Singularity When?

Mike Deering, Executive Director of Singularity Awareness, joins us again to discuss his ideas about when the Singularity might happen. Whether you agree with his conclusions or not, he makes some good arguments about how the nature of exponential change is likely to be sudden and disruptive, rather than gradual.

~~~

With the current state of medical technology at your local hospital, factory automation where you work, entry level desktop PC technology that you can buy from Dell, things don't seen to be moving very fast. And we all know that we have a long way to go. Even if you look at the leading edge in: robotic technology, gene sequencing of viruses in a few hours, mass production of carbon nanotubes, stopping light particles in the lab, paper thin display monitors, sub-wavelength microscopy, MEMS, microfluidics, qubits, self-assembly, and nano-particles, you would see that we have just barely begun to get a handle on the enabling technologies for molecular super-computers, medical nanobots, and desktop nano-factories. The most frustrating thing about the Singularity is that it is an exponential process rather than a linear process. This means that we are not going to gradually and predictably move from where we are to God-like technologies. It's going to be a very sudden change.

The way technology has advanced in our experience is by degrees. We make electronic circuits a little smaller all the time. We go from vinyl albums to cassette tapes to CD's to DVD's to memory cards to story our music. Each advance has advantages and improved performance and new functionalities, but the changes are small and gradual. The Singularity is different. The Singularity is a technological phase change. When you can manufacture large products by placing every atom where you want it, rather than taking a large block of material with randomly positioned atoms and cutting away everything you don't want, or melting the material and pouring it into a mold to solidify into an object with randomly positioned atoms, this changes everything about the product. By controlling the position of every atom you can control the physical characteristics of the material. You can make materials fifty times stronger than steel, virtually indestructible. You can make products perfect down to the atomic scale. You can build new functionalities into the materials at the molecular level, paints that spread themselves on the wall, change colors on command, and never get dirty. When you have software with general reasoning capabilities, that changes everything. Suddenly all of your computers are doing exactly what you want them to, just by asking them, cars driving themselves, robots taking your job, finding exactly what you are looking for on the internet the first time, personal digital assistants that manage your money for you.

Either one of these two technological developments, molecular manufacturing, or artificial general intelligence, would almost instantly change every aspect of the way you live your life, but guess what? Both of these capabilities are going to occur simultaneously, along with gaining the complete knowledge of all biological processes at the molecular level. Nanotechnology will first be used to build computers millions of times more powerful than are available today at almost zero cost. This will result in the almost immediate development of super-human artificial general intelligence (SAGI). Just as there are many paths to nanotech, there are many paths to SAGI and all of them are greatly accelerated by more powerful computers. The combination of nanotech and SAGI will rapidly develop a complete knowledge base of molecular genetic and proteomic biological functions. At this point, whoever controls the technology can do whatever they want.

A.I. and nanotech are inextricably linked. They are advancing in lock-step because the one cannot advance without the other. This interdependence means that they will maintain the same level of progress toward the final goals of molecular machinery (MM) and AGI. It is understandable that you would think that AGI is at least a generation away if you are not following closely the developments in A.I. research. Just as you would think that MM is at least decades away if you only knew what you read on CNN. Here is some stuff you should know about AGI research:

---Biology is an existence proof of molecular machinery and computational intelligence.

---AGI's like nanobots will not be reverse engineered copies of their biological counterparts, but rather original engineered designs using some of the same concepts used by biology and some wholly new concepts.

---The design stage of development of AGI is at the same or more advanced level as the design of the assembler.

---Both the assembler and the AGI are very complex design challenges that some people believe are beyond our intellectual capability. They are wrong in both cases.

---Experts in both fields, Richard Smalley and Marvin Minsky, claim that they will not be developed for a long time if ever. They are both wrong.

AGI's do not have to be people. This is a very important point. There are many different cognitive architectures that can support intelligence, not all of them are conscious, self motivated, self serving entities. It is very possible to design a piece of software with general intelligence and problem solving ability without ego. This would significantly reduce the dangers of AGI. Even though this would remove the danger of the AGI taking over the world for its own purposes, it would not remove the danger of the person controlling the AGI taking over the world, or making a mistake in the use of the AGI if the AGI was significantly more intelligent than the user. If the AGI is significantly more intelligent than the user, the user might not understand all of the implications of the results produced by the AGI.

The combination of general purpose human-like reasoning ability and computer speed, complex accurate serial computation, data storage, and reprogramability will make human level AGI's automatically super-human level intelligences.

The leading contenders in the AGI race that I am aware of are Ben Goertzel's Novamente project and James Andrew Rogers secret AGI project. Both claim to be no more than twelve months away from a human-like reasoning working prototype.

What is the world going to look like a year before the Singularity? A month before? A day before? Well, pretty much like it does today. We are gradually moving toward having these capabilities, but before we have them, we can't do any of the amazing things we will be able to do after we have them. Technology is advancing faster and faster because of feedback, the better tools we have the faster we can build even better tools. This feedback function results in an exponential curve of technological advancement. The nature of exponential advancement is that almost all of the change occurs very near the end of the curve.

The Singularity is going to be a complete surprise to the vast majority of persons on this planet, even the vast majority of highly educated people. It's not going to make itself obvious until it is upon us. Anyone who makes "realistic" predictions of the timing of the Singularity will be laughed at, because conventional wisdom and the consensus of expert opinion will always be based on linear models of gradual change, whereas the Singularity is an exponential process of advancement with a phase change at the end. You don't get respect by telling people the truth about the Singularity. I realize that my prediction is outlandish and ridiculous by generally accepted standards. Nevertheless, I think that getting the truth out is more important than building my reputation. It is critical that people hear that the Singularity is coming much sooner than they think. We are one vital breakthrough from the Singularity.

I think this whole Singularity phenomenon is driven by miniaturization of electro-mechanical systems. All the Singularity technologies are based on it. Of all of Kurzweil's graphs, this one is the scariest :
http://www.kurzweilai.net/articles/images/chart18.jpg

Admittedly, there are not many data points, but it is clear that miniaturization is leading the charge to the Singularity. The most accessible way to see this is by watching the market with the greatest economic return on miniaturization, computers. And in the computer market, the slice with the most fiscal pressure (largest dollar total, largest number of units) is the entry level desktop computer. You may have noticed prices dropping while power increases.

Doubling time - the time it takes for the power of a computer to double for the same dollar.

example: suppose you purchased a computer on a certain date, and at a later date, you found another computer with double the power for the same price, or you found another computer with the same power at half the price, or some mathematical computation resulting in double the power per dollar.

The reason we would be interested in this doubling time value is that available computer power is an important factor in the timing of the Singularity.

For instance, Eliezer Yudkowsky writes in CFAI: 4.1.2.

"The intelligence required to create AI. Decreases with increasing computing power." Eliezer also writes in CFAI: 4.1.2. "The total processing power available to an average research project will increase faster than chip clock speeds (i.e., maximum parallel speeds increase faster than maximum serial speeds). The total networked processing power on the planet will increase even faster than that; a doubling time of nine months is probably an underestimate."

Ben Goertzel writes on 1/4/2003 [AGI],

"We are limited tremendously by CPU speed and RAM capacity. Either greater CPU speed or greater RAM capacity would be valuable, but the biggest boost would be both together. We could utilize essentially any amount of CPU speed or RAM capacity. No limit in sight. Having a CPU with (for example) 10x greater speed would have a HUGE positive impact on some of the work we're doing. If we had vastly better CPU's and vastly more RAM, the amount of time to get to a complete working implementation of a Novamente system might be reduced to 2/3 what it is right now."

I accept these experts opinions that the availability of more powerful computers would significantly push up the date of the Singularity.

So what is the doubling time? How do we calculate it? On the http://www.bjklein.com/sing/default.htm website it says, "Computer power is doubling at a slowly accelerating rate: every 18 months currently." This seems to be the general consensus based on a little google statistics:

google search -----------------------------------number of sites

• singularity doubling "24 months" --------------------60
• singularity doubling "20 months" --------------------4
• singularity doubling "18 months" --------------------163 ****
• singularity doubling "12 months" --------------------78
• singularity doubling "11 months" --------------------8
• singularity doubling "10 months" --------------------8
• singularity doubling "9 months" --------------------28
• singularity doubling "6 months" --------------------49
• singularity doubling "3 months" --------------------25
• singularity doubling "1 month" --------------------10

For purposes of assessing the effects on the Singularity, we need to use the computer with the fastest doubling time out there. AGI will come out of the computer market with the best power to dollar ratio. This market is the entry level desktop. The April 2003 entry level desktop sells for $500 and has a 2 gig Celeron processor with 256 meg of ram and a 80 gig hard drive and other standard peripherals. As you can see from this graph the price of desktop computers has been dropping since 1990 and continues to drop:

While at the same time the power has been steadily increasing:

Based on available data, how are we to calculate the doubling time extrapolation into the future?

On 1/6/2003 [AGI] Stephen Reed writes:

"Progressing from -50 db HEC to 0 db HEC in 22 years is equivalent to Moore's Law doubling every 16 months. [ 2^16.61 = 100025, 22/16.61*12 = 15.9 ]"

A careful examination of this formula shows that Stephen is merely averaging the doubling time over the past 22 years and applying that constant to the next 22 to arrive at his crossover date of 2021. A constant extrapolation of an average doubling time is not the correct method to project an exponentially changing value. Unfortunately I haven't been able to get good historical data on the entry level computer market. I would welcome any assistance. This is my current extrapolation:

DATE ------ DOUBLING TIME ------ DROPPING RATE

• 1900 ----------48 months
• 1915 ----------42 months --------------------6/180 (6 months in 180 months)
• 1930 ----------36 months --------------------6/180
• 1945 ----------30 months --------------------6/180
• 1960 ----------24 months --------------------6/180
• 1975 ----------18 months --------------------6/180
• 1980 ----------17 months --------------------1/60
• 1990 ----------15 months --------------------2/120
• 1999.5 -------12 months --------------------3/114
• 2001 ----------11 months --------------------1/18
• 2002.3 -------10 months --------------------1/15
• 2003.3 --------9 months ---------------------1/12
• 2004 -----------8 months ---------------------1/10
• 2004.7 --------4 months ---------------------1/8
• 2004.6 --------3 months ---------------------1/6
• 2004.9 --------2 months ---------------------1/5
• 2005.2 --------1 month -----------------------1/3
• 2005.3 ------<1 month -----------------------1/1 --------------------Singularity!

Actually, April 28, 2005 (plus zero, minus 11 days)

Mike can be reached at deering9 at mchsi dot com

Posted by paul at April 26, 2004 01:47 AM | TrackBack