Pages

January 18, 2010

Ray Kurzweil Responds to the Issue of Accuracy of His Predictions

Ray Kurzweil responds to Michael Anissimov and I got a copy of the letter as well. The complete letter is below. After the letter is a discussion of Ray's technology predictions and the work needed for accurate, precise and unambiguous predictions.

Here is Michael Anissimov's original article

Here is a link to Chapter 9 of the "Age of Spiritual Machines", which has the exact original wording of most of Ray's predictions for 2009. (H/T again to Michael Anissimov at Accelerating Future)

The Text of Ray Kurzweil's Letter
January 17, 2010

Dear Michael,

I want to respond to your Blog post “Reviewing Kurzweil Predictions from 1999 for 2009.”

This starts out “Michael Anissimov notes that Ray Kurzweil had several predictions from 1999 for 2009 and those predictions are in general wrong.”

You also write “Ray Kurzweil’s Failed 2009 Predictions. In May 2008, a poster on ImmInst (the life extension grassroots organization I co-founded in 2002) pointed out that it looked like Kurzweil’s 1999 predictions for the year 2009 would fail. Now that 2009 is over, we can see that he was mostly correct.”

Your review is biased, incorrect, and misleading in many different ways.

First of all, I did not make “several predictions” for 2009. I made 108 predictions in The Age of Spiritual Machines (TASM), which, incidentally, I wrote in 1996 to 1997. It takes a year to publish, so the book came out at the end of 1998. It is very misleading to take 7 predictions out of 108 and present that as all of my predictions for 2009.





I am in the process of writing a prediction-by-prediction analysis of these, which will be available soon and I will send it to you. But to summarize, of these 108 predictions, 89 were entirely correct by the end of 2009.

An additional 13 were what I would call “essentially correct” (for a total of 102 out of 108). You will note that the specificity of my predictions in TASM was by decades. There were predictions for 2009, 2019, 2029, and 2099. The 2009 predictions were providing a vision of what the world would be like around the end of the first decade of the new millennium. My critics were not saying “Kurzweil’s predictions for 2009 are ridiculous, they will not come true until 2010 or 2011.” Rather, they were saying that my predictions were off by decades or centuries or would never happen. So if predictions made around 1996 for 2009 come true a year or a couple of years after 2009, given that the specificity was by decade, and the critics were saying that they were wrong by decades or centuries, then I would consider them to constitute an essentially accurate vision of what the world would be like around now.

My critics are very quick to jump on and exaggerate the slightest issue with my predictions. For example, earlier this year, one critic wrote that my prediction (made in 1996) that by 2009 there would exist a supercomputer that would be capable of performing 20 petaflops (quadrillion operations per second)” was “not just a little bit wrong, but wildly, laughably wrong.” I wrote back that IBM’s 20 petaflop Sequoia supercomputer was already under construction and that IBM has announced that it will be operational in 2012. Since that time, another 20 petaflop supercomputer has been announced that will be operational next year, in 2011. Is it fair or reasonable to call this prediction “wildly, laughably wrong?”

I make this very point in my movie The Singularity is Near, A True Story about Future. One of my key (and consistent) predictions is that a computer will pass the Turing test by 2029. The first long-term prediction on the Long Now website (www.longnow.org) is a bet that I have with Mitch Kapor regarding this prediction. Mitch and I put up $20,000, and this amount plus interest will go to the foundation of the winner’s choice. I will win if a computer passes the Turing test by 2029 (and we have elaborate rules that we negotiated) and Mitch will win if that does not happen. In the movie, I create an AI-based avatar named Ramona and she fails the test in 2029 and Mitch wins the bet. However, she goes on to pass the test in 2033. If that is indeed what happens in the future, whose vision of the future can we say was correct?

From a strictly literal point of view and in terms of the rules of the bet, Kapor will have won the wager. But Kapor’s critique is not that “Kurzweil’s prediction of a computer passing the Turing test in 2029 is ridiculous, it won’t happen until 2033.” Rather he is saying I am off by centuries if it ever happens at all. My point is that if a computer passes the Turing test by 2033 rather than 2029 my vision of the future would be “essentially correct.” And so it is with the 13 predictions out of 108 that I made in TASM that are likely to come true in the next year or couple of years. By my calculation, 102 out of 108 predictions are either precisely correct or essentially correct.

Another 3 are partially correct, 2 look like they are about 10 years off, and 1, which was tongue in cheek anyway, was just wrong.

So for starters, your list of 7 predictions is misleading and is the result of severe selection bias. Moreover, most of these are not actually wrong. You have also changed the wording in ways that change the meaning of the predictions, or have just misinterpreted either the prediction or the current reality.

Take, for example, the first one you cite. The correct prediction was “Personal computers are available in a wide range of sizes and shapes, and are commonly embedded in clothing and jewelry…” When I wrote this prediction, portable computers were large heavy devices carried under your arm. Today they are indeed embedded in shirt pockets, jacket pockets, and hung from belt loops. Colorful iPod nano models are worn on blouses as jewelry pins or on a sleeve while running, health monitors are woven into undergarments, there are now computers in hearing aids, and there are many other examples. The prediction does not say that all computers would be small devices, just that this would be "common," which indeed is the case.. And “computers” should not be restricted to the current category we happen to call “personal computers.” All of these devices – iPods, smart phones, etc. are in fact sophisticated “computers.” By a reasonable interpretation of the prediction and the current reality, it is correct, not “false.”

There are indeed “computer displays that project images directly onto the eyes.” The prediction did not say that all displays would be this way or that it would be the majority, or even common.

You cite the prediction that “three-dimensional chips are commonly used” as false. But it is not false. Many if not most semiconductors fabricated today are in fact 3D chips, using vertical stacking technology. It is obviously only the beginning of a broad trend, but it is the case that three-dimensional chips are commonly used today.

“Translating Telephone technology” was indeed available only in prototype form earlier in 2009, but now is a popular iPhone app and the technology is available on Symbian phones and on Google's popular new Nexus One, using Google's voice translation server. My prediction was that it would be “commonly used,” not that it would be ubiquitous. I suppose we could argue how “common” its use is, but it is already a popular app. Having been introduced late in 2009, it is likely to become quite popular on many phones worldwide in 2010.
“Warfare is dominated by unmanned intelligent airborne devices” is certainly true in Afghanistan. As Wired recently noted, "The unmanned air war ... has escalated under McChrystal’s watch....” Also there are munitions that are about the size of birds that can be released from larger aircraft and that have their own intelligent navigation.
So even of this highly selective list, your interpretation of the predictions is rigid and idiosyncratic. You have a certain vision of how these types of developments will or should manifest themselves, but under a reasonable interpretation, most of your selected predictions are in fact not false.

The status of these predictions changes very quickly. In November 2009, the idea of large-vocabulary, continuous, speaker-independent speech recognition on a cell phone was still off in the future. Just one month later, this became one of the most popular free apps for the iPhone (Dragon Dictation from Nuance, which used to be Kurzweil Computer Products, my first major company) as well as the popular Google Search on iPhones and in Google Droid and Nexus One phones.

Two or three years from now is a very long way off, and the world will again be quite different, so for the handful of my 108 predictions for 2009 that are not literally true now, most will likely become true over that time.

So I agree with you that there should be accountability for predictions, but such reviews need to be free of bias, fair, and not subject to selection bias and myopic interpretations of both the words used and the current reality.

In this essay I am working on, I will also review my predictions written in the mid 1980s in The Age of Intelligent Machines, which were also very accurate.

I am not saying that there are no misses, but it I believe it is fair to say that the vision of the future that I have painted in the past for the current world is quite accurate, especially compared to the critics who at the time said that these predictions were off by decades or centuries.

Best,

Ray Kurzweil

My Analysis of Ray Kurzweil's predictions
I had reviewed the same set of Ray's predictions 7 out of 108 total predictions. I had previously updated the article with the fact that Ray had made more predictions than I covered and there was a fairly comprehensive list of them at wikipedia.

Ray's letter does also get into the aspect of what is meant by the qualifying terms and definitions of certain key words, which I had already noted. A clear positive or negative prediction would specify a quantified data range and more specificity in what the likely range of the qualifying word is. Having tried to apply that amount of rigor myself means that a prediction ends up taking 1-2 pages to describe and takes five to one hundred times as much research to make in order to get to explicitly define a suitable confidence range.

Technology That could Enable Electronics as Clothing

Stretchable, Porous, and Conductive Energy Textiles are in the labs now and making a wearable high resolution display will be possible as previously noted with more advanced OLEDs. However, this does not change the fact that there is less purpose to have a wearable high res display to show things to other people. Wearable electronics are coming and will be used and picoprojectors (so that the person carrying and others can all see it) are commercial.

There will be increased power for personal electronics from energy harvesting





This is an example of the level of detail that you need for really useful predictions. I would have to keep going for another couple of pages to specifically describe what is possible, what is coming and how it is reasonable that it be used and combined. A substantial part of it would be laying out the basic physic parameters and broad usability and functionality categories.

I made about 156 predictions in 2006 and had a series of articles explaining and later reviewing my predictions.

I did not fully qualify all of my predictions in 2006, but this and my long bet which is more qualified have shown me the effort needed to make predictions that can be unambiguously judged.









blog comments powered by Disqus