One way to track scientific and technological progress is to compare outcomes to predictions that were made by futurists. So I pulled out my copy of Ray Kurzweil’s The Age of Spiritual Machines, written 20 years ago, which has predictions for 2009 and every decade thereafter. Are there predictions that he made for 2019 that came true sooner? Are there predictions that he made for 2009 that came true much later?
I will get into some specifics below, but some general points that occur to me from re-examining these predictions are the following.
1. Relative to his predictions, I can think of few “upsides” (something that appeared sooner or turned out better than predicted) and many “downsides.” Most of his predictions for 2009 have come to pass only in the past few years, and some are still remote. But to offer a more positive take, the fact that most of his milestones for 2009 have been hit as of 2017 is probably a better outcome than most other prognosticators would have been willing to bet on in 1999.
Roughly 50 percent of his predictions for 2019 now appear likely to be realized between 2025 and 2030, and the remaining 50 percent ought to be pushed back even farther.
2. I sense that a lot of progress that he expected has been held back by ergonomic issues. For example, language translation software may be effective, but people find ear buds and microphones to be uncomfortable. Similarly, augmented reality has been held back by the poor ergonomics of what goes over your eyes and ears. Collaboration across distance is not nearly as effortless as Kurzweil anticipated, even though software firms have put a lot of resources into “collaboration tools.”
Maybe more venture capital resources ought to be focused on finding breakthroughs in ergonomics.
3. Progress also has been slow in developing applications that react to the emotional state of human users. This is particularly important if computers are going to contribute outstanding value in education.
4. There is considerable cultural drag. Kurzweil predicted that by 2019 there would be parts of the road system dedicated exclusively to self-driving cars. One can argue that the technology is here to do that, but the culture is not ready to accept the idea.
I think that this cultural drag is becoming increasingly important. William Gibson’s saying that “The future is already here. It’s just not very evenly distributed” is even more apt than when he said it.
For 2009, Kurzweil predicted that we would have “at least a dozen computers on and around our bodies.” This seems wrong, but if you look at the functions that he expected these computers to perform, they are embedded in smart phones. Some of these capabilities matured after 2009, though, so I think Kurzweil was somewhat optimistic.
Also for 2009, he wrote,
The majority of text is created using continuous speech recognition (CSR) dictation software. . .
That prospect seems remote as I write this in 2017, so I have to count that as a really big miss. The software for speech recognition on my phone works quite well, but the ergonomics for, say, blogging, using speech recognition are not there.
Kurzweil saw distance learning becoming commonplace by 2009. I think that distance learning started to take off a few years after that, and even now it is not comfortable for most teachers and students.
Kurzweil foresaw translating telephone technology by 2009, but it only now seems to be emerging. He foresaw telemedicine in wide use in 2009, but in fact it is just starting to penetrate the health care industry.
About the only “upside” I can come up with is the Kindle reader, which appeared in 2007 and resembles something he predicted for 2009.
For 2019, Kurzweil predicted virtual reality glasses being in “routine” use. That does not look like it will happen.
For 2019, he thought that virtual teachers would have replaced real teachers. Maybe if you count YouTube videos as virtual teachers, but otherwise I think that this has not happened.
For 2019,
You can do virtually anything with anyone regardless of physical proximity. The technology to accomplish this is easy to use and ever present.
Nope.
“Phone” calls routinely include high-resolution three-dimensional images projected through the direct-eye displays and auditory lenses. . .users feel as if they are physically near the other person.
Relative to where we were when Kurzweil wrote this, FaceTime and Google Hangouts have taken us part of the way toward his vision, but we do not seem close to getting all the way there.
“The majority of text is created using continuous speech recognition (CSR) dictation software…”
“That prospect seems remote as I write this in 2017, so I have to count that as a really big miss. The software for speech recognition on my phone works quite well, but the ergonomics for, say, blogging, using speech recognition are not there.”
The simple problem here is that one can type faster than one can speak and one can read faster than one can hear. There is no reason to replace typing with speaking. In addition, in crowded places, typing is more private than speaking.
The one time I do a lot of textspeech is while driving and an idea occurs to me and I want to dictate it, when I use speech to text to send a short message to someone, or hear my email messages read to me (text to speech). Of course, I won’t be doing that much longer, as the car will be doing the driving, and I can read transcripts instead of listening to podcasts (so maybe there will be more speech to text after all)
I have worked on several speech recognition projects, all were for professionals that did not type (e.g. pathologists) and had to write many reports. The software increased their productivity by quite a bit.
“The simple problem here is that one can type faster than one can speak and one can read faster than one can hear.”
People say I speak slowly–and I do, especially to people who speak English as a second language–but I speak much, much faster than I type. (Especially with all the go-backs to correct mistakes when I type. :-))
Per a quick look on the Internet, the average speaking rate is 110-150 words per minute…virtually no one types that fast! (Another search indicates that professional typists at about 70 words per minute, and amateurs at 40 wpm.)
But have you evaluated how fast you can dictate error-free text to your phone? I end up dictating more slowly, and error-correction is abysmal.
“But have you evaluated how fast you can dictate error-free text to your phone? I end up dictating more slowly, and error-correction is abysmal.”
Well, I can’t type on my phone with my thumbs, like the kids these days can ;-), so I’m probably under 10 words per minute either way.
But I have a lot more confidence that voice recognition-to-text will improve dramatically, than I’ll start typing quickly with my thumbs. 🙂
P.S. When voice recognition-to-text works well for me, it’s like wonderful magic. But I agree that it’s tremendously frustrating with error-correction.
Seems like a lot of his predictions have somewhat arrived, but have significant headwinds when it comes to “Last Mile” implementation, which really slows down their ability to be transformative in the way he imagined.
A lot of this is probably due to cultural inertia, but a good bit more of it is probably just due to increasing technical difficulties that come with entirely closing the loop. Might be well be the case that we can quickly reach a point where his proposed technologies augment our existing solutions, but only very slowly replace them entirely.
I think I’ve learned far more online from blogs (such as your own)and even Twitter than live in any class room. Does that not count as virtual teaching? But I also suppose this is a pseudowin because you could make the same argument for reading paper books.
Still, I wonder about the aggregate quantities. How many total pages of education are read online everyday vs offline? How many educational lectures or videos (education doesn’t have to come from a lecturer!) are watched online vs in person? To me, these questions suggest he may be on point.
The text issue is well commented on. Frankly, writing and speech are rather different.
The issue of looking at projected images turns out to be much more challenging than one would imagine. Having just spent a relatively long time _really looking_ when I talk, I’ve come to believe that few people really do this, and that it is extraordinarily powerful. Further, that it is nowhere close for VR.
I agree with Will Minshew. I have learnt from askblog, from econlog, from Robin Hanson, from Scott Alexander, from Mencius Moldbug and countless other blogs much-much more than in university. In fact, over these years I more and more think that 99% of formal education is waste. Technology is here, but cultural drag (signaling in this case?) holds us back.
I have learned a lot for youtube videos, from how to cook x to physics to econ.
I wrote in my Amazon review of The Singularity Is Near that Kurzweil was not taking into account many sources of friction, political and technological. I had not considered “cultural”.
I am generally a “believer” in the Singularity story. Certainly his charts of exponential growth since the beginning of technology (or even before) are compelling. Kurzweil’s actual point about, say, virtual reality glasses, is that they are dependent on cheap processing power, and his prediction is unlikely to miss by more than a couple of years because growth of computer power is easy to predict. He may miss the 2019 date, but unlikely to miss 2022, with computer power another 3-4x cheaper.
The one thing that is still completely missing is computer “initiative”. The fender bender of the self-driving car in Vegas this week is symptomatic. The car drives well or better than humans almost all the time – but humans are still way better in edge cases, like when a truck is backing up onto your vehicle. I’d like to know if a computer driver would run a red light when directed to by a traffice cop. However, “initiative” will get here eventually.
Computers cannot ever have true initiative. They are completely logical. Every possible scenario has to be anticipated and then programmed by their developers. It is entirely possible for the programmers to anticipate a traffic cop directing a self driving car to do something at odds with its logic. But, then the self driving car has to have sensors capable of detecting that the object in the road is a cop and detecting his gestures. The software then can decode the gestures, and knowing his directions take precedence over other logic, act accordingly. That sensor capability doesn’t seem near at hand.
If self driving cars need dedicated roadways, I don’t think they are really “self driving”.
This isn’t true, and generally not how current self driving systems work. It’s a lot of ML which can learn from arbitrary inputs, and take novel action. Sometimes very clever, novel action, sometimes super stupid novel action. Hopefully less of the latter.
I’ve always thought Kurzweil was a bit silly because he didn’t account for diminishing returns and limits to exponential growth.
You know what really happens when you put a grain of rice on the first square of a chessboard, and two grains on the next square, and keep doubling for every square?
You run out of rice.
Kurzweil discusses limits to exponentially increasing computer power in The Singularity is Near and in talks. His argument is that “the limits aren’t very limiting” for at least the rest of this century.
“I’ve always thought Kurzweil was a bit silly because he didn’t account for diminishing returns and limits to exponential growth.”
Per Ray Kurzweil, circa 2025, a $1000 computer will be able to perform the same number of calculations as a human brain. Even if from 2025 onward, if the cost of a computer capable of doing the same number of calculations as a human brain dropped by say $20 per year (i.e. a linear change, rather than an exponential change) the world would be transformed in decades.
But there’s little evidence that the exponential increase in computing power per $1000 is likely to halt anytime soon.
http://www.theequitykicker.com/2010/08/17/kurzweil-predicts-personal-computers-with-the-power-of-the-human-brain-by-2025/
Actually, you run out of the board space of each square long before you run out of rice. Which dovetails nicely with your previous comment about limits to growth, and the original post’s mention of ergonomic, political, and cultural limits.
Get a bigger board and you’ll still run out of rice. 🙂
I wonder what the real world limit is?
You start filling your chess board, you empty your bags of rice. You go out to buy more, you empty all the rice from local warehouses. Have you run out of space yet? Or are you in a field? You need more rice, so you need people buying and shipping it in. Are you out of money? Are you blocking local roads so you can’t bring rice in fast enough? By the time you’re sinking enough money into it to buy significant percentages of all available rice, you’re driving the price up, which makes your growing piles more attractive to to thieves and hungry people. Each square you stack has to stack higher, with more care, can you glue the grains? Then they can be as high as the glue is strong and as the ground is supportive – you expand the chessboard lop-sidedly, you hire more people to stack, you try to coordinate with international rice sellers in multiple languages, you dominate local shipping routes, sentiment is turning against you, something is going to give out first, but what actually does?
“The car drives well or better than humans almost all the time – but humans are still way better in edge cases, like when a truck is backing up onto your vehicle.”
The simple solution there is to have all trucks operated by computers. Then they won’t back into one’s vehicle.
Human drivers kill 30,000+ people every year in the U.S., and close to 90 percent of the deaths are caused by human drunkenness, speeding, or inattention. The sooner all vehicles are driven by computers, the better.
What Kurzweil overlooked (or underestimated) IMHO: (1) ‘needs’ as opposed to capabilities; that is even if we could develop X, if there is not a great need for it, it will not develop much; (2) entrenched interests; disruptive technologies can be opposed, slowed down, blocked etc by parties who stand to lose.
But your point is excellent about the huge impact of ‘ergonomics’ or more generally ease of casual use (e.g. Apple products’ success). People are impatient, and if there are obstacles in the way they give up quickly. No, I do not want to figure it out. No, I do not want to learn how to use this. Either it works right away, or I give up.
This from 1965 by the inventor of PCM is still wide of the mark.
https://youtu.be/F8hqgwQPJ1E
One of the problems with telepresence seems to be that people who take the time, inconvenience, cost and risk of attending a meeting “personally in person” usually get their point through as against alternative views of people whose time is too valuable to spend it in traffic jams or airport waiting rooms. The irony is that people whose time is valuable are likely to have input to the meetings that is also valuable. However this isn’t an anomaly that technologists have managed to resolve so far.
And what about what people WANT to do? What about what biology and evolution demand they do?
I don’t care if I can dictate what I write, I prefer typing it, and need to *read* it rather than hear it in order to edit it. So even if speech-to-text is “perfect” it is not what I want. Likewise, for most emails, I need to *read* rather than hear them in order to use them. (Indeed, the pathologists noted above dictate reports to be written, rather than recording a report to be listened to. Significant I think.)
Likewise, “self driving” cars will face huge headwinds from people who do NOT want to pick a known location at a known address and go there in the most efficient way. There is a great deal more driving where people are making up their minds as they go, looking for something they don’t know the location of, driving past something to look at it, and so forth.
As for education – a very great deal of early education is the biologically limited process of socializing children and teaching them key behavoirs and technologies. How to talk inside, how to keep your feet to yourself, how to write the alphabet, and so forth. This process involves a lot of emotional and physical interaction (which is why parents are so key.) This most important part of education will never be remote.
College has a kind of demand for “reverse remoteness” – that is, a large part of the value of college is that the students leave their parent’s household, go somewhere “far enough” away, and go be embedded in this other experience. “Remoting” college back to the student’s parent’s dwelling is the exact opposite of what is required.
Don’t think of college as being about calculus or even critical thinking. A large part of college is about dealing with a landlord, joining or not joining various groups, learning to eat one’s own cooking, and so forth. None of that can be “remote”.
To the extent that virtual K-12 education has been tried, the results have been extremely poor.
https://fredrikdeboer.com/2017/06/12/study-of-the-week-of-course-virtual-k-12-schools-dont-work/
“Roughly 50 percent of his predictions for 2019 now appear likely to be realized between 2025 and 2030, and the remaining 50 percent ought to be pushed back even farther.”
This is a throw away sentence without specifying which predictions will be met by 2025 to 2030 and which “ought to be pushed back even further.”
I think Kurzweil discounts all the things that make Apple a great company. The fact that something works in a lab under narrow circumstances doesn’t mean it’s practical or useful for consumers. That takes a lot longer, and the design process often cannot even get started until after the technology appears.
Microsoft and Apple both launched tablet computers years before they were practical and useful. Then the iPad finally “had it all” and has wildly succeeded. But there was a decade there where tablet computers could exist in some limited form but lacked some ancillary technologies that finally make it compelling.
I feel like a lot of his predictions are “close enough” for predicting lab prototypes. He misses just how complicated and messy the real world is though and how the technologies have to cover all the edge cases too before they really take off. Self-driving cars are like that too. The Stanford self-driving car challenge was “won” ten years ago, but Waymo is only just going to market now.
Umm.. he revised all these up and down in “The singularity is near” https://en.wikipedia.org/wiki/The_Singularity_Is_Near