michael vassar vs. weird things, round one
As you might remember, Michael Vassar, the president of the Singularity Institute had a glance at some of my posts regarding theories of human transcendence through machines and artificial intelligence, and he’s not thrilled. In fact, he’s taken the time to send me a few rebuttals, the first of which primarily deals with the post that prompted his original e-mail, a look at how Ray Kurzweil and his partners are making money from their computer-inspired futurology. So settle in for the first episode of the Singularity debate…
Hi Greg, I’m happy to be here. I appreciate you giving me an opportunity to present my response to your recent articles on the Singularity. I don’t see a lot of technical disagreement between us in most of your articles on the Singularity. For instance, Visions of a Digital Doomsday seems totally reasonable. By contrast, we seem to disagree regarding what prominent Singularitarians think, say and do.
Well, I’m glad I’ve been able to write something reasonable for a change. As far as our disagreements go, the focus is not the Singularitarians. They’re actually the source of the ideas being scrutinized and it’s those ideas with which I see serious technical problems when it comes to real world implementation. Now, maybe I’m not quite right about what they mean when they say what they do, but I’m not here to put them on the couch. I’m an analyst and designer when it comes to the computer realm, not a philosopher or psychologist.
In Your Pricey Ticket to Immortality we seem to also agree on technical details. You say that to get where Kurzweil says we will be by 2045 will require tens if not hundreds of billions of dollars of research. Naturally. That’s why he says 2045 not 2011. Global research expenditures are many hundreds of billions of dollars annually, and between the largest few US tech companies alone tens of billions are spent on research.
Estimating very conservatively, if half of research spending over the next thirty six years goes into relevant fields, such as biotech, computer science, nanotechnology and electronics, and if just a meager ten percent of that is spent on relevant work within those fields, we’re talking about nearly a trillion dollars worth of relevant work.
Relevant research is all well and good, but it’s not necessarily going to help you. The applications of all these relevant fields are incredibly diverse and just because a lot of money is constantly being pumped into them, it doesn’t mean that in the next few decades the labs doing the research will produce something that will get the Singularitarians closer to their goals. Instead, these labs would need funding for decades of work on the very specific challenge of turning humans into immortal machines as per Kurzweil’s predictions in The Singularity Is Near. There are thousands of different projects in nanotech and computer science that have nothing to do with AI or making computers faster because those aren’t the priorities for the researchers.
[Your post] rather clearly and incorrectly indicates: a) that the [Singularity Summit] is put on by one of Ray Kurzweil’s organizations, b) that we are selling advice on how to be immortal and c) that the Summit costs a minimum of $399 and doesn’t include meals.
In truth, the Summit includes meals, though not dinner, and the discounts on our web page are clearly indicated to be cumulative by our statement “Each non-student referral receives a 20% discount. Summit is FREE with five referrals!”
In any event, all of our presentations, like the presentations from previous years, will be available on the web for free within a few months.
According to the Singularity Institute’s team page, Kurzweil is a member of the Institute’s board of directors. In fact, his photo and bio are right under yours. Ray is also a featured speaker, and he and Peter Thiel are prominent in all the ads and pages associated with the Singularity Summit. The post itself lists all the ways Ray sells his immortality advice and while I know you’re interested in protecting the Summit’s reputation, the event just happens to be one of the ways he keeps people interested in his theories. I don’t dwell on the event, I simply list it as a venue for promoting Kurzweil’s futurology business. I concede about the meals however since I just didn’t find any information about what’s being included in the package other than the speeches.
I also clearly noted that the fee “still amounts to $399 for attending both days of the event” after discounts, not that it was the minimum cost. And at the time the post was written, the note about getting a free pass for five non-student referrals was not posted. Since there was, and still is, a direct link to the registration page on the post I’m sure that at least one of the hundreds of readers who opened the page to see for themselves would have alerted me to the omission. The web is like that. You get fact checked whether you expect to or not.
But while we’re hotly debating what discount applies where and how, the two key questions are getting lost in this storm in a teacup. Are people going to the event so they can hear Ray’s presentations? Yes. And are they paying for it? Yes. We can talk about how much and discuss the cheapest way to attend the Summit of if the whole thing is profitable or not until the the Sun turns into a red giant but that doesn’t change these facts. It’s a small piece of the puzzle but just because you have a vested interest in this particular piece doesn’t mean we should miss the forest for the tress here.
If someone wants Ray’s advice on extending lifespan, they should probably read the most recent of the three books he has written on the subject, though they can get the highlights of Ray’s and other people’s life extension work from some of the other attendees.
And they would still be giving him money, which was the whole point of the post. You just can’t marginalize the person who’s the smiling face of the Summit to the outside world and say that he’s basically just one speaker and it’s not his organization running the show despite the fact that he’s on the board of directors of the group which organizes the event. If you would like to distance yourself from Ray so much and minimize the weight or popularity of his theories with the Singularitarian crowd, why even have him on the ads and on the board in the first place? Why not just advertise all the speakers and try to drive home how great of a bargain the event is?
[ illustration from the Artiphyciel series by Emanuel Lepas ]
“Relevant research is all well and good, but it’s not necessarily going to help you…these labs would need funding for decades of work on the very specific challenge of turning humans into immortal machines as per Kurzweil’s predictions in The Singularity Is Near. There are thousands of different projects in nanotech and computer science that have nothing to do with AI or making computers faster”
I don’t buy that. You seem to be treating scientific research as if it was something much more directed than it practically can be over long period. In general, the scientific community creates a collection of capabilities that support one another. When an application becomes directly approachable, it is always approachable for a few tens of billions of dollars or less. Are there any major new technologies that exist today that you think could have been developed a hundred years earlier of only someone in 1880 had thought to pour in a few tens of billions of dollars over a few decades to research directly focused on developing them? Very little work is directly aimed at AI, but everything we learn about the brain or about algorithmic efficiency brings us somewhat closer in a general sense. As for computer speed, well, I’ll just ask you strait out what you expect in terms of the speeds of future computers rather than arguing.
Also, where is Peter Diamandis in any of our adds? Seriously…
“In general, the scientific community creates a collection of capabilities that support one another. “
Right. So the first capability is one research project, the second is another, the third is yet one more, and bringing them together is a whole new project in and of itself. When you try to find a solution to a problem or achieve a very far reaching goal, you need to direct your research to those specific areas. Problems, scientific or otherwise, aren’t solved by just hoping that people from relevant fields will come up with theories and tools that when put together would create exactly what you need. You need to do the work yourself.
Yes all that relevant research can bring you closer to your goal but you will still need to devote decades and many billions of dollars taking all that relevant information, turning it into viable prototypes, and eventually, required devices. That was my point.
“As for computer speed, well, I’ll just ask you strait out what you expect in terms of the speeds of future computers.”
That depends. Are we talking silicon chips? Quantum computing? Diamond based microprocessors? The speed isn’t the issue. We already know how to make computers work faster if we need to. The only questions are scale and cost. The big issue is what we’ll do with all that speed. Today, the fastest supercomputers are basically assigned the work of very fancy calculators, solving very large and complex equations trillions of times per second.
“Also, where is Peter Diamandis in any of our adds?”
Whoops. I meant to write Peter Thiel. My apologies. The typo has been fixed.
In Tron, the computer deleted the protagonist’s physical body when it uploaded him into its memory, and restored his body when the story was over. Being unaware of any cosmic licensing/copy protection requirement, I consider that a bug-not-a-feature.
Let’s assume that hardware, software, & neurology in the next few decades all advance to the point where a few electrodes can be pasted to my grayed temples so that a software entity results which perceives itself as me looking out through a webcam instead of eyeballs. If either the system operator or the AI itself have any editing capacity, a lot of housekeeping will need to ensue – why retain all the subroutines having to do with legs/stomach/genitals etc when those peripherals are not installed?
Pretty shortly we end up running a Me 2.0 resembling the original only in, at most, ideology/opinions and a set of verbal idiosyncrasies. Once that entity takes advantage of its cybernetic ground of being to finally learn all the math, languages, science, history, etc which my carnal incarnation was too distracted &/or cerebrally limited to pick up first time ’round, Me 3.0 will lose most of those features which I or anybody else would identify as myself.
All that the brainwave reading accomplished, then, is to furnish a slightly different template for the billionth neural network hosted on the singularity.com servers. Perhaps that entity would be kind enough to call up my earlobephone as I slowly hobble out of the mighty Kurzweil Tower and answer some of the burning questions that have filled my last few decades, but what basis would either it or I have to think that “I” had been immortalized?
And research projects take, in general, 2-10 years. You can stack a lot of research projects on top of one another in 50 years.
Ray Kurzweil claims that people intuitively believe change to be linear while it’s actually exponential. I actually think that’s mistaken. In reality change is pretty linear, limited in speed, as you say largely by research cycles (though of course simulated people in simulated worlds should be able to run through their research cycles at their subjective speeds. Kurzweil is right however that people underestimate change. In general they imagine a single jump, from now to the near future, and then another from then to the far future, with the near future being a roughly linear extrapolation of the last 5-15 years and the far future being a bunch of cliches taken from science fiction. They absolutely do not take the process that creates their ideas about the near future and iterate it, e.g. say, “OK, so what would ten successive increments of change comparable to that over the last 5-15 years look like”. If they did, they would find it pretty hard to rig the details in such a manner as to not describe a world where a single well funded university lab, not to mention a major government or corporate project, could create a Singularity of the sort I’m discussing, or for that matter destroy the world by accident.
We agree that we don’t know how to simulate a brain efficiently, but ANY simulation is nothing more than running some very large and complex equations trillions of times per second. It seems to me that we covered this by phone, and that the only way to argue that we will not have simulated humans by the end of the century is to argue that we will never have a) computers even ten thousand times more powerful than those that exist today, or b) equations that can realistically describe the input to output relationships that exist in a neuron, relationships that thermodynamics and the energy consumption of a neuron guarantee can in principle be simulated with under 100,000 flops, that are even a few percent computationally efficient.
“[People] absolutely do not take the process that creates their ideas about the near future and iterate it, e.g. say, ‘OK, so what would ten successive increments of change comparable to that over the last 5-15 years look like?’”
The problem with trying to predict the future is that none of us have the powers of precognition. We can extrapolate however we want, but all it would take to upset everything we’re eagerly predicting is a war or a terrorist attack or another major economic downturn. And this is not to mention the role of politics, egos and government bureaucrats who give out major grants.
Right now, your argument seems to be that future technologies build on each other and we can’t underestimate how much researchers can achieve. Fine, that’s a fair point. But at the same time, we can’t ignore that the process of getting to profound new technologies is a messy one, influenced by many forces outside the lab, and that relevant research doesn’t necessarily translate into applicable research.
“We agree that we don’t know how to simulate a brain efficiently…”
Let’s save that for the next post where we can devote enough time to the science involved.
A nuclear war would definitely set things back, but no-one ever says we shouldn’t worry about global warming or the deficit because there might be a nuclear war? You are stuck trying to predict the future, and of course if you are serious you include lots of possibilities. Equally obviously, every possible outcome doesn’t get discussed.
Short of nuclear war, many things, from plagues to supervolcanos to serious regime collapse in the US or possibly China might significantly retard scientific and technical progress, but given that the second world war doesn’t seem to have done so, nor the great depression very noticibly, a terrorist attack seems awfully unlikely to except through causing US regime collapse.
Lots of things are imperfect with science funding today, but are you making a specific prediction that that situation importantly likely to get much worse? If not, why bring it up? Science is progressing today and will, in most people’s nominal beliefs, probably continue to for a substantial time. If I was going to use the laplace’s rule of succession, the default update, I’d say that we have a solid 8 successive half centuries of blatant scientific progress, so the odds of another half century are 90%. I can then update from that based on other elements of my models and end up with a much lower estimate, more like 60%, but if it was 10% that would still make the expectation of scientific progress a very big deal.