Smart Watches, Smarter People?
Are smart watches making people dumber or smarter? If they aren't making people smarter, why would anyone buy them? Because they think they could make you smarter. Smart watches today actually outsell Swiss-made mechanical watches (in volume not in total dollar sales, yet: smart watches sold $1.3B in 2014 and $5B in 2015, Swiss watches were about $6B in 2015), but in the typical waves of disruptive innovation that happens every day, smart watches will sell in a much larger volume and probably surpass market size of Swiss watches next year.
Swiss watches are the best analog computers that money can buy. Think Steampunk computers. I've had a few of them, and they have their own issues. They wear out. They need adjusted. Their accuracy is good but not great. They can do a lot, but one smart watch can do everything that every Swiss watch ever made could do, and more. Swiss watches are now officially art; just fancy complications, and surpassed by orders of magnitude by the intelligence of the smart watch. We'll calculate by exactly how much later and we'll calculate exactly how smart smart watches will be (and when.)
So what is a Smart Watch?
The market has settled on a color display, connected to the Internet by a phone. Typically worn on your wrist, but you could have a pocket smart watch. It typically has very few knobs and buttons, but always has an interactive touch screen that recognize some gestures. The screen is small, maybe one square inch, and there are not many gestures that you can make. Swipe, left, up, right, down. Tap. Double Tap. Tap & Hold. Drag. There's basically 12 positions you can differentiate and 10 moves, so there's a one-dimensional communications channel of about 120 bits per second. This is not very sophisticated, which is one reason why smart watches seem so dumb.
Smart watches require smart phones. Humanity has in their possession about 2 billion smart phones. Humanity owns about 3.4 billion analog phones. You can tell this by looking at the two charts below. The first chart shows how many smart phones have been sold - by my definition, any phone that shipped with a data connection is a smart phone.
The first image, by Chris55 - Own work, CC BY-SA 4.0, shows the number of smart phones by region per year. In 2014, 33 % of the world had a smart phone, which is about 2 billion smart phones.
How many smart phones exist?
How many analog phones plus smart phones exist?
The second graph, also from Wikipedia, tells us how many mobile phones of any kind exist per year in different regions, including world-wide. Since phones are either smart phones or analog phones, the difference tells us how many non-smart phones are around at any point in time.
Smart phone possession is going up at 5% of the population per year, starting from a base penetration of 22 % in 2014, if this continues, everyone on earth will have a smart phone in 8 years or 2022. Not so far away.
This means the market for smart watches will soon be everyone on the earth. That's a pretty big market. And smart watches depend on the most valuable thing on earth: intelligence. We've spent more money making more intelligence per cubic inch than on almost anything else you can think of. And smart watches are only going to get smarter. Lots of people think Moore's law which predicted that the density of processing power doubles every 2 years or so is about to expire, but it has at least another decade to go in my opinion and in Intel's (Moore's company.)
This means the market for smart watches will soon be everyone on the earth. That's a pretty big market. And smart watches depend on the most valuable thing on earth: intelligence. We've spent more money making more intelligence per cubic inch than on almost anything else you can think of. And smart watches are only going to get smarter. Lots of people think Moore's law which predicted that the density of processing power doubles every 2 years or so is about to expire, but it has at least another decade to go in my opinion and in Intel's (Moore's company.)
How much information can a Smart Watch send you?
First, let's talk about how well a smart phone can understand you. A smart phone interface has about 500 - 1000 hit positions and about 10 gestures, so the bandwidth is about 5 - 10 kilobits per second (kb/s). That bandwidth is 10 times the bandwidth of a smart watch. It's no wonder that smart watches seem dumb compared to smart phones. Actually they are blind, deaf and dumb.
A smart watch screens is 10 times smaller than a smart phone's screen. The amount of information available to a human from a smart watch is also 10 times smaller. How much information would that be? Screen resolutions are about 400 pixels/inch. For a smart phone that's about a high resolution television screen, or about 2 million pixels. Each pixel can hold about 3 bytes or 24 bits of information, so about 6 megabytes (MB) or 48 megabits (Mb) of information per screen. Since the screen can be updated 30 times per second, the total bandwidth would be 1500 Mb/s or 300,000 times as much as your channel to it. A smart watch is 10 times smaller, and it's bandwidth is about 150 Mb/s or about 3000 times faster than the fastest analog modem made 20 years ago in the 1990's. A smart watch can talk to you you 300,000 times faster than you can talk to it. That might be why it spends a lot of the time doing nothing. It's waiting for you to say something.
Why is my Smart Watch so dumb?
Actually let's first see if there is any faster way to communicate with the smart watch instead of just using gestures on the screen (which is 500 to 1000 b/s)? You could talk to it, for instance. Speech has a total information throughput about the same as one of those modems built in the 1990's. That would be about 56 kb/s. You can see examples of this type of interface in Siri or on 'Okay, Google'. Today, it's like talking to an imbecile, they barely understand you most of the time and it takes them forever, but things will only get better. Within 10 years, smart watches will understand speech at normal speeds. Most people don't know this but computers actually understand individual spoken words better than humans today, computers understand about 95%, while humans are below 90% [citation needed]. It's the context and the gestures and the tone that humans get and computers don't. Humans have great feedback systems for communication, computers don't, but they will. It'sis just a matter of time. When computers pass the Turing test, smart watches will too; see what Ray Kurzweil thinks.
Another alternative would be to do some type of gesture interface, but use a camera as a sensor, not a touch screen. In that case you can do more precise measurements, so you can have a much larger input bandwidth. Cameras easily do HDTV style information flows, which is similar to smart phone screens. So it seems that the input to a smart watch will one day be as good as the output of your smartphone. So eventually a smart watch won't be deaf, dumb and blind. I think that will happen in about 8 years.Why do I say that? People are willing to pay about $500 dollars for a smart phone every two years. When smart phones are used by everyone in the whole world there will be money to pay about 24 million very smart people people to work on smart phones. That's a lot. They will get better. And these same people will be working on smart watches. Granted that's not as big an industry as automobiles or furniture or energy, but it's plenty big.
Another alternative would be to do some type of gesture interface, but use a camera as a sensor, not a touch screen. In that case you can do more precise measurements, so you can have a much larger input bandwidth. Cameras easily do HDTV style information flows, which is similar to smart phone screens. So it seems that the input to a smart watch will one day be as good as the output of your smartphone. So eventually a smart watch won't be deaf, dumb and blind. I think that will happen in about 8 years.Why do I say that? People are willing to pay about $500 dollars for a smart phone every two years. When smart phones are used by everyone in the whole world there will be money to pay about 24 million very smart people people to work on smart phones. That's a lot. They will get better. And these same people will be working on smart watches. Granted that's not as big an industry as automobiles or furniture or energy, but it's plenty big.
Smart Watches have other problems, too.
The current batch of Android based smart watches have all of the problems of a new and cheap technology. Instead of an analog watch that tends to wear out after a decade or two, and can only tell you a few 100 things per second, at most, and needs to be rewound every night; you get a very high bandwidth output that can emulate any known watch. A smart watch has a bandwidth of 20 to 15 million times better than an analog watch. Even though they are cheap and crappy, they will win because they deliver more of what they end user wants for less money. They deliver information much better than an analog watch can. And can be programmed to deliver different information at different times. They will take over the market.
Today a smart watch is better than an analog watch, except... it runs out of power faster, it crashes randomly, it only fully works when connected to a network, it's hard to upgrade, it only works with a single type of smart phone, ...
Here's what they have to fix:
Today a smart watch is better than an analog watch, except... it runs out of power faster, it crashes randomly, it only fully works when connected to a network, it's hard to upgrade, it only works with a single type of smart phone, ...
Here's what they have to fix:
- The narrow bandwidth back to the Internet. It is interminable that this is essentially a Blue Tooth connection maxed out at about 1 Mb/s. This is too small by a factor of at least 500.
- It needs to be simple to connect it to its smart phone, and the channel bandwidth between the two needs to be safe and secure and speedy. It'd be nice if this didn't use too much energy.
- It shouldn't require rebooting to get something too work.
- It shouldn't reboot on its own at random times.
- It needs to use less energy or have a higher density storage mechanism.
- It needs to be flatter and larger so it can have a bigger screen.
- It needs to be able to listen and understand you.
- It one day needs to be able to see you, understand you and understand the surrounding environment. Hmm. The same requirements as a self-driving-car.
Smart Watches will rule the world, someday.
Remember those 24 million smart people? They will be working tirelessly away at making smart phones and smart watches smarter. Today these smart watches are cheap and fairly useless. However we sell many more of them that we do of analog watches. This just happened last year in 2015. It's only going to get better as time goes on. Smart phones with smart watches will take over the world, because digital is so much more useful that analog. In fact, this has been happening in many, many industries around the world. The transition to digital has made all kinds of industries smarter. Digital processors are the way that industry packages intelligence into machines.
How much machine intelligence is our there? Enough to simulate a human brain? You betcha!
How much machine intelligence do you need to emulate a human's brain?
The human brain has about one 100 billion neurons. Each neuron can fire at around 25 Hz, and it can be connected to 1000's of other neurons. So each neuron has about 25 kB/s of bandwidth of communication. How many bits of information does it take to represent a neuron? Just to find a neuron takes about 40 bits. Each neuron is connected to a 1000 other neurons, so this connectivity matrix takes about 5,000 bytes, or 5kB as us computer scientists say. Then each neuron has some internal state to remember and some algorithm to generate pulses given the inputs, say another 5 kB. This means that you need about a million billion bytes to represent the brain (10 kB x 100 billion.) The maximum memory today's computers can hold is about 8,000 GB or 8,000 billion bytes. Since we need a million bytes we'll need about 125 fully decked out machines to hold that neuron map. Today that memory would cost you about $8000 per machine. So assuming there's another $2000 for everything else, you could simulate an entire brain for about $1 million dollars in hardware.
Wait, we have to check something else. How fast can these neurons talk to each other? Looking at our numbers above, there's about 1 billion neurons on each machine and they are sending altogether about 25,000 billion bytes of information out each second. The fastest ethernetconnections top out at about 100 billion bytes per second. Could I fit 256 of these in one computer? Probably you can buy 1000 Gb ethernet adapter for $25 if they use the USB connector. This puts us down by a factor or 10 from what we need. So we either run the simulation 10 times slower than real life, or we split the memory up over 10 times as many computers to run it in real time.
So to simulate a human brain I need 1000 computers with 800 GB of memory each and 250 1000 Gb Ethernet connections each. These computers are still around $10,000 each, with the memory and networking cards. So I can simulate a human brain with about $10 million of hardware. Of course that's assuming that each neuron connects to 1000 other neurons on average, if the average is twice as high, I've got to buy twice as many computers. And I need to have a lot of fast networking switches to keep this all working in real time. They cost about $5000 for each computer, so they add about 50% to the cost, then I have to power and house these computers, say another $5000. So the total costs of simulating one human brain: about $20 million. That's today.
And this is real stuff. This Japanese team just used one of their supercomputers to simulate one second of one percent of your brain. They used the open source software: Neural Simulation Technology to do the simulation. The needed about a 750,000 processors, 1.4 GB of memory and it took 40 minutes. You can easily buy 12 core chips that are smart enough to do this processing, so we would need 60 processing chips on each computer, although, these cores are smarter by a factor of at least 10 so 32 processing chips should be 4x overkill, no problem there. Oh wait, I forgot that factor of 250 (40 minutes to one second.) So it seems like processing power is the bottleneck. Even though these machines can provide the networking bandwidth, it takes 60 more times cpu power then they currently have. Rats! I either need 60 times as many machines or I run the simulation 60 times slower than real time. 60 times as many machines would mean that the investment would be 60,000 computers and cost closer to $1.2 billion dollars. Or we need to make that simulation software algorithm 60 times faster. For a billion dollars, I think I might put my efforts there.
Coincidentally, Google now sells compute engine. For $2M per day you can get about 700,000 cores, which was what our Super computer Japanese friends were using. That's only for the cpu cores, you have to pay more for networking. Let's guess that the networking doubles the prices. So you can buy 40 minutes of this time (to simulate one second of a human brain) for about $100,000. That seems like a better deal that buying your own hardware and building your own data centers. Google is rumored to have over a million computers 5 or 6 years ago. Certainly with all the capital spending they do they have more than that today.
Think about all the computers in the world, there's about 2 billion out there today. And I bet there's about one backend server for each one. So that would be about 4 billion computers. You could simulate about 70,000 human brains at the 250 slow down rate or about 250 in real time.
So exactly what is the Singularity? By my definition the Singularity is already here in a bunch of places. To me, the Singularity means the time when computers are smarter than humans at some task. Like Chess, or Go. Already happened. Or coaching baseball. Or coaching basketball. The Singularity is already here in a bunch of industries: search, cedit cards, advertising, simulating rockets; the list is really, really long. And in 2056 computers will be able to be as good as everyone in the world at everything. If that doesn't meet the definition of the Singularity, nothing will.
And it's only going to happen if the aliens don't get here first. In my opinion, the main reason that the aliens will be here soon.
Wait, we have to check something else. How fast can these neurons talk to each other? Looking at our numbers above, there's about 1 billion neurons on each machine and they are sending altogether about 25,000 billion bytes of information out each second. The fastest ethernetconnections top out at about 100 billion bytes per second. Could I fit 256 of these in one computer? Probably you can buy 1000 Gb ethernet adapter for $25 if they use the USB connector. This puts us down by a factor or 10 from what we need. So we either run the simulation 10 times slower than real life, or we split the memory up over 10 times as many computers to run it in real time.
So to simulate a human brain I need 1000 computers with 800 GB of memory each and 250 1000 Gb Ethernet connections each. These computers are still around $10,000 each, with the memory and networking cards. So I can simulate a human brain with about $10 million of hardware. Of course that's assuming that each neuron connects to 1000 other neurons on average, if the average is twice as high, I've got to buy twice as many computers. And I need to have a lot of fast networking switches to keep this all working in real time. They cost about $5000 for each computer, so they add about 50% to the cost, then I have to power and house these computers, say another $5000. So the total costs of simulating one human brain: about $20 million. That's today.
And this is real stuff. This Japanese team just used one of their supercomputers to simulate one second of one percent of your brain. They used the open source software: Neural Simulation Technology to do the simulation. The needed about a 750,000 processors, 1.4 GB of memory and it took 40 minutes. You can easily buy 12 core chips that are smart enough to do this processing, so we would need 60 processing chips on each computer, although, these cores are smarter by a factor of at least 10 so 32 processing chips should be 4x overkill, no problem there. Oh wait, I forgot that factor of 250 (40 minutes to one second.) So it seems like processing power is the bottleneck. Even though these machines can provide the networking bandwidth, it takes 60 more times cpu power then they currently have. Rats! I either need 60 times as many machines or I run the simulation 60 times slower than real time. 60 times as many machines would mean that the investment would be 60,000 computers and cost closer to $1.2 billion dollars. Or we need to make that simulation software algorithm 60 times faster. For a billion dollars, I think I might put my efforts there.
How many human minds could you simulate today?
Now how many companies have 60,000 machines? That's about the size of one large data center. Microsoft claims to have 1 million computers, and that was three years ago. One million computers with the right networking can be used to simulate about 16 human brains.Coincidentally, Google now sells compute engine. For $2M per day you can get about 700,000 cores, which was what our Super computer Japanese friends were using. That's only for the cpu cores, you have to pay more for networking. Let's guess that the networking doubles the prices. So you can buy 40 minutes of this time (to simulate one second of a human brain) for about $100,000. That seems like a better deal that buying your own hardware and building your own data centers. Google is rumored to have over a million computers 5 or 6 years ago. Certainly with all the capital spending they do they have more than that today.
Think about all the computers in the world, there's about 2 billion out there today. And I bet there's about one backend server for each one. So that would be about 4 billion computers. You could simulate about 70,000 human brains at the 250 slow down rate or about 250 in real time.
What does this say about the Singularity (the nerd rapture)?
Let me say that again. Given the current number of computers on the earth, you could simulate 250 human brains in real time. And it's only going to get faster (remember Moore's Law?) If this continues to double every two years, computers will arguably be smarter than all of humanity in 2056 since they could simulate all of humanity at that time.So exactly what is the Singularity? By my definition the Singularity is already here in a bunch of places. To me, the Singularity means the time when computers are smarter than humans at some task. Like Chess, or Go. Already happened. Or coaching baseball. Or coaching basketball. The Singularity is already here in a bunch of industries: search, cedit cards, advertising, simulating rockets; the list is really, really long. And in 2056 computers will be able to be as good as everyone in the world at everything. If that doesn't meet the definition of the Singularity, nothing will.
Why do we call it the nerd rapture?
It's called the nerd rapture because it's basically the equivalent of the various religious raptures that ares supposed to occur at some random time in the future when God takes everyone to heaven, or some such thing where everyone is transcended. There's one big different between the nerd rapture and religious rapture. One of them is actually going to happen, it isn't myth, it isn't made up, it's predictable, and it's going to happen in your lifetime. And it's going to have a big effect on the world. What effect? Who knows. It could be nirvana or it could be hell. And it's up to you which one it is. It all depends on who you leave in charge for the next 40 years. This could arguable be the most important turning point for humanity. I hope we pay attention and use it for good. Maybe we could use all that compute power to intelligently design morals instead of reading what others have said who didn't have this huge advantage. It could give humanity a real chance to make things better in the universe. One can only hope.And it's only going to happen if the aliens don't get here first. In my opinion, the main reason that the aliens will be here soon.
Thanks for reading.
-Dr. Mike
-Dr. Mike
Comments
Post a Comment