Qubits and binary data successfully squirted down the same fiber, at the same time: Here comes quantum cryptography

Toshiba Research and Cambridge University in England are reporting that they have succeeded in building the first conventional fiber-optic network that’s capable of transmitting and receiving both quantum data (for encryption) and normal high-speed binary data. This breakthrough means that the world’s fiber networks can now be secured with theoretically unbreakable encryption.

The basic concept behind quantum cryptography is that Alice sends a crypto key to Bob via a stream of single photons (quantum key distribution, QKD). If a man-in-the-middle attacker somehow manages to intercept the photons, this interrupts the transmission in such a way that can be detected. In theory, quantum crypto should result in totally secure communications. (In reality, there are other attack vectors that bypass the inherent security provided by the photons.)

Now, quantum-secured networks aren’t particularly new, but until now the single photons (qubits) have required their own dedicated optical fiber for transmission. In conventional fiber-optic networks, the transmission of data is very intense, with over 1 million photons carrying a single bit of binary data. Sending single photons down the same fiber simply wasn’t feasible; it was impossible to extract that single photon at the other end. Until now!

Toshiba/Cambridge QKD + fiber data network diagram. A = the overall setup. B = the quantum transmitter. C = the quantum receiver.

The Toshiba and Cambridge researchers have overcome this restriction by transmitting the quantum photons and data signals at different wavelengths, and by using a special photodetector at the receiving end that turns on for just 100 millionth of a microsecond (a few hundred femtoseconds). The different wavelengths mean that the signals don’t clash, and the photodetector only turns on when it expects to receive a single photon.

The end result is a fiber network that can transmit binary data at 1Gbps in both directions, and perform quantum key distribution at 500Kbps at the same time, over a 90km (56mi) length. This is apparently 50,000 times faster than the previous best QKD over a fiber network of this length. In theory, this means that the next generation of fiber networks — assuming this femtosecond photodetector can be implemented commercially — could be secured with quantum cryptography. Most importantly, we’re only talking about new routers — this method of QKD could be performed on existing (and very expensive) fiber networks.

The immediate benefit will be for military and police/security networks, and domestic infrastructure (think secure communications in smart cities). With fiber steadily rolling out to consumers, though, you and I might soon be downloading torrents and watching cat videos that have been encrypted with quantum cryptography.

Now read: Quantum teleportation lays the foundation for a global quantum internet — or check out our popular story on the secret world of submarine fiber-optic cables.

Research paper: 10.1103/PhysRevX.2.041010 – “Coexistence of High-Bit-Rate Quantum Key Distribution and Data on Optical Fiber”

Source

EU seeks to reduce world’s energy usage, and double data center speeds, by replacing copper with fiber

Data centers are big and costly. Engineers all over the world are working hard at making servers and networking more efficient. Processors are using less power, cooling is getting easier, and evenrouters are reducing their footprint. Sadly, data centers are still using a gigantic amount of power, so the European Union is funding a trend away from traditional electrical data connections. Headed by the Fraunhofer Institute in Germany, project PhoxTroT aims at reducing power consumption by using light-based data connections, while at the same time increasing transfer speeds to two terabits per second (Tbps).

An article from Fraunhofer explains that this four-year project isn’t about reinventing the wheel — optical data transfer is already used around the world. Instead, PhoxTroT will be focused on taking existing technologies, combining them, and refining them into a system that will save money and use less energy while doubling connection speeds. “They will realize the optical transmission on a printed circuit board (‘on-board’), ‘board-to-board’ and also ‘rack-to-rack’. By combining these interfaces, it will also be possible to bridge longer distances within the foreseeable future,” says the article. This isn’t just a dolled-up fiber optic cable — this is taking the technology to the next level by integrating light-based data transfer throughout entire data centers on the individual server level, while increasing the effective range to hundreds of kilometers.

Not only is optical networking more power efficient and faster than its copper counterpart, but it’s also more robust in the face of disaster. After Hurricane Sandy took out a non-trivial amount of communications on the east coast of the United States, telcos went through and replaced copper lines with fiber-optic cables to update their network speeds and reliability. Electrical data transfer like typical coaxial and Ethernet cables still have a place, but it is slowly being overtaken in usefulness by optical data transfer. If PhoxTroT is a success, copper wiring will become even more of a niche.

With a little under twelve million dollars invested by the European Union, and eighteen different companies working together over the next four years, PhoxTroT can transform the data center into a much more eco-friendly and cost effective endeavor. Google‘s data centers alone draw 260 million watts continuously. A single Amazon data center in 2011 drew eight million watts continuously. Worldwide, data centers account for around 30 billion watts — a few percent of the world’s total power usage.

If these engineers can double the data throughput while using a small fraction of the power traditional networking uses, we’re talking savings of tens of millions of dollars per data center. The EU should be applauded for its efforts, and other countries and organizations should take a page out of its handbook in this instance. We’re saving money and saving the planet one data center at a time.

Now read: Will 100Mbps internet connections destroy the web as we know it?

[Image credit: Adrienne Serra & Alex]

Source

Harvard creates self-assembling DNA nanostructures from tiny DNA ‘Lego bricks’

Harvard’s Wyss Institute, which brought us 700-terabytes-per-gram-of-DNA data storage earlier in the year, has now produced DNA Lego bricks — three-dimensional DNA building blocks that self-assemble into more than 100 different, three-dimensional structures (pictured above).

These DNA Lego bricks are short strands of DNA that have been specially crafted to join with other DNA bricks at a 90-degree angle — just as if you had pushed two eight-stud Lego bricks on top of each other at 90 degrees. By joining more and more of these DNA bricks together, a 3D structure emerges. In this case, the DNA Legos are built into 25-nanometer cubes, which consist of around 1,000 voxels, with each voxel consisting of DNA strands that are just 2.5nm. A voxel (volumetric pixels) is a term borrowed from graphics; it’s essential the 3D equivalent of a 2D pixel.

The Wyss Institute call these cubes the “master molecular canvas.” By restricting which DNA bricks are available during self-assembly, 102 distinct 3D shapes were formed. In the image at the top of the story you can see the simulated 3D models of these 102 shapes, and below is an actual microscopic view from above. As you can see, the level of detail is really quite astonishing — and even better, some of the shapes include intricately detailed tunnels and cavities. “This is a simple, versatile and robust method,” says Peng Yin, who led the project.

Essentially, Peng Yin is now an architect of, quite possibly, the world’s smallest building blocks. Intel alters features that are perhaps 30 nanometers in size, while Yin has the power to alter a single 2.5nm voxel. This is important and exciting because changing a single voxel could alter the function of the DNA cube, much as moving a single transistor alters the function of a computer chip.

As for what these self-assembled DNA cubes will actually be used for, the answer is probably medicine. DNA molecules are (obviously) biocompatible, and Harvard’s Wyss Institute is generally oriented towards medical research. The general idea is that you could somehow fashion a DNA structure that interacts with the human body in a curative or preventative way — or, more simply, you might fashion a DNA cube that can carry medicine to a specific region of the body.

Moving forward, there could be non-medical applications too. In much the same way thatHarvard’s DNA data storage could be used for storage in computer systems, these DNA building blocks might one day form the basis of biological (or digital-biological hybrid) computers.

Now read: Living organ-on-a-chip could soon replace animal testing

Research paper: DOI: 10.1126/science.1227268 – “Three-Dimensional Structures Self-Assembled from DNA Bricks”

Source

Spaun, the most realistic artificial human brain yet

A group of neuroscientists and software engineers at the University of Waterloo in Canada are claiming to have built the world’s most complex, large-scale model simulation of the human brain. The simulated brain, which runs on a supercomputer, has a digital eye which it uses for visual input, a robotic arm that it uses to draw its responses — and it can pass the basic elements of an IQ test.

The brain, called Spaun (Semantic Pointer Architecture Unified Network), consists of 2.5 million simulated neurons, allowing it to perform eight different tasks. These tasks range from copy drawing to counting, to question answering and fluid reasoning. At this point, you should watch the video below to get a rough idea of how Spaun works — and then read on to find out why Spaun is so interesting.

Now, the nitty-gritty details. Spaun has a 28×28 (784-pixel) digital eye, and a robotic arm which can write on some paper. Every interaction with Spaun is through its 784-pixel eye. The scientists flash up a bunch of numbers and letters, which Spaun reads into memory, and then another letter or symbol acts as the command, telling Spaun what to do with its memory. The output of the task is then inscribed by the robotic arm.

Spaun’s brain consists of 2.5 million neurons that are broken down into a bunch of simulated cranial subsystems, including the prefrontal cortex, basal ganglia, and thalamus, which are wired together with simulated neurons that very accurately mimic the wiring of a real human brain. The basic idea is that these subsystems behave very similarly to a real brain: Visual input is processed by the thalamus, the data is stored in the neurons, and then the basal ganglia fires off a task to a part of the cortex that’s designed to handle that task.

All of this computation is performed in a physiologically accurate way, with simulated voltage spikes and neurotransmitters. Even the limitations of the human brain are simulated, as you can see in the video below, with Spaun struggling to store more than a few numbers in its short-term memory.

The end result is a brain that is mechanistically simple (2.5 million neurons isn’t really much to write home about), but which is surprisingly flexible. By implementing just a handful of very basic tasks, it’s interesting to see how complex behavior begins to emerge. There are some tantalizing hints as to how the brain evolved: starting with simple tasks, and then building upon and weaving them together to build complex functionality. In the video below, Spaun recognizes the pattern of a number sequence — the kind of question you would find on an actual IQ test.

Moving forward, the research team, led by Chris Eliasmith, wants to imbue Spaun with adaptive plasticity — the ability to rewire its neurons and learn new tasks simply by doing, rather than being pre-programmed. As for the ultimate end goal, Eliasmith is excited about Spaun’s prospects. “It lets us understand how the brain, the biological substrate, and behavior relate. That’s important for all sorts of health applications,” he says in an interview with PopSci. In testing he has “killed” synthetic neurons and watched performance degrade, which could provide an interesting insight into natural aging and degenerative disorders.

Spaun is built upon Nengo, a graphical open-source software package for building simulated neural systems. You can actually download the Spaun neural model, if you want to simulate your own brain — though I suspect it might require a little more processing power than your desktop PC.

Now read: Hackers backdoor the human brain, successfully extract sensitive data

Research paper: DOI: 10.1126/science.1225266 – “A Large-Scale Model of the Functioning Brain”

Source

Rice University creates graphene/nanotube hybrid material that could redefine electronics and energy storage

What could possibly be cooler than graphene or carbon nanotubes? Rice University’s new material that consists of forests of carbon nanotubes grown on sheets of graphene, of course!

This graphene/nanotube hybrid is as awesome as it sounds, too; we’re talking about a material that might be the single best electrode interface possible, potentially revolutionizing both energy storage (batteries, supercapacitors) and electronics.

To create the hybrid material, the James Tour Group at Rice University began with a copper substrate coated in a single layer of carbon atoms (graphene). From here, the process is a little bit mystical — it sounds like they place a mixture of aluminium oxide and an “iron catalyst” on the graphene, and heat the whole thing in a furnace. Within a few minutes, carbon nanotubes skyscrapers spring up from the graphene.

As you can see in the picture below, we’re quite literally talking about a sheet of graphene with carbon nanotubes growing upwards from it — up to a distance of 120 microns (0.12mm), which is really rather impressive at this scale. If we scaled it up to actual trees, they would rise into outer space. As you can see in the image at the top of the story, the carbon nanotube forest is also very dense. The most important thing, though, is that the bonds between the graphene and nanotubes are completely seamless — as far as electrons are concerned, there is absolutely no resistance when transitioning between graphene and nanotube.

Why is this important? Because this hybrid material has a ridiculously vast surface area: A single gram of the new material has a surface area of 2,000 square meters (21,500 sq ft) — half an acre of the most conductive material in the world. When it comes to energy storage, there is a direct correlation between energy density and the surface area of the electrodes — this new graphene/nanotube hybrid could result in significantly smaller batteries, or larger batteries that can do more work. In testing, Rice University created a supercapacitor with the new material that matches “the best carbon-based supercapacitors that have ever been made,” which is impressive because “we’re not really a supercapacitor lab, and still we were able to match the performance because of the quality of the electrode.”

Moving forward, the next step for advances such as this is production of the new material in commercial quantities. In all likelihood, the research baton will now pass to commercial companies, such as Intel, Sony, or Samsung, who will try to develop real components and batteries using the graphene/carbon nanotube hybrid material.

Now read: IBM creates breathing, high-density, light-weight lithium-air battery

Research paper: doi:10.1038/ncomms2234 – “A seamless three-dimensional carbon nanotube graphene hybrid material”

Source

Caltech and UVic set 339Gbps internet speed record

Engineers at Caltech and the University of Victoria in Canada have smashed their own internet speed records, achieving a memory-to-memory transfer rate of 339 gigabits per second (53GB/s), 187Gbps (29GB/s) over a single duplex 100-gigabit connection, and a max disk-to-disk transfer speed of 96Gbps (15GB/s). At a sustained rate of 339Gbps, such a network could transfer four million gigabytes (4PB) of data per day — or around 200,000 Blu-ray movie rips.

These record-breaking demonstrations took place at Caltech’s booth at the SuperComputing 2012 (SC2012) conference in Salt Lake City, Utah. In the booth, Caltech set up the mother of all high-speed networks: A handful of IBM x3650 M4 servers, each equipped with 16 OCZ Vertex 4 SSDs, connected to a Juniper MX 480 router (which has a total capacity of 1.92Tbps, or 300GB/s). From there, three 100-gigabit fiber-optic links connected the Caltech booth in Utah to the CANARIE, BCNet, Internet2, StarLight, and CENIC networks, which in turn transported data to end points at the Caltech university campus, the University of Victoria, and the University of Michigan. At each end point there’s another bunch of IBM servers, loaded up with SSDs.

Updated: The University of Victoria contacted us with more information about the hardware setup. There was also a Data Direct Networks (DDN) system, with a total of 288 SAS (serially-attached SCSI) 15,000 rpm drives. There was also two 24-bay 2U SuperMicro chassis, housing yet more OCZ Vertex 4 drives. Finally, there were 10 1.2TB Fusion-io PCIe storage cards, installed in a couple of SuperMicro servers. Just FYI: A single 1.2TB Fusion-io card costs somewhere in the region of $30,000.

In the case of the 339Gbps memory-to-memory record, we are talking about an aggregate of all three universities connecting to the Caltech booth in Utah. If you look at the graph below, and add together the inbound and outbound peaks, you get 339Gbps. The peak aggregate disk-to-disk transfer speed across the three 100-gigabit links was 187Gbps.

Over a single duplex 100-gigabit link between the University of Victoria and Salt Lake City, a memory-to-memory transfer rate of 187Gbps was obtained — just beating out last year’s record of 186Gbps.

The max disk-to-disk transfer rate of 96Gbps (1.5 gigabytes per second) was achieved between Salt Lake City and the University of Victoria. If we break it down, each IBM server was able to read data at 38Gbps (4.75GB/s) and write at 24Gbps (3GB/s) — so in the case of 96Gbps, we’re probably looking at 3 or 4 IBM servers working in tandem, on each end of the link.

These LANd (ha) speed records are all very impressive, but what’s the point? Put simply, the scientific world deals with vasts amount of data — and that data needs to be moved around the world quickly. The most obvious example of this is CERN’s Large Hadron Collider; in the past year, the high-speed academic networks connecting CERN to the outside world have transferred more than 100 petabytes of data. It is because of these networks that we candiscover new particles, such as the Higgs boson.

In essence, Caltech and the University of Victoria have taken it upon themselves to ride the bleeding edge of high-speed networks so that science can continue to prosper. There’s also the distinct possibility that advances to Internet2, which is slowly expanding to encompass hundreds of educational and scientific institutions around the world, might eventually trickle down to consumers, too — much in the same way that the original ARPAnet became the internet.

Now read: Will 100Mbps internet connections destroy the web as we know it?

Source

MIT creates carbon nanotube pencil, doodles some electronic circuits

A team of MIT chemists have created a carbon nanotube “lead” that can be used to draw freehand electronic circuits using a standard, mechanical pencil.

In a normal pencil, the lead is usually fashioned out of graphite and a clay binder. Graphite, as you may already know, is a form of carbon that is made up of layer after layer of the wonder material graphene. When you write or draw with a graphite pencil, a mixture of tiny graphene flakes and clay are deposited on the paper, creating a mark. (Incidentally, pencil leads never contained lead; it’s just that when graphite was first used in the 1500s, they thought it was lead ore, and the name stuck).

With MIT’s carbon nanotube pencil, the lead is formed by compressing single-walled carbon nanotubes (SWCNT), until you have a substance that looks and behaves very similarly to graphite. The difference, though, is that drawing with MIT’s pencil actually deposits whole carbon nanotubes on paper — and carbon nanotubes have some rather exciting properties.

In this case, MIT is utilizing the fact that SWCNTs are very electrically conductive — and that this conductivity can be massively altered by the introduction of just a few other atoms, namely ammonia.

In the picture above, electricity is applied to the gold electrodes (which are imprinted in the paper). The carbon nanotube pencil is used to fill in the gaps, and effectively acts as a resistor. When ammonia gas is present, the conductivity of the nanotubes decreases, and thus resistance increases — which can be easily measured. Carbon nanotubes are so sensitive that MIT’s hand-drawn sensor can detect concentrations of ammonia as low as 0.5 parts per million (ppm).

There are two main takeaways here. The first is that MIT has found a form of carbon nanotubes that is stable, safe, and cheap to produce. Second, carbon nanotubes have been used in sensors before, but usually the process involves dissolving SWCNTs in solvents, which can be dangerous. Here, creating a carbon nanotube sensor is as simple as drawing on a piece of paper — either by a human, or an automated process.

The team will now work on other carbon nanotube leads that can be used to detect other gases, such as ethylene (produced by fruit as it ripens) and sulfur (for detecting natural gas leaks). It’s also worth noting that the research was partly funded by the US Army/MIT Institute for Soldier Nanotechnologies — so it wouldn’t be surprising if military personnel are eventually outfitted with these sensors… or perhaps their very own carbon nanotube pencil, for MacGyver-like sensor fabrication in the field.

Now read: Hype-kill: Graphene is awesome, but a very long way from replacing silicon

Research paper: DOI: 10.1002/anie.201206069

Source

Clean, limitless fusion power could arrive sooner than expected

Good news, denizens of Earth: If the findings from two premier research labs are to be believed, commercial nuclear fusion is feasible — and could arrive sooner than expected.

The first breakthrough comes from Sandia National Laboratories (the same engineers who brought us the fanless heatsink). At SNL, a research team has been working on a new way of creating fusion called magnetized liner inertial fusion (MagLIF). This approach is quite similar to the National Ignition Facility at the LLNL in California, where they fuse deuterium and tritium (hydrogen isotopes) by crushing and heating the fuel with 500 trillion watts of laser power. Instead of lasers, MagLIF uses a massive magnetic pulse (26 million amps), created by Sandia’s Z Machine (a huge X-ray generator), to crush a small cylinder containing the hydrogen fuel. Through various optimizations, the researchers discovered a MagLIF setup that almost breaks even(i.e. it almost produces more thermal energy than the electrical energy required to begin the fusion reaction).

Probably more significant is news from the Joint European Torus (JET), a magnetic confinement fusion facility in the UK. JET is very similar to the ITER nuclear fusion reactor, an international project which is being built in the south of France. Whereas NIF and Sandia create an instantaneous fusion reaction using heat and pressure, ITER and JET confine the fusing plasma for a much longer duration using strong magnetic fields, and are thus more inclined towards the steady production of electricity. JET’s breakthrough was the installation of a new beryllium-lined wall and tungsten floor inside the tokamak — the doughnut-shaped inner vessel that confines 11-million-degrees-Celsius plasma (pictured above).

Carbon is the conventional tokamak lining (and the lining that had been chosen for the first iteration of ITER) but now it seems the beryllium-tungsten combo significantly improves the quality of the plasma. Hopefully this information will allow ITER to skip the carbon tokamak and jump straight to beryllium-tungsten, shaving years and millions of dollars off the project.

Moving forward, JET will actually try full-blown fusion with the optimum mix of deuterium and tritium (16 megawatts, for less than a second). At this point, JET is practically an ITER testbed, so its results from the next year or two will have a large impact on the construction of ITER’s tokamak, which should be completed by 2019.

Before today, magnetic confinement fusion was generally considered to be more mature and efficient than inertial confinement fusion — but Sandia’s new approach might change that. ITER is one of the world’s largest ongoing engineering projects (it’s expected to cost around $20 billion), and yet critics are quick to point out that we still don’t know if it will actually work. ITER isn’t expected to fuse D-T fuel until 2027 (producing 500 megawatts for up to 1,000 seconds) — and an awful lot can happen in 15 years. Still, the main thing is that we’re actually working on fusion power — when we’re talking about limitless, clean power, it’s probably worth investing a few billion dollars, even if it doesn’t work out.

Fusion reactors are some of the most beautiful constructions you’ll ever see, so be sure to check out our galleries of the National Ignition Facility and the Princeton Plasma Physics Lab.

Source

SpaceX launches ISS resupply spacecraft, despite mid-launch engine explosion

Just over a year since the final launch of Space Shuttle Atlantis and the retirement of the Space Transportation System, SpaceX has successfully launched a Dragon spacecraft aboard a Falcon 9 rocket — the first ever commercial resupply mission (CRS-1) to the International Space Station.

You can watch the launch in full, technicolor 1080p glory below. Watch out for the 90 second mark, where you can see one of the nine Merlin engines explode.

Yes, you read that right: During last night’s launch, an engine exploded — and yet the Falcon 9 rocket continued its climb into the stratosphere unabated. Believe it or not, Falcon 9 is designed for such “engine out” occurrences. When you’re burning RP-1 and liquid oxygen to produce 138,000 pounds of thrust, explosions happen. In this case, the Falcon 9′s on-board computers almost instantly register the engine failure, and then alter the output of the remaining eight engines to compensate. If you watch the video clip again, you’ll see that the engine outage is a non-issue; the rocket continues on course, as if nothing happened.

Both the Space Shuttle and Saturn 5 launch vehicle had similar “engine out” capabilities, though in their cases the solution was to shut down the faulty engine before it could explode. It isn’t clear why the Merlin engine exploded (SpaceX will hopefully hold a press conference later today with more info), but in any case it seems that the Falcon 9 rocket was constructed in such a way that an exploding engine isn’t a catastrophic failure.

According to SpaceX’s president, Glynne Shotwell, the launch itself was “picture perfect.” The Dragon capsule was placed in exactly the right orbit, and will now maneuver into position for docking with the International Space Station on Wednesday.

The mission, CRS-1, is the first of 12 commercial supply missions to the ISS, which SpaceX is carrying out under a $1.6 billion contract with NASA. The next 11 launches are expected to occur between 2013 and 2015.

The SpaceX Dragon launch is vital to the ongoing health of the International Space Station. Dragon will carry around 900 kilos of (2,000lbs) of cargo to the ISS, including vital supplies for both the crew and ongoing scientific experiments. More importantly, though, Dragon is capable of returning cargo from the ISS back to Earth — an ability that we have lacked since the retirement of the Space Shuttle. Except for the Space Shuttle and Soyuz, other cargo ships are simply jettisoned from the ISS and then burnt up during atmospheric re-entry — and Soyuz, with its cosmonaut occupants, doesn’t have space for return cargo.

Dragon is expected to return 905 kilos of cargo, 400kg of which is scientific experiments that have been stored on the ISS for the past year, unable to make the journey back to Earth. A lot of the cargo being returned is actually blood and urine samples, which ISS astronauts have been collecting over the last year, with the hope that we can better understand the effects of low-gravity environments. For a full list of the cargo being delivered and returned by Dragon, hit up the NASA/SpaceX CRS-1 mission manifest [PDF].

NASA administrator Charles Bolden, speaking to the press after after the successful launch, said that CRS-1 is “a historical event in the annals of space flight.” Elon Musk, CEO and CTO of SpaceX, said “We still have a lot of work to do, of course, as we guide Dragon’s approach to the space station. But the launch was an unqualified success.”

Updated: SpaceX has issued an official response. Apparently the engine did not explode — rather, “panels designed to relieve pressure within the engine bay were ejected to protect the stage and other engines.”

Now read: The Space Shuttle legacy in pictures

Source

The vast gulf between current technology and theoretical singularity

An interesting pair of news posts caught my eye this week, and they’re worth presenting for general discussion. First, VentureBeat has an interview with futurologist Ray Kurzweil, who made waves in 2005 with his book The Singularity Is Near. In it, Kurzweil posits that we’re approaching a point at which human intelligence will begin to evolve in ways we cannot predict.

The assumption is that our superintelligent computers (or brains) will allow us to effectively reinvent what being human means. In our present state, we are, by definition, incapable of understanding what human society would look like after such a shift.

Mrow

Meanwhile, Google is working to put its neural network technology to work on different sorts of problems. This past summer, the company taught its network how to recognize a cat by showing it YouTube videos. Specifically, it showed 16,000 processors enough cat videos that the network itself learned how to “see” cat without human intervention. Total visual accuracy, according to the initial paper, is about 16%. The announcement is about applying similar strategies to language processing and how computers can “learn” to understand the specifics of human speech.

Kurzweil, as you can see in the video at the bottom, is a persuasive speaker and Google’s success with teaching a network to recognize cats really is impressive. Reading stories like these, however, I come away skeptical. It’s not that I doubt the individual achievements, or that they can be improved, but focusing on specific achievements ignores the greater problem:We have no idea how to build a brain.

Kurzweil uses advances in scanning resolution and genetic engineering together as proof that at some point, we’ll be able to either program cell structures to do the things we want far more effectively than we can currently, or that we’ll simply be able to build mechanical analogs. On some scale, this is probably true. The nematode worm Caenorhabditis eleganshas 302 neurons. We could build a neural network (or neural network analog) with 302 nodes fairly easily — Google’s neural node structure is far more complex than that.

Unfortunately, just having nodes isn’t enough. The human brain has an estimated 100 billion neurons and 100 trillion synapses. Different neurons are designed for different tasks and they respond to different stimuli. They respond to and release an incredibly complex series of neurotransmitters, the functions of which we don’t entirely understand. It’s not enough to say “Yes, the brain is complex” — the brain is complex in ways that dwarf the best CPUs we can build, and it does its work while consuming an average of 20W.

That’s a monkey brain. We’ve got more.

This is where Moore’s Law is typically trotted out, but it’s a wretchedly terrible comparison. Scientists have already demonstrated transistors as small as 10 atoms wide. Your average neuron is between 4 and 100 microns. If groups of transistors equals neural networks, brains would be no problem. It’s not that simple. We don’t know how to build synapse networks at anything like the appropriate densities. We don’t even know if consciousness is an emergent property of sufficiently dense neural structures or not.

Self-driving cars (an example Kurzweil mentions) are a sophisticated application of refined models, meshed with sensor networks on the vehicle and additional positional data gathered from orbit. They’re an example of how being able to gather more information and correlate that information more quickly allows us to create a better program — but they aren’t smart. Our best neural networks are single-task predictors that gather information at a glacial pace compared to the brain.

The idea that we’ll strike some sort of tipping point within the next 33 years seems farcical. 33 years ago, scientists were well aware that genetic engineering, molecular biology, and cell phone-like devices might all be possible. Fast forward to today, and we have cell phones. We have better neural networks, certainly. We can dump far more data down the pipe, access it more quickly, and process the results — but creating human or superhuman intelligences? No way.

Source