade, you likely sprang for sixteen gigabytes of memory. Then thirty-two. Then sixty-four.”

— Humans Need Not Apply by Jerry Kaplan

“New systems of the first class, many of which are already deployed, learn from experience. But unlike humans, who are limited in the scope and scale of experiences they can absorb, these systems can scrutinize mountains of examples at blinding speeds. They are capable of comprehending not only the visual, auditory, and written information familiar to us but also the more exotic forms of data that stream through computers and networks. Imagine how smart you would be if you could see through thousands of eyes, hear distant sounds. You’ll get an idea of how these systems experience their environment.” — Jerry Kaplan

“In a nutshell, after fifty years of effort and billions spent on research, we’re cracking the code on artificial intelligence. It turns out it’s not the same as human intelligence, or at least it looks that way right now. But that doesn’t matter.” — Jerry Kaplan

In the words of computer scientist Edsger Dijkstra “The question of whether machines can think is about as relevant as whether submarines can swim.” — Jerry Kaplan

Humans Need Not Apply by Jerry Kaplan begins with a simple puzzle to solve. Can you solve it? Take the numbers 100, 1,000, and 10,000. What about the numbers 32, 64, and 128? What do these numbers have in common? Answers to both exponential growth? The powers of 10 and 2, respectively. The puzzle has many solutions, however. Another suitable answer is that eventually, the numbers grow so unfathomable, the answer evolves from a quantitative topic to a qualitative one. The Rubik’s Cube doesn’t end there. What do you get by the eightieth step from the numbers of 100, 1,000, 10,000? Here’s a hint: the answer morphs from a quantitative solution to a qualitative one. The solution is not a number, but an observation. Need more time? Another hint? Look up. Preferably at night, with a telescope. The answer is that by eightieth step, the number has likely surpassed “the estimated number of atoms in the universe.” Humans Need Not Reply uses the analogy for how technology, in the domains of what we call “big data” will so transform our lives that everything from our government policy to how homes get painted will be unrecognizable. This is not a book on understanding “big data”, but how physical manifestations of artificial intelligence will finally see light. This is nothing short of AI robotics, what Humans Need Not Apply calls the fusion of synthetic intellects and forged laborers. These involve robots, able to cut grass and paint houses, go to war and stand in line. Forged laborers in particular, will rewrite not just the rules guiding traffic but society itself. Should Russia ever fire a nuclear missile on the United States, we should much prefer that than what synthetic intellects can do, says Jerry Kaplan. The Russian nuke will be merciful enough to give us a few extra minutes to plan a move.

“Synthetic intellects will soon know more about you than your mother does, be able to predict your behavior better than you can, and warn you of dangers you can’t even perceive. The second class of new systems arises from the marriage of sensors, and actuators. They can see, hear, feel, and interact with their surroundings. When bundled together, you can recognize these systems as robots.” — Jerry Kaplan

All animated beings, that we know of, have four categories that separate them from non-living things, and relatively inanimate beings like plants. They are:

1. Energy
2. Awareness
3. Reasoning
4. Means

All living things require energy. Awareness is the ability to sense environment, which doesn’t require brains in biology. Simple creatures like the enormous Japanese Nomura’s jellyfish fall under categories 1 and 2. The Nomura’s also fall under category 4, as does every creature. Means is an ability to move, or at least physically get what you need for food, etc. The magic is in category 3, as higher thinking mammals that include Orcas, dolphins, elephants, primates all composite all four categories. No biological construct has ever surpassed human abilities in category 3. We are now starting to see autonomous robots in factories performing mainly, rote and singular tasks. You can compare them to jellyfish, unable to think or exhibit range of movement. Just as DNA evolved out of living creatures, branching out to more complex species, artificial and mechanical machinations are following behind, at a much accelerated pace. Which goes to category 3. How exactly did an AI have the ability to think?

“About 120 years ago, after millions of years of evolution, something magical happened: through us, life suddenly developed the means to to burst free of the locality constraint. Guglielmo Marconi figured out how to use electromagnetic radiation — more commonly called radio waves — to transmit information instantly between distant locations with no evident physical connection. And Thomas Edison figured out how to move energy, in the form of electricity, through wires at a relatively low cost. We’re still sorting out what this will ultimately mean.” — Humans Need Not Apply

AI, which is defined as synthetic intellects, widely known as algorithms, neural networks, cognitive systems, and machine learning emerged out of models to simulate the human brain. Humans Need Not Apply tells the tale of Nathaniel Rochester, one of the most talented adopters of AI in the 1950s. Definitely the most renowned organization in the field, IBM, gave Rochester full autonomy to run IBM’s Watson Research Lab to go ahead with AI. Rochester met with peers, scientists, researchers, and academic leaders in 1956, in what is likely the world’s first ever conference on the subject. The conference lasted over two months, which the attendees believed “a significant advance can be made in one or more of these problems if carefully selected scientists work on it together for a summer.” No such advance happened, though Rochester returned to the Watson Lab with tingling ideas. But Rochester only discovered horror. Higher ups at IBM, thought Rochester’s work on whatever it was Watson Labs was building, could cannibalize existing sales teams and powerhouse computers. These fears reached so far, than an internal IBM report essentially said to shut down the Watson Lab, as managers worried about their own longevity. Irony aside, at the time computers were not much more than high school lockers with nobs and buttons. Most couldn’t imagine them used for anything higher than calculators for military equipment. The thinking on AI was ruled by what was called “symbolic systems” following linear, “you do this, then you do that” models. This worked on logical problem solving, involving board games, etc. This thinking branched out into “heuristics”, allocating resources to solutions at reasonable length, before moving on to find different solutions. Symbolic systems had trouble with complexity. The analogy is GPS, but attempting to figure out a destination by starting on a side street, and one street at a time trying to find the destination.

“It’s one thing for these systems to recommend what music you should listen to or what toothbrush you should buy. It’s quite another when we permit them to take action on their own — make them autonomous — because they operate on timescales we can barely perceive, with access to volumes of data we can’t comprehend. “ — Humans Need Not Apply by Jerry Kaplan

Researchers resorted to philosophy. What exactly are talent, skills, and expert domain? The answer is repeated experience in a subject, with much trial and error. On a cellular level, how are these things obtained? The Talent Code by Daniel Coyle talks about sausage like wraps called myelin, that wrap around nerves like pigs in a blanket. As humans exhibit behaviors of intellectual or physical exercise, the fibers wrap around nerves like coverings on cable wire. This is the basis of the ten thousand hours, and muscle memory. But back in the 1960s, scientists only knew that some combination of electrical, chemical exchanges of neurons — building blocks of the brain — communicated to each other. Researchers attempted to replicate the neuron in computer form, block by block building models that looked like neurons. Children learn by inputs, through sensory feedback. So researchers wondered if by pumping inputs in the models, observing a response output: whether the artificial intelligence could learn faster, better. It earned the name “neural networks.” In the early 1960s, a researcher named Frank Rosenblatt managed to build a bundle of these so called “neural networks” that he called them “perceptrons.” These models reached a point that they could recognize patterns, perhaps the first ever neural networks to do so. Rosenblatt’s perceptrons, unfortunately, were constrained by the technology of the time. Contemporaries at MIT published works denouncing the creation, causing the perceptrons to die a slow death. Funding for projects like these went away. Of course, those perceptrons and its kin are forerunner to artificial intelligence today. The 1990s and the 2000s brought in computers with the power, to actualize what Gordon Moore once acknowledged fifty years ago. What Moore couldn’t have known, is how the doubling of computing power hasn’t stopped since.

“As sensors and wireless communication continue to improve, they will vanish from view as surely as computer technology has. I’m old enough to recall a day when you could pick up a piece of computer memory and literally see each bit called “core memories.” Today we perceive gigabytes of computer memory as flat, black rectangles the size of a postage stamp, if they are physically separate from other components at all. Someday you may be walking through what looks like a pristine wilderness, blissfully unaware that an extensive network of self-organizing collaborative devices are maintaining the environment.” — Jerry Kaplan

The maturation of artificial intelligence, made possible by computing powers capable of housing infinite data, are two of the four planets that have aligned in the Manhattan Project of our time. The other two are near completion, the third planet being the physical manifestation. The body. The book gives an example of the author, observing someone inside the Stanford AI, sparring in a sword battle vs. an automaton. The author himself duels vs. the automaton. The robot was capable of tracking its opponents moves, adjusting accordingly. That was back in 2011, and you can watch it here. Though only just a limb, kind of like robotic snakes, it shows you the power of such beings when fused with potential military funding. More important, perhaps more dangerous, is how these “robots” will not appear in the forms we foresee. They will not be encased by what engineers call “locality.” They will not have bodies. Biology places the eyes close to the brain, in all species, because it is the fastest way for nerves to connect. “Because biological creatures by themselves can’t communicate or transmit energy over long distances, their body parts have to be near each other” says Kaplan. But artificial intelligence has no such constraints. What if your eyes were feet apart, instead of mere inches? Instead of two eyes, what if you had 6, 8, several dozens? The same with ears. It’s more likely, that future police officers will not be androids running on foot, but hundreds of sensors hovering over highways, self-driving patrol cars, all powered by an artificial intelligence located somewhere on a data farm. The term “swarm robotics” is named after robotic insects like these. These forms will do anything from painting houses, guided by an app someone opens up and connected by paint nozzles, to the next generation of war. Today, there are drone pilots in the Nevada desert piloting drones in the Middle East. The robots as bodily humanoid forms, are largely myth. Which brings it all to the last planet to align. The last planet?

“As we amass data from an expanding array of sensors that monitor aspects of the physical world — air quality, traffic control, ocean wave heights — as well as our own electronic footprints such as ticket sales, online searches, blog posts, and credit card transactions, these systems can gleam insights inaccessible to humankind.” — Humans Need Not Apply by Jerry Kaplan

The final planet to align is the eyes. The eyes of the beholder, although that one is a misnomer too. Depth of perception through superior cameras, add machine learning, armed with unlimited information, powering our physical robots will uproot society. What is interesting about the evolution of life forms vs. the advances in technology, is that they grow in opposing matter. Species grow more specialized with evolution, differing out and creating variations of the original forerunner. Technology does the opposite. It absorbs all the genes of various traits, becoming more singular as it grows. The Apple II computer: once the most powerful machine of its time, but it could only hold one second worth of music on a CD. In turn, most smart phones can hold over twelve days worth of music on a CD. As computing power increases, it absorbs tasks that other gadgets specialize in. All the apps on our phones were once individual industries, the books, the music, the video, the photos, in the forms of print, vinyl and CDs, camcorders, cameras, etc. The recognition of faces, voices of people, and what activity those people are engaging in will all fuse into some distant offspring that will make today’s smart home devices the vinyl and CD. Golfers in Silicon Valley know robots as caddies. Some wonder if autonomous vehicles will move themselves after a two hour park limit, imposed when laws thought it would be inconvenient for people to do so. In an emergency, rules are accepted to be broken. Ambulances, etc. Will self-driving cars do the same, or obey traffic laws to the rule? The Aesop’s Fable of the Donkey come to life, will these vehicles operating under different AI compete against each other? This is a lot of imagination for a camera, the AWS DeepLens, releasing in the summer. But it’s never too soon to scan the horizon.

“Even self-driving cars aren’t going to be nearly as self-contained or autonomous as they appear. Standards for vehicles and roadside sensors to share information, essentially becoming one interconnected system of eyes and ears, are close to completion. The U.S. Department of Transportation, among other institutions, is developing so called V2V (vehicle to vehicle) communications protocols. Integrated with traffic control and energy management systems, your future car will simply be the visible manifestation of an integrated flexible public transportation system.” — Humans Need Not Apply

Humans Need Not Apply moves away from the utopian or dystopian future, and explores more practical applications of AI. Digital marketing and ads. The online advertising field is nothing less than total war. War often begins with the most benign of circumstances, be it competition for some unused land or disagreement in worldview with neighbors. This war is for prize to our eyes, once again, but the battle starts innocently with a pixel. The book observes how someone, in the early days of the internet, noticed how one pixel was all but invisible. Camouflage. By incorporating one pixel, the AI had all it needed to harness the most important commodity in media…Information about you. While we can’t see these pixels, one pixel on a website acts as a sticky note on your computer. To put it better:

“Why display this if you can’t see it?”

“That’s the whole point. You can’t see it, but that single pixel may come from anywhere, in particular from someone who would like to take note of when and from where you visited that particular page.”

“Because the pixel comes from someone else’s server, they get the automatic right to make a notation on your hard drive. You might wonder who the parties are that trail you around the Internet.”

— Humans Need Not Apply

Those parties belong to the most powerful industries in the private sector. Before getting acquired, the former AI advertising agency Rocket Fuel estimates that at time of book’s writing (2015), it has implanted over 90 percent of all personal computers in the United States with cookies. Like chestbursters popping out as Xenomorphs, they multiply spreading across populations of demographics, markets, nations, generations, anyone with access to the personal computer. Of course, you can always “turn third party” cookies off but… Cookies are tasty, especially those enormous ones that sit inside Subway with macadamia nuts. If they called them “meatloaf” or some other name like “Vienna sausage” or “poison berries” that would be funny. Anyways, the information that carefully watches your activities on the web, becomes the most valuable asset for advertisers. Hence why multi-million dollar mercenaries like Rocket Fuel exist, before its acquisition by bigger fish.

For example, let’s say you visit websites about vegetarian or healthy cooking. By itself, there’s not much to go off of. But data shows that you are more likely to look around for yoga studios in your neighborhood. AI scrambles to put up bids, or auctions for the rights to sell you advertisements that will then pop up. The next website you visit, might well display studios for yoga that was inferred by your searches in vegan cooking. What’s called “impressions” in the media business, can become alarmingly accurate with the data accumulated from your activities on the web. Chief Technology Officer of Rocket Fuel, Mark Torrance, shows a “heat map” based off his AI models on targeted advertising. The map revealed an index for one of his company’s clients, a national pizza chain. For a test group, the map’s index guessed between 9.125 and 11.345 % of the group, will order a pizza after seeing the impression. The pizza chain has no way of knowing this prior, but later reported back to Rocket Fuel the actual figure. 10.9 percent of the people who saw the ad, ordered pizza within two weeks. This is gathered from mass accumulation of data, pieced together like a puzzle that eventually, reveals who you are. It will be interesting to see what happens as AI advances on advertising agencies, against each other for the prize of your eyes.