Ramez Naam, author of Nexus and Crux (two books I enjoyed and recommend), has recently put together a few guest posts for Charlie Stross (another author I love). The posts are The Singularity Is Further Than It Appears and Why AIs Won't Ascend in the Blink of an Eye.
They're both excellent posts, and I'd recommend reading them in full before continuing here.
I'd like to offer a slight rebuttal and explain why I think the singularity is still closer than it appears.
But first, I want to say that I very much respect Ramez, his ideas and writing. I don't think he's wrong and I'm right. I think the question of the singularity is a bit more like Drake's Equation about intelligent extraterrestrial life: a series of probabilities, the values of which are not known precisely enough to determine the "correct" output value with strong confidence. I simply want to provide a different set of values for consideration than the ones that Ramez has chosen.
First, let's talk about definitions. As Ramez describes in his first article, there are two versions of singularity often talked about.
The hard takeoff is one in which an AI rapidly creates newer, more intelligent versions of itself. Within minutes, days, or weeks, the AI has progressed from a level 1 AI to a level 20 grand-wizard AI, far beyond human intellect and anything we can comprehend. Ramez doesn't think this will happen for a variety of reasons, one of which is the exponential difficulty involved in creating successively more complex algorithm (the argument he lays out in his second post).
I agree. I don't see a hard takeoff. In addition to the reasons Ramez stated, I also believe it takes so long to test and qualify candidates for improvement that successive iteration will be slow.
Let's imagine the first AI is created and runs on an infrastructure of 10,000 computers. Let's further assume the AI is composed of neural networks and other similar algorithms that require training on large pools of data. The AI will want to test many ideas for improvements, each requiring training. The training will be followed by multiple rounds of successively more comprehensive testing: first the AI needs to see if the algorithm appears to improve a select area of intelligence, but then it will want to run regressive tests to ensure no other aspect of its intelligence or capabilities is adversely impacted. If the AI wants to test 1,000 ideas for improvements, and each idea requires 10 hours of training, 1 hour of assessment, and averages 1 hour of regressive testing, it would take 1.4 years to complete a round of improvements. Parallelism is the alternative, but remember that first AI is likely to be a behemoth, require 10,000 computers to run. It's not possible to get that much parallelism.
The soft takeoff is one in which an artificial general intelligence (AGI) is created and gradually improved. As Ramez points out, that first AI might be on the order of human intellect, but it's not smarter than the accumulated intelligence of all the humans that created it: many tens of thousands of scientists will collaborate to build the first AGI.
This is where we start to diverge. Consider a simple domain like chess playing computers. Since 2005, chess software running on commercially available hardware can outplay even the strongest human chess players. I don't have data, but I suspect the number of very strong human chess players is somewhere in the hundreds or low thousands. However, the number of computers capable of running the very best chess playing software is in the millions or hundreds of millions. The aggregate chess playing capacity of computers is far greater than that of humans, because the best chess playing program can be propagated everywhere.
So too, AGI will be propagated everywhere. But I just argued that those first AI will require tens of thousands computers, right? Yes, except thanks to Moore's Law (the observation that computing power tends to double every 18 months), the same AI that required 10,000 computers will need a mere 100 computers ten years later and just a single computer another ten years after that. Or an individual AGI could run up to 10,000 times faster. That speed-up alone means something different when it comes to intelligence: to have a single being with 10,000 times the experience and learning and practice that a human has.
Even Ramez agrees that it will be feasible to have destructive human brain uploads approximating human intelligence around 2040: "Do the math, and it appears that a super-computer capable of simulating an entire human brain and do so as fast as a human brain should be on the market by roughly 2035 - 2040. And of course, from that point on, speedups in computing should speed up the simulation of the brain, allowing it to run faster than a biological human's."
This is the soft takeoff: from a single AGI at some point in time to an entire civilization of that AGI twenty years later, all running at faster than human intellect speeds. A race consisting of an essentially alien intelligence, cohabiting the planet with us. Even if they don't experience an intelligence explosion as Verner Vinge described, the combination of fast speeds, aggregate intelligence, and inherently different motivations will create an unknowable future that likely out of our control. And that's very much a singularity.
But Ramez questions whether we can even achieve an AGI comparable to a human in the first place. There's this pesky question of sentience and consciousness. Please go read Ramez's first article in full, I don't want you to think I'm summarizing everything he said here, but he basically cites three points:
1) No one's really sure how to do it. AI theories have been around for decades, but none of them has led to anything that resembles sentience.
This is a difficulty. One analogy that comes to mind is the history of aviation. For nearly a hundred years prior to the Wright Brothers, heavier than air flight was being studied, with many different gliders created and flown. It was the innovation of powered engines that made heavier than air flight practically possible, and which led to rapid innovation. Perhaps we just don't yet have the equivalent yet in AI. We've got people learning how to make airfoils and control services and airplane structure, and we're just waiting for the engine to show up.
We also know that nature evolved sentience without any theory of how to do it. Having a proof point is powerful motivation.
2) There's a huge lack of incentive. Would you like a self-driving car that has its own opinions? That might someday decide it doesn't feel like driving you where you want to go?
There's no lack of incentive. As James Barrat detailed in Our Final Invention, there are billions of dollars being poured into building AGI, both in big profile projects like the US BRAIN project and Europe's Human Brain Project, as well as countless smaller AI companies and research projects.
There's plenty of human incentive, too. How many people were inspired by Star Trek's Data? At a recent conference, I asked attendees who would want Data as a friend, and more than half the audience's hands went up. Among the elderly, loneliness is a very real issue that could be helped with AGI companionship, and many people might choose an artificial psychologist for reasons of confidence, cost, and convenience. All of these require at least the semblance of opinions.
More than that, we know we want initiative. If we have a self-driving car, we expect that it will use that initiative to find faster routes to destinations, possibly go around dangerous neighborhoods, and take necessary measures to avoid an accident. Indeed, even Google Maps has an "opinion" of the right way to get somewhere that often differs from my own. It's usually right.
If we have an autonomous customer service agent, we'll want it to flexibly meet business goals including pleasing the customer while controlling cost. All of these require something like opinions and sentience: goals, motivation to meet those goals, and mechanisms to flexibly meet those goals.
3) There are ethical issues. If we design an AI that truly is sentient, even at slightly less than human intelligence we'll suddenly be faced with very real ethical issues. Can we turn it off?
I absolutely agree that we've got ethical issues with AGI, but that hasn't stopped us from creating other technology (nuclear bombs, bio-weapons, internal combustion engine, the transportation system) that also has ethical issues.
In sum, Ramez brings up great points, and he may very well be correct: the singularity might be a hundred years off instead of twenty or thirty.
However, the discussion around the singularity is also one about risk. Having artificial general intelligence running around, potentially in control of our computing infrastructure, may be risky. What happens if the AI has different motivations than us? What if it decides we'd be happier and less destructive if we're all drugged? What if it just crashes and accidentally shuts down the entire electrical grid? (Read James Barrat's Our Final Invention for more about the risks of AI.)
Ramez wrote Infinite Resource: The Power of Ideas on a Finite Planet, a wonderful and optimistic book about how science and technology are solving many resource problems around the world. I think it's a powerful book because it gives us hope and proof points that we can solve the problems facing us.
Unfortunately, I think the argument that the singularity is far off is different and problematic because it denies the possibility of problems facing us. Instead of encouraging us to use technology to address the issues that could arise with the singularity, the argument instead concludes the singularity is either unlikely or simply a long time away. With that mindset, we're less likely as a society to examine both AI progress and take steps to reduce the risks of AGI.
On the other hand, if we can agree that the singularity is a possibility, even just a modest possibility, then we may spur more discussion and investment into the safety and ethics of AGI.
Showing posts with label artificial intelligence. Show all posts
Showing posts with label artificial intelligence. Show all posts
Recent Rate of Computer Processing Growth
I was having a discussion with a group of writers about the technological singularity, and several asserted that the rate of increasing processor power was declining. They backed it up with a chart showing that the increase in MIPS per unit of clock speed stalled about ten years ago.
If computer processing speeds fail to increase exponentially, as they have for the last forty years, this will throw off many different predictions for the future, and dramatically decreases the likelihood of human-grade AI arising.
I did a bit of research last night and this morning. Using the chart of historical computer speeds from Wikipedia, and I placed a few key intervals in a spreadsheet and found:
By no means is the list of MIPS ratings exhaustive, but it does give us a general idea of what's going on. The data shows the rate of CPU speed increases has declined in the last ten years.
I split up the last ten years:
Five years isn't much of a long term trend, and there are some processors missing from the end of the matrix. The Intel Xeon X5675, a 12 core processor isn't shown, and it's twice as powerful as the Intel Core i7 4770k that's the bottom row on the MIPS table. If we substitute the Xeon processor, we find the growth rate from 2008 to 2012 was 31% annually, a more respectable improvement.
However, I've been tracking technology trends for a while (see my post on How to Predict the Future), and I try to use only those computers and devices I've personally owned. There's always something faster out there, but it's not what people have in their home, which is what I'm interested in.
I also know that my device landscape has changed over the last five years. In 2008, I had a laptop (Windows Intel Core 2 T7200) and a modest smartphone (a Treo 650). In 2013, I have a laptop (MBP 2.6 GHz Core i7), a powerful smartphone (Nexus 5), and a tablet (iPad Mini). I'm counting only my own devices and excluding those from my day job as a software engineer.
It's harder to do this comparison, because there's no one common benchmark among all these processors. I did the best I could to determine DMIPS for each, converting GeekBench cores for the Mac, and using the closest available processor for mobile devices that had a MIPS rating.
When I compared my personal device growth in combined processing power, I found it increased 51% annually from 2008 to 2013, essentially the same rate as for the longer period 1996 through 2011 (47%), which is what I use for my long-term predictions.
What does all this mean? Maybe there is a slight slow-down in the rate at which computing processing is increasing. Maybe there isn't. Maybe the emphasis on low-power computing for mobile devices and server farms has slowed down progress on top-end speeds, and maybe that emphasis will contribute to higher top-end speeds down the road. Maybe the landscape will move from single-devices to clouds of devices, in the same way that we already moved from single cores to multiple cores.
Either way, I'm not giving up on the singularity yet.
If computer processing speeds fail to increase exponentially, as they have for the last forty years, this will throw off many different predictions for the future, and dramatically decreases the likelihood of human-grade AI arising.
I did a bit of research last night and this morning. Using the chart of historical computer speeds from Wikipedia, and I placed a few key intervals in a spreadsheet and found:
- From 1972 to 1985: MIPS grew by 19% per year.
- From 1985 to 1996: MIPS grew by 43% per year.
- From 1996 to 2003: MIPS grew by 51% per year.
- From 2003 to 2013: MIPS grew by 29% per year.
By no means is the list of MIPS ratings exhaustive, but it does give us a general idea of what's going on. The data shows the rate of CPU speed increases has declined in the last ten years.
I split up the last ten years:
- From 2003 to 2008: MIPS grew by 53% per year.
- From 2008 to 2013: MIPS grew by 9% per year.
Five years isn't much of a long term trend, and there are some processors missing from the end of the matrix. The Intel Xeon X5675, a 12 core processor isn't shown, and it's twice as powerful as the Intel Core i7 4770k that's the bottom row on the MIPS table. If we substitute the Xeon processor, we find the growth rate from 2008 to 2012 was 31% annually, a more respectable improvement.
However, I've been tracking technology trends for a while (see my post on How to Predict the Future), and I try to use only those computers and devices I've personally owned. There's always something faster out there, but it's not what people have in their home, which is what I'm interested in.
I also know that my device landscape has changed over the last five years. In 2008, I had a laptop (Windows Intel Core 2 T7200) and a modest smartphone (a Treo 650). In 2013, I have a laptop (MBP 2.6 GHz Core i7), a powerful smartphone (Nexus 5), and a tablet (iPad Mini). I'm counting only my own devices and excluding those from my day job as a software engineer.
It's harder to do this comparison, because there's no one common benchmark among all these processors. I did the best I could to determine DMIPS for each, converting GeekBench cores for the Mac, and using the closest available processor for mobile devices that had a MIPS rating.
When I compared my personal device growth in combined processing power, I found it increased 51% annually from 2008 to 2013, essentially the same rate as for the longer period 1996 through 2011 (47%), which is what I use for my long-term predictions.
What does all this mean? Maybe there is a slight slow-down in the rate at which computing processing is increasing. Maybe there isn't. Maybe the emphasis on low-power computing for mobile devices and server farms has slowed down progress on top-end speeds, and maybe that emphasis will contribute to higher top-end speeds down the road. Maybe the landscape will move from single-devices to clouds of devices, in the same way that we already moved from single cores to multiple cores.
Either way, I'm not giving up on the singularity yet.
The Danger of DARPA's Crowdsourced Cybersecurity
The Pentagon's research arm, DARPA, wants to crowdsource a fully automated cyber defense system, and they're offering a two million dollar prize:
On the other hand, this is scary. They're asking competitors to marry artificial intelligence with cyber defense systems. Cyber defense requires a solid understanding of cyber offense, and aggressive defensive capabilities could be nearly as destructive as offensive capabilities. Cyber defense software could decide to block a threatening virus with a counter-virus, or shut down parts of the Internet to stop or slow infection.
Artificial intelligence has taking over stock trading, and look where that's gotten us. Trading AI has become so sophisticated it is described in terms of submarine warfare, with offensive and defensive capabilities.
I don't doubt that the competition will advance cyber defense. But the side effect will be a radical increase in cyber offense, as well as a system in which both side operate at algorithmic speeds.
Full information about the Cyber Grand Challenge, including rules and registration, is available on DARPA's website.
The so-called "Cyber Grand Challenge" will take place over the next three years, which seems like plenty of time to write a few lines of code. But DARPA's not just asking for any old cyber defense system. They want one "with reasoning abilities exceeding those of human experts" that "will create its own knowledge." They want it to deflect cyberattacks, not in a matter of days—which is how the Pentagon currently works—but in a matter of hours or even seconds. That's profoundly difficult.On the one hand, this is brilliant. I can easily imagine some huge leaps forward made as a result of the contest. The Netflix Prize advanced recommendation algorithms while the DARPA Grand Prize gave us autonomous cars. Clearly competitions work, especially in this domain where the barrier to entry is low.
On the other hand, this is scary. They're asking competitors to marry artificial intelligence with cyber defense systems. Cyber defense requires a solid understanding of cyber offense, and aggressive defensive capabilities could be nearly as destructive as offensive capabilities. Cyber defense software could decide to block a threatening virus with a counter-virus, or shut down parts of the Internet to stop or slow infection.
Artificial intelligence has taking over stock trading, and look where that's gotten us. Trading AI has become so sophisticated it is described in terms of submarine warfare, with offensive and defensive capabilities.
I don't doubt that the competition will advance cyber defense. But the side effect will be a radical increase in cyber offense, as well as a system in which both side operate at algorithmic speeds.
Full information about the Cyber Grand Challenge, including rules and registration, is available on DARPA's website.
The Last Firewall is here!
I'd like to announce that The Last Firewall is available!
In the year 2035, robots, artificial intelligences, and neural implants have become commonplace. The Institute for Applied Ethics keeps the peace, using social reputation to ensure that robots and humans don't harm society or each other. But a powerful AI named Adam has found a way around the restrictions.
Catherine Matthews, nineteen years old, has a unique gift: the ability to manipulate the net with her neural implant. Yanked out of her perfectly ordinary life, Catherine becomes the last firewall standing between Adam and his quest for world domination.Two+ years in the making, I'm just so excited to finally release this novel. As with my other novels, I explore themes of what life will be like with artificial intelligence, how we deal with the inevitable man-vs-machine struggle, and the repercussions of using online social reputation as a form of governmental control.
 |
| The Last Firewall joins its siblings. |
Buy it now: Amazon Kindle, in paperback, and Kobo eReader.
(Other retailers coming soon.)
(Other retailers coming soon.)
I hope you enjoy it! Here is some of the early praise for the book:
“Awesome near-term science fiction.” – Brad Feld, Foundry Group managing director
“An insightful and adrenaline-inducing tale of what humanity could become and the machines we could spawn.” – Ben Huh, CEO of Cheezburger
“A fun read and tantalizing study of the future of technology: both inviting and alarming.” – Harper Reed, former CTO of Obama for America, Threadless
"A fascinating and prescient take on what the world will look like once computers become smarter than people. Highly recommended." – Mat Ellis, Founder & CEO Cloudability
“A phenomenal ride through a post-scarcity world where humans are caught between rogue AIs. If you like having your mind blown, read this book!” – Gene Kim, author of The Phoenix Project: A Novel About IT, DevOps, and Helping Your Business Win
“The Last Firewall is like William Gibson had a baby with Tom Clancy and let Walter Jon Williams teach it karate. Superbly done.” – Jake F. Simons, author of Wingman and Train Wreck
How to Predict the Future
Everyone would like a sure-fire way to predict the future. Maybe you’re thinking about startups to invest in, or making decisions about where to place resources in your company, or deciding on a future career, or where to live. Maybe you just care about what things will be like in 10, 20, or 30 years.
This is a repost of an article I originally wrote for Feld.com. If you enjoyed this post, please check out my novels Avogadro Corp: The Singularity Is Closer Than It Appears and A.I. Apocalypse, near-term science-fiction novels about realistic ways strong AI might emerge. They’ve been called “frighteningly plausible”, “tremendous”, and “thought-provoking”.
There are many techniques to think logically about the future, to inspire idea creation, and to predict when future inventions will occur.
I’d like to share one technique that I’ve used successfully. It’s proven accurate on many occasions. And it’s the same technique that I’ve used as a writer to create realistic technothrillers set in the near future. I’m going to start by going back to 1994.
Predicting Streaming Video and the Birth of the Spreadsheet
There seem to be two schools of thought on how to predict the future of information technology: looking at software or looking at hardware. I believe that looking at hardware curves is always simpler and more accurate.
This is the story of a spreadsheet I've been keeping for almost twenty years.
In the mid-1990s, a good friend of mine, Gene Kim (founder of Tripwire and author of When IT Fails: A Business Novel) and I were in graduate school together in the Computer Science program at the University of Arizona. A big technical challenge we studied was piping streaming video over networks. It was difficult because we had limited bandwidth to send the bits through, and limited processing power to compress and decompress the video. We needed improvements in video compression and in TCP/IP - the underlying protocol that essentially runs the Internet.
The funny thing was that no matter how many incremental improvements researchers made (there were dozens of people working on different angles of this), streaming video always seemed to be just around the corner. I heard “Next year will be the year for video” or similar refrains many times over the course of several years. Yet it never happened.
Around this time I started a spreadsheet, seeding it with all of the computers I’d owned over the years. I included their processing power, the size of their hard drives, the amount of RAM they had, and their modem speed. I calculated the average annual increase of each of these attributes, and then plotted these forward in time.
I looked at the future predictions for “modem speed” (as I called it back then, today we’d called it internet connection speed or bandwidth). By this time, I was tired of hearing that streaming video was just around the corner, and I decided to forget about trying to predict advancements in software compression, and just look at the hardware trend. The hardware trend showed that internet connection speeds were increasing, and by 2005, the speed of the connection would be sufficient that we could reasonably stream video in real time without resorting to heroic amounts of video compression or miracles in internet protocols. Gene Kim laughed at my prediction.
Nine years later, in February 2005, YouTube arrived. Streaming video had finally made it.
The same spreadsheet also predicted we’d see a music downloading service in 1999 or 2000. Napster arrived in June, 1999.
The data has held surprisingly accurate over the long term. Using just two data points, the modem I had in 1986 and the modem I had in 1998, the spreadsheet predicts that I’d have a 25 megabit/second connection in 2012. As I currently have a 30 megabit/second connection, this is a very accurate 15 year prediction.
Why It Works Part One: Linear vs. Non-Linear
Without really understanding the concept, it turns out that what I was doing was using linear trends (advancements that proceed smoothly over time), to predict the timing of non-linear events (technology disruptions) by calculating when the underlying hardware would enable a breakthrough. This is what I mean by “forget about trying to predict advancements in software and just look at the hardware trend”.
It’s still necessary to imagine the future development (although the trends can help inspire ideas). What this technique does is let you map an idea to the underlying requirements to figure out when it will happen.
For example, it answers questions like these:
When will the last magnetic platter hard drive be manufactured?
|
2016. I plotted the growth in capacity of magnetic platter hard drives and flash drives back in 2006 or so, and saw that flash would overtake magnetic media in 2016.
|
When will a general purpose computer be small enough to be implanted inside your brain?
|
2030. Based on the continual shrinking of computers, by 2030 an entire computer will be the size of a pencil eraser, which would be easy to implant.
|
When will a general purpose computer be able to simulate human level intelligence?
|
Between 2024 and 2050, depending on which estimate of the complexity of human intelligence is selected, and the number of computers used to simulate it.
|
Wait, a second: Human level artificial intelligence by 2024? Gene Kim would laugh at this. Isn’t AI a really challenging field? Haven’t people been predicting artificial intelligence would be just around the corner for forty years?
Why It Works Part Two: Crowdsourcing
At my panel on the future of artificial intelligence at SXSW, one of my co-panelists objected to the notion that exponential growth in computer power was, by itself, all that was necessary to develop human level intelligence in computers. There are very difficult problems to solve in artificial intelligence, he said, and each of those problems requires effort by very talented researchers.
I don’t disagree, but the world is a big place full of talented people. Open source and crowdsourcing principles are well understood: When you get enough talented people working on a problem, especially in an open way, progress comes quickly.
I wrote an article for the IEEE Spectrum called The Future of Robotics and Artificial Intelligence is Open. In it, I examine how the hobbyist community is now building inexpensive unmanned aerial vehicle auto-pilot hardware and software. What once cost $20,000 and was produced by skilled researchers in a lab, now costs $500 and is produced by hobbyists working part-time.
Once the hardware is capable enough, the invention is enabled. Before this point, it can’t be done. You can’t have a motor vehicle without a motor, for example.
As the capable hardware becomes widely available, the invention becomes inevitable, because it enters the realm of crowdsourcing: now hundreds or thousands of people can contribute to it. When enough people had enough bandwidth for sharing music, it was inevitable that someone, somewhere was going to invent online music sharing. Napster just happened to have been first.
IBM’s Watson, which won Jeopardy, was built using three million dollars in hardware and had 2,880 processing cores. When that same amount of computer power is available in our personal computers (about 2025), we won’t just have a team of researchers at IBM playing with advanced AI. We’ll have hundreds of thousands of AI enthusiasts around the world contributing to an open source equivalent to Watson. Then AI will really take off.
(If you doubt that many people are interested, recall that more than 100,000 people registered for Stanford’s free course on AI and a similar number registered for the machine learning / Google self-driving car class.)
Of course, this technique doesn’t work for every class of innovation. Wikipedia was a tremendous invention in the process of knowledge curation, and it was dependent, in turn, on the invention of wikis. But it’s hard to say, even with hindsight, that we could have predicted Wikipedia, let alone forecast when it would occur.
(If one had the idea of an crowd curated online knowledge system, you could apply the litmus test of internet connection rate to assess when there would be a viable number of contributors and users. A documentation system such as a wiki is useless without any way to access it. But I digress...)
Objection, Your Honor
A common objection is that linear trends won’t continue to increase exponentially because we’ll run into a fundamental limitation: e.g. for computer processing speeds, we’ll run into the manufacturing limits for silicon, or the heat dissipation limit, or the signal propagation limit, etc.
I remember first reading statements like the above in the mid-1980s about the Intel 80386 processor. I think the statement was that they were using an 800 nm process for manufacturing the chips, but they were about to run into a fundamental limit and wouldn’t be able to go much smaller. (Smaller equals faster in processor technology.)
Semiconductor
manufacturing
processes
|
Source: Wikipedia
But manufacturing technology has proceeded to get smaller and smaller. Limits are overcome, worked around, or solved by switching technology. For a long time, increases in processing power were due, in large part, to increases in clock speed. As that approach started to run into limits, we’ve added parallelism to achieve speed increases, using more processing cores and more execution threads per core. In the future, we may have graphene processors or quantum processors, but whatever the underlying technology is, it’s likely to continue to increase in speed at roughly the same rate.
Why Predicting The Future Is Useful: Predicting and Checking
There are two ways I like to use this technique. The first is as a seed for brainstorming. By projecting out linear trends and having a solid understanding of where technology is going, it frees up creativity to generate ideas about what could happen with that technology.
It never occurred to me, for example, to think seriously about neural implant technology until I was looking at the physical size trend chart, and realized that neural implants would be feasible in the near future. And if they are technically feasible, then they are essentially inevitable.
What OS will they run? From what app store will I get my neural apps? Who will sell the advertising space in our brains? What else can we do with uber-powerful computers about the size of a penny?
The second way I like to use this technique is to check other people’s assertions. There’s a company called Lifenaut that is archiving data about people to provide a life-after-death personality simulation. It’s a wonderfully compelling idea, but it’s a little like video streaming in 1994: the hardware simply isn’t there yet. If the earliest we’re likely to see human-level AI is 2024, and even that would be on a cluster of 1,000+ computers, then it’s seems impossible that Lifenaut will be able to provide realistic personality simulation anytime before that.* On the other hand, if they have the commitment needed to keep working on this project for fifteen years, they may be excellently positioned when the necessary horsepower is available.
At a recent Science Fiction Science Fact panel, other panelists and most of the audience believed that strong AI was fifty years off, and brain augmentation technology was a hundred years away. That’s so distant in time that the ideas then become things we don’t need to think about. That seems a bit dangerous.
* The counter-argument frequently offered is “we’ll implement it in software more efficiently than nature implements it in a brain.” Sorry, but I’ll bet on millions of years of evolution.
How To Do It
This article is How To Predict The Future, so now we’ve reached the how-to part. I’m going to show some spreadsheet calculations and formulas, but I promise they are fairly simple. There’s three parts to to the process: Calculate the annual increase in a technology trend, forecast the linear trend out, and then map future disruptions to the trend.
Step 1: Calculate the annual increase
It turns out that you can do this with just two data points, and it’s pretty reliable. Here’s an example using two personal computers, one from 1996 and one from 2011. You can see that cell B7 shows that computer processing power, in MIPS (millions of instructions per second), grew at a rate of 1.47x each year, over those 15 years.
A
|
B
|
C
| |
1
|
MIPS
|
Year
| |
2
|
Intel Pentium Pro
|
541
|
1996
|
3
|
Intel Core i7 3960X
|
177730
|
2011
|
4
| |||
5
|
Gap in years
|
15
|
=C3-C2
|
6
|
Total Growth
|
328.52
|
=B3/B2
|
7
|
Rate of growth
|
1.47
|
=B6^(1/B5)
|
I like to use data related to technology I have, rather than technology that’s limited to researchers in labs somewhere. Sure, there are supercomputers that are vastly more powerful than a personal computer, but I don’t have those, and more importantly, they aren’t open to crowdsourcing techniques.
I also like to calculate these figures myself, even though you can research similar data on the web. That’s because the same basic principle can be applied to many different characteristics.
Step 2: Forecast the linear trend
The second step is to take the technology trend and predict it out over time. In this case we take the annual increase in advancement (B$7 - previous screenshot), raised to an exponent of the number of elapsed years, and multiply it by the base level (B$11). The formula displayed in cell C12 is the key one.
A
|
B
|
C
| |
10
|
Year
|
Expected MIPS
|
Formula
|
11
|
2011
|
177,730
|
=B3
|
12
|
2012
|
261,536
|
=B$11*(B$7^(A12-A$11))
|
13
|
2013
|
384,860
| |
14
|
2014
|
566,335
| |
15
|
2015
|
833,382
| |
16
|
2020
|
5,750,410
| |
17
|
2025
|
39,678,324
| |
18
|
2030
|
273,783,840
| |
19
|
2035
|
1,889,131,989
| |
20
|
2040
|
13,035,172,840
| |
21
|
2050
|
620,620,015,637
|
I also like to use a sanity check to ensure that what appears to be a trend really is one. The trick is to pick two data points in the past: one is as far back as you have good data for, the other is halfway to the current point in time. Then run the forecast to see if the prediction for the current time is pretty close. In the bandwidth example, picking a point in 1986 and a point in 1998 exactly predicts the bandwidth I have in 2012. That’s the ideal case.
Step 3: Mapping non-linear events to linear trend
The final step is to map disruptions to enabling technology. In the case of the streaming video example, I knew that a minimal quality video signal was composed of a resolution of 320 pixels wide by 200 pixels high by 16 frames per second with a minimum of 1 byte per pixel. I assumed an achievable amount for video compression: a compressed video signal would be 20% of the uncompressed size (a 5x reduction). The underlying requirement based on those assumptions was an available bandwidth of about 1.6mb/sec, which we would hit in 2005.
In the case of implantable computers, I assume that a computer of the size of a pencil eraser (1/4” cube) could easily be inserted into a human’s skull. By looking at physical size of computers over time, we’ll hit this by 2030:
Year
|
Size
(cubic inches)
|
Notes
|
1986
|
1782
|
Apple //e with two disk drives
|
2012
|
6.125
|
Motorola Droid 3
|
Elapsed years
|
26
| |
Size delta
|
290.94
| |
Rate of shrinkage per year
|
1.24
| |
Future Size
| ||
2012
|
6.13
| |
2013
|
4.92
| |
2014
|
3.96
| |
2015
|
3.18
| |
2020
|
1.07
| |
2025
|
0.36
| |
2030
|
0.12
|
Less than 1/4 inch on a side cube. Could easily fit in your skull.
|
2035
|
0.04
| |
2040
|
0.01
|
This is a tricky prediction: traditional desktop computers have tended to be big square boxes constrained by the standardized form factor of components such as hard drives, optical drives, and power supplies. I chose to use computers I owned that were designed for compactness for their time. Also, I chose a 1996 Toshiba Portege 300CT for a sanity check: if I project the trend between the Apple //e and Portege forward, my Droid should be about 1 cubic inch, not 6. So this is not an ideal prediction to make, but it’s still clues us in about the general direction and timing.
The predictions for human-level AI are more straightforward, but more difficult to display, because there’s a range of assumptions for how difficult it will be to simulate human intelligence, and a range of projections depending on how many computers you can bring to pair on the problem. Combining three factors (time, brain complexity, available computers) doesn’t make a nice 2-axis graph, but I have made the full human-level AI spreadsheet available to explore.
I’ll leave you with a reminder of a few important caveats:
- Not everything in life is subject to exponential improvements.
- Some trends, even those that appear to be consistent over time, will run into limits. For example, it’s clear that the rate of settling new land in the 1800s (a trend that was increasing over time) couldn’t continue indefinitely since land is finite. But it’s necessary to distinguish genuine hard limits (e.g. amount of land left to be settled) from the appearance of limits (e.g. manufacturing limits for computer processors).
- Some trends run into negative feedback loops. In the late 1890s, when all forms of personal and cargo transport depended on horses, there was a horse manure crisis. (Read Gotham: The History of New York City to 1898.) Had one plotted the trend over time, soon cities like New York were going to be buried under horse manure. Of course, that’s a negative feedback loop: if the horse manure kept growing, at a certain point people would have left the city. As it turns out, the automobile solved the problem and enabled cities to keep growing.
So please keep in mind that this is a technique that works for a subset of technology, and it’s always necessary to apply common sense. I’ve used it only for information technology predictions, but I’d be interested in hearing about other applications.
This is a repost of an article I originally wrote for Feld.com. If you enjoyed this post, please check out my novels Avogadro Corp: The Singularity Is Closer Than It Appears and A.I. Apocalypse, near-term science-fiction novels about realistic ways strong AI might emerge. They’ve been called “frighteningly plausible”, “tremendous”, and “thought-provoking”.
Subscribe to:
Posts (Atom)