Some people think much faster computers are required for Artificial Intelligence, as well as new ideas. My own opinion is that the computers of 1974 were fast enough if only we knew how to program them.
- John McCarthy
So begins the book After the Software Wars, by Keith Curtis. At least, that’s how he says on a blog he started his book; I haven’t read it (but i’m sure it’s a fine book). But with all due respect to John McCarthy – which is a lot – what the …?
How could he possibly know if computers of 1974 were fast enough for AGI – i’ll go ahead and assume that’s what he means – if he doesn’t (even now, presumably) know what code they would even be running? And, further, if we don’t know how to program them, wouldn’t some new ideas help a bit? I’m glad i ran across this quote. Not for an opportunity to diss an AI grand master — that’s not my intention at all. Rather, the concept of field physics has been rumbling around in my head for a while, and this quote made it bubble to the surface. So, thanks John.
Like probably every other programmer interested in AI, i wrote my fair share of neuron simulations. Not “neurode” simulations, but actual high level sims of action potentials and neurotransmitter release and input summing (over time) and blah blah blah… And as we all know the Blue Brain Project has taken this to an unprecedented and fascinating level, right down to axon ion channels and all of the other stuff the rest of us thought (and may still suspect) irrelevant. And god bless them for it, because now we know that simulating even just roughly 10,000 neurons in a cortical column takes at least 4 inter-operating super computers.
But, likely, John thinks that biological simulation is unnecessary. We merely need to determine the “cortical algorithm”, lisp it into a turbo-charged ENIAC, and then settle into our subsequent lives of leisure. (Sorry John, couldn’t help that. And hey, if you’re actually reading this, call me!)
But I suspect there is more to it than that; something of a middle road. Even within a single cell – neuron or otherwise – there are uncountable chemical interactions happening all at once. These interactions can be summed up statistically, but although this might provide some high level information on what is going on it could never recreate it. The problem is the field interactions. I wrote previously about some research i did on Lotka-Volterra models, and how adding dimensionality to the problem completely changed how it worked. Often statistics is a way of eliminating dimensionality and simplifying a problem, but as some guy named Einstein said, make things as simple as possible, but no simpler. And statistics makes things too simple. Now, go beyond a single cell into a brain of billions, and tell me that computers of Captain and Tennille fit the bill.
Here’s the thing: massive computing parallelism is a powerful thing. And biology tends to like to reproduce (seemingly) simple working things on a massive scale. Bottom line: brains use massive computing parallelism… because they can. So, maybe the cortical algorithm is actually quite simple (which i doubt, even if it exists, which i doubt), and maybe enough ENIACs could be wired together to achieve the necessary computing power (which i doubt way more), but 1974 is reaching a bit too far back. Personally, i wonder if 2018 may be reaching too far back. I hope not.
Hmm, but i didn’t really explain the field physics thing… What wigs me out is that some dust particle on the other side of the galaxy is, as we speak, affecting the gravity of our planet. I’m not a physicist, but my understanding is that in our universe every mass affects every other mass gravitationally simultaneously. (If i’m totally out to lunch on this, please let me know so that i can quickly delete this post.) We currently don’t have any means of computation that can replicate this, so we’re reduced to having to run huge time-consuming double “for” loops to calculate the affect that every particle has on every other particle in our virtual universes. Yes, i know there are clever ways to optimize this, but the fact is the real universe gets it for free. There are certainly ways in which brains are taking advantage of the field physics free lunch. Just considering classic economics: if your brain isn’t doing it, you’ll be at a fatal disadvantage to a brain that is. Proof, i have not. But, sorry John, i’ll be putting my money on a number higher than 1974.
Hi;
There is no canonical FOSS codebase containing AGI that has hundreds of people working together on it. This is a social problem, not a technical one.
That could be a whole new blog post…
I understand where you’re coming from, but i mostly must disagree. There are certainly social issues involved in making AGI into a collaborative work, but there are far more technical problems. For example, where to begin? If i knew how to answer this question i would have created a FOSS project a long time ago. (Look me up… I’ve been doing it for years.) In fact, that’s in a way what GOID is.
Right from project conception there’s a problem with input, being, what is the input? Text? Images? Data streams? Brains use huge numbers of concurrent pulse streams, so i might propose we start there, but as far as i can tell not many people would agree. Without being able to decide on this simple architectural point, how could we possibly decide on the thousands that follow on?
I agree that 1974 computers will not be sufficient for real time AGI.
Your point about gravity is essentially correct, as far it effects what you are discussing. But it would be more accurate to say that: every mass effects every other mass – in it’s future light cone – gravitationally *at the speed of light*. (Ie, not simultaneously, gravity propagates at c, and masses can be too far appart to interact.)
I agree with you 100% that’s it’s amazing that all masses 10 billion light years away (and ago!) effect the mass of my chair.
But as you well know, brains operate perfectly well in microgravity, and it is really the EM fields that are of interest. And these fields can be treated as simultaneously effecting each other on the size scale (and speed scale) of the brain.
The vast majority of the EM fields in the brain do not meaningfully effect anything even a fraction of a synapse length away. (Think of the field around a typical protein, lipid, water, etc.) The only time field effects extend meaningfully beyond that, is of course when neurons fire. But even this doesn’t extend much beyond a single neuron except when many neurons are spatially aligned, and temporally correlated. In short, brain waves. We know EM fields can have long distance effects on neural functioning, since we’ve used them to meddle with people’s heads. (There are short range local field effects that we might have to worry about. For example, E fields leaking out the nodes of Ranvier in the myelin sheath and effecting nearby unsheathed dendrites. But those kinds of things aren’t the sort of everything-effecting-everything effects you are discussing.)
If our intent was to simulate a brain, then handling the long range EM effects wouldn’t increase the overall computational demands by even a factor of 10; perhaps less.
However, it seems like your real point was actually less about fields, and more about parallelism – and how nature gets it for free. Yes? That point I easily grant.
So the bad news is that nature gets parallelism for free. But the good news is that the brain is modular, so it’s not doing n^2 operations per unit time (i.ie, we don’t need doubly nested FOR’s over all the components). And the other good news is that we wont’ (as you say) have to model neurons in full electrochemical detail.
@Matthew:
The AGI system can have multiple kinds of inputs. The Linux kernel gets input from multiple places. Supporting one kind of input never precludes another.
Creating a FOSS project is not enough, you need to get other people working on it. Take a look at computer vision. There is no canonical codebase for it: http://www.cs.cmu.edu/~cil/v-source.html. Everyone knows the name Wikipedia and Linux, but no one knows the name of the de facto computer vision or AGI codebase.
Once you have this social problem solved, then people can solve the technical ones. There are lots of to be solved, and there needs to be a place for the software knowledge to accumulate. It is that simple.
As for whether we could do AGI in 1974, we don’t know yet because we don’t have the software. I think it wouldn’t be any slower, but it would be dumber. What is the smallest AGI you can build? How much data would it have?
I discuss computer vision hardware requirements in my book.
Hey Keith,
In the same way that there are a ton of computer vision projects, there are also a ton of AI projects, some of which claim to be AGI. They don’t coalesce into a common effort because there is so little technical similarity in the approaches.
You try to compare it to a Linux kernel, but that’s misleading. I tried talking to my Linux box, but it didn’t respond. I even tapped on the box in morse code, but it still pretended not to hear me. What gives? In fact, the Linux kernel only handles an extremely limited set of input. And it only handles input in extremely limited ways, constrained by its von Neumann architecture. If it were as robust as you suggest, we would have had AGI a long time ago.
The bottom line is that very smart people have been trying to figure out AGI for over 60 years, and yet the opening chapter in AI texts for university courses is still “Search”. We only know a bit more about building AGI than we do about building a time machine. There is certainly no shortage of people who want to figure this out, so i disagree that it’s a social problem. We just don’t know where to start.
If you still disagree, i suggest you address the social problem that you detect. By all means, count me in.
Hey Forrest,
Thanks for setting me straight. I was, yes, only using gravity as an example, also silently including other field forces like EM – and you called me out on that too. Parallelism is more what i’m after, as you said.
It’s of course true that not all brain cells talk to every other cell. But then, there are purkinje cells with dendrites touching up to 10K other cells – that’s a lot of parallelism. And the interconnectivity of the cerebellum is just awesome.
But, even inside a cell there are unknown numbers of second messenger systems that regulate the behaviour of the cell over multiple time periods. It’s possible that these systems could be reduced to statistical calculations of the x(t) = f(x(t-1)) type representing the entire cell. I seriously hope so, although i’ve read some stuff suggesting that it might not be quite that simple. Even still, this means that the coding style can’t be event-based (i.e. only execute when there’s an event to process), but rather must be continuous to simulate as analog an environment as possible. We might be able to avoid huge double nested for loops, but we might have to substitute them with a single massive time-slice loop.
Yes, agreed again that an event-based simulations isn’t going to cut it; and with your previous comment that doing every single ion channel is overkill. OTOH, Markram is leaving out things like adult neurogenesis, which might matter for learning and memory.
Would be great if you could point me to any strong arguments for exactly what level of modeling detail is required.
m@:
The reason we have lots of codebases for AGI is the same reason Linux has lots of RSS readers. There are tons of technical similarities between AGI codebases. We have separate AGI codebases because the humans are choosing not to work together.
Software is infinitely malleable. 99% of the time, software can support both features. And in the 1% where two people have different and incompatible ideas for how something can be done, you can still support both methods and create a configuration parameter telling it which one to use.
I was just making the point that an AGI system may need to support multiple kinds of inputs like text, images, data streams — and that Linux already handles these things today. The Linux kernel, and the distro, handle these things.
The worry about how an AI system will handle multiple kinds of input is such a basic high-level discussion. There are thousands of more detailed and complicated issues to be worked through. The questions you ask, and many harder ones, will need to be answered. Software never writes itself.
The Von Neumann architecture can build machines to run any software: http://en.wikipedia.org/wiki/Turing_completeness Building a Turing complete machine is a very simple requirement. Given that baseline, you can build as much software complexity on top as you want, including a AI system that can input audio, video, and text, and output text. The multiple data streams problem is well understood.
I realize that very smart people have been trying to figure out AGI for 60 years. The thing I keep saying is that there is no place for the knowledge to accumulate.
As for the hardware, remember that we have SSE, and GPUs at our disposal. We can parallelize an algorithm if we want to. Note that our processors today are very fast. If you thought maybe 1974 computers would be too slow, then surely today’s serial hardware, 16 million times faster, is fast enough.
I am trying to address the social problem. My advice to everyone is to join the SciPy community.
-Keith
Though I agree with a lot you say, Keith, I’m going to have to say that I think you’re probably naive in your idea of how easy it is to collaborate. I was talking with a stranger the other day about ROS (the Robot Operating System), and he said that he rejected it for his company because he didn’t want to have to invest in supporting and learning the vast existing architecture. It was quicker and cheaper for him to work from scratch. On a more AGI track, which I think it would be possible to merge the codebases for Adaptive AI, Numenta, and Novamente, the time and energy to do this would likely increase with the product of the sizes, or even exponentially.
What’s worse is that the efficiency of a collection of code, and the cost of maintenance (bug hunting, etc) certainly increase exponentially with the size of the codebase, so I think that it’s almost always a better idea to keep AI architectures separate unless they are very similar. (This says nothing about whether they’re open, though. And I’d certainly like to see more projects along the line of OpenCog.)
While we’re talking about efficiency, I feel like I must point out that while a Von Neumann machine can emulate any system, it can not necessarily do it quickly. The whole point of this post is that the hardware is a vital component in AGI because of the speed (and size).
And lastly, if the computers in the 70′s were a trillion times too slow, then a 16 million fold increase isn’t going to do it. We don’t know how much power we need yet, but there’s a general consensus that we’re not there yet.
– Max
Hi Max:
I’ve written product code in big codebases and teams so I’m idealistic but not naive. What I’ve seen is that that some codebases are easier to collaborate in than others. And I’ve found that a language like Python makes code much easier to be shared.
You criticize ROS, and my response is that it is written in C/C++ and perhaps the code is flawed. When I looked at ROS I saw that it had re-invented the world on a lot of things that having nothing to do with robots. You find the same problem with OpenCV, which defines an entire world down to the matrix class. If OpenCV or ROS was written in Python, a lot of the code would disappear, and integration for newcomers would be easier.
I agree that merging C++ codebases is very difficult. I don’t ever recommend doing that. I would recommend a native re-write at the same time as that would make it much easier. What you would do is use a foundation of Python, and then pull the interesting bits out of the C/C++, port them, and chuck the mass of duplicate infrastructure that already exists in SciPy. The codebase will get about 10x smaller. It is a big task that needs planning. But you guys are all very smart. So you can prove it by porting to Python and create a lingua franca for our scientists to work together in.
Now if everyone were to agree on using a C++ codebase, they could iron out all the problems with it. The problem is that the codebase is such a mess that it isn’t getting huge amounts of volunteers to help solve it. The complexity is preventing it from reaching a critical mass. But we should have abandoned C/C++ years ago anyway, so it shouldn’t be considered for any new re-engineering investment.
Do you know where our software would be if we had been doing 100% free software in Lisp since 1959? The distraction of proprietary software, C, then C++, and then so many new languages now are killing progress. You can look at all the inefficiencies in the worldwide software community and conclude that they are trying to lose. Standardizing on some language for scientists, and then the rest of user mode, is such a huge advantage.
There were a lot of people in 1973 writing code in Lisp and making AT demos and advancements. They weren’t saying: “This computer is too slow for me to do my work”. They were saying: “give me more time and I can do something better.” (Or, “I’d like to find another person to work on this with.”)
Also, an AI system can only be as smart as the data fed into it. A computer back then couldn’t be very smart because it didn’t have access to the Internet. We’ll get to find out the hardware requirements. Note that there are two separate issues: memory and CPU.
I realize there is a consensus that the hardware isn’t ready yet — that is what I’m trying to counteract
I realize that it should be easier to get our AI people working in Python than the Linux desktop so I am trying to start by talking to the scientists first.
Computers also have parallel capabilities like the GPU. There are even people working on adapting GPUs for AI, but the effort is again scattered amongst a bunch of separate codebases and so not accumulating.
So while I do say that it is a social problem, I do agree there is a technical issue in using a modern programming language that helps with collaboration. I try to discuss just one aspect at a time, and Python already exists, and so the problem is again ultimately a social thing.
Regards,
-Keith
Keith,
You clearly have a lot of experience with software development. But i have to ask the question: have you ever tried to implement anything like AGI? Not an expert system, or neural network, or fuzzy logic, or a Bayesian network, or any other kind of narrow AI meant for a very specific purpose but which, as Minsky so aptly put it, wouldn’t know to come in from the rain? Max suggested you may be naive, and that’s a fighting word, but unless you’ve tried to solve an AI problem that in any way also scales to solve other unrelated, non-trivial problems, i’d have to concur.
If the answer is yes, then i’m eager to learn what it is you did, and join any collaborative effort that you start. But if the answer is no, you need to appreciate that there are no specifications, no standards, no protocols. Hell, there aren’t even any agreed upon requirements. Do a search on “what is intelligence”; you’ll find as many unique definitions as there are definers.
Note: i’ve tried to build a scalable AGI many times, and so far have failed exactly as many times (minus 1 – my current work, which hasn’t failed just yet). My own research code bases are all written in the same language, but they are not interoperable at all because my approach changes significantly each time. (The language is not Python, but you can’t seriously be suggesting that that language will solve all my problems.) I’d like to think i’m converging on a solution, but only time will tell. Yes, give me more time and i can do something better – but when is that not the truth? And i’ve spoken at length with many people about it, but opinions and ideas are more varied than my code bases. I even spent some time at Numenta talking to Jeff and Dileep, but even their approach is unsatisfactory to me, enough that i can’t use their code either. So, if nothing else, take my word for it that AGI is still a very, very large technical problem, still deep in the realm of research, not engineering.
But i also agree that there is a big social problem here. The economic and social issues surrounding the existence of an AGI are enormous. The person or group that “controls” the AGI may be in an unprecedented position of power. It’s obvious why researchers might hold their cards close to their chests. (Including myself: i mentioned in a previous post that i’m working on an idea with promise that i didn’t want to discuss.) The problem, of course, is that corporations like Microsoft et al that have little regard for intellectual property, licenses, and even – as we’ve seen – patents wouldn’t waste any time applying whatever resources they felt necessary to get to the finish line first. No one knows how it will really play out, but there’s an argument here that this is akin to a Manhattan project.
Hey Forrest,
“Would be great if you could point me to any strong arguments for exactly what level of modeling detail is required.”
Sadly, i know of none but my own, and i don’t know if it clears the “strong” bar. Actually, i know it doesn’t. Yet. Only a proof in code would change that, and i’m not there yet.
m@:
Microsoft has plenty of regard for software licenses and intellectual property It values its own!
Saying Strong AI is like a Manhattan project is intellectually lazy.
The fact that you think to question my resume points out a flaw in your approach. Remember Einstein, the patent clerk?
It also makes me mad as I realize that if you aren’t open-minded then typing up these comments is a waste of my life. I actually don’t care what codebases get used, nor which of you does the work that is relevant. My interest in typing this up is not because of any debt I owe you, so if you want to have a discussion, I suggest you not insult.
As to your specific question about building an AGI, no, I haven’t tried to implement an AGI. However, that is just one question, and a bad one. A better question is whether I’ve looked at the source code to an AGI. You don’t have to write something to understand it. In fact, it only takes reading for understanding. If you care, I’ve done enough reading of software and English to be comfortable in my understanding of the concepts.
AGI is just one problem domain in software. I’ve worked and studied many problem domains, and I see connections between those experiences and AGI. So even having no knowledge of Strong AI would could still mean that I have something to contribute.
I appreciate you think I’m naive. I can recommend you read my book which has a lot more meat than I can put in these boxes.
There are standards, specs and protocols for Strong AI. We have it for Strong AI just like we have it for countless other problem domains. Above you wrote that we had been working on AGI for 60 years. So how can you also think we still have none of that? I may not have the same experience you have, but I can recognize a spec when I see one.
The thing about Strong AI, is that it is a collection of many other kinds of intelligences. It is an integration problem, which is why Python is so important. I am not suggesting Python will solve all of your problems. I’m saying it is a good language that could provide a lingua franca for our scientists. What are you using now?
I think you overstate the remaining technical difficulty of AGI by your comments above, but I’ll accept for argument it is still a research problem. My response is: just get the researchers working together so knowledge can accumulate. That advice works for any type of software problem, no matter whether engineering or research. Now I happen to think that every research idea needs a lot of engineering to be implemented and tested, so that every problem is always more engineering than research, no matter where it is in your false continuum.
I don’t worry about anyone controlling AGI just like I don’t worry about anyone controlling Linux. Maybe some people want to control it, but they won’t be able to. Furthermore, I think no proprietary effort will succeed in building it. I also see the free software movement as taking over.
It might be true the human instinct to be secretive of your plans. Don’t worry, I don’t care what you are doing. I do find it ironic that the paranoia about getting an idea stolen can cause someone to fail in their efforts to be a part of something that makes a significant accomplishment. We need to get all the paranoid people crouching in the shadows, scribbled secrets clutched in their fists, working together, for any of them to succeed. Proprietary software encourages such thinking. Many people in have forgotten the scientific tradition.
Keith, I think we’re all in agreement that open software, open designs, and lots of collaboration would be awesome. GoiD was (and is) an attempt to bring researchers together for just that reason.
I think our only contention is in what I might call the “Goertzel approach” (though this is perhaps a bit unfair), where the key to AGI lies in gluing together a bunch of different attempts, and figuring out how each can wire into the others. After re-reading your comments, I’m not so sure you’re advocating this approach as much as calling for a “standard” codebase to eliminate hand-coding little things like matrix multiplication, or whatever.
I’d also just like to quickly say that I don’t mean to offend or give you the impression you’re wasting your life. I’m trying to assist and learn, same as you.
(P.S. Einstein may have worked as a Patent Clerk, but he continually wrote about physics throughout his life. It’s not like he suddenly knew general relativity after a life spent writing about other stuff. I think Matt was asking what you’ve spent time developing, professionally OR in your free time. That’s just how I read it.)
I am not familiar with the Goertzel approach versus other approaches. I don’t follow the discussions that closely. My major focus is on free software and Python. I am arguing for a standard lingua franca software foundation for our scientists. Our scientific community can’t afford to invest in tons of different programming languages.
It was m@ who made me wonder whether I was wasting time. He thought that writing AGI system from scratch was a pre-requisite for contributing to this discussion. Thinking you can write a codebase for a massive problem domain by yourself is a big part of the proprietary mindset.
Einstein released his first big paper while he was a patent clerk. His later work hadn’t been done yet. That is a big point. Anyway, software engineering is not physics. If you can understand AGI, you can understand garbage collection and the value of SciPy.
I could tell you of my experiences but the key is that it was in a number of problem domains. That is the biggest difference I have compared to most other people in the FOSS community. I can say that I did a lot of researching and thinking while writing my book, and so I have pages of strong arguments for Python, for example.
-Keith
Keith, you certainly have a way of making friends. I never questioned your resume, i respectfully asked a relevant question, and you respond by berating me – an odd tactic for someone pushing (some would say Pollyannishly) the wild benefits of collaboration. Having personally run 5 FOSS projects and contributed to several others, i can assure you it’s not all that.
I’m relieved that you’ve read about AGI. Me too. As i’ve offered twice already, when you start your FOSS AGI project that shows some promise, you can count me in. Until then it may be premature to believe you know of what you speak.
@Matthew
I’m not trying to make friends or enemies, just discuss how to get technology faster.
I just explained why I got mad. You had “concluded” that I wasn’t qualified to know what I’m talking about. That came after your question. It isn’t about me, it is about getting our scientists to work together in a standard GC codebase so knowledge can accumulate.
I don’t need to have started an AGI project to know what I’m talking about. And I don’t need to start an AGI project to fix this problem. All I need to do is talk to enough AI programmers.
I reminded myself of Pollyanna. It comes from my belief that our hardware is more than fast enough, that we already have more than enough scientists, that we could have built Strong AI decades ago, and would have if Lisp had taken off for user mode.
Hey Keith,
While the gloves are off…
My favourite line is, “no, but i’ve thought about it a lot”. (Second best is, “i don’t need to have [actually done anything]…”.) Well, that settles it then. How did your resume to Microsoft read: “I’ve thought about software development a lot, so you should hire me.” Or, maybe Barnum and Bailey: “I’ve thought about lion taming a lot, give me center ring.” Or, Delta Airlines: “I’ve thought about flying airliners a lot, so i’ll take the NY to LA route, thanks.” No one cares what you’ve thought about Keith, only what you’ve done. And by your own admission, that’s nothing.
But your ego isn’t the only thing that’s only in your head. So is your book’s thesis. Theoretically it’s all flowery and nice: why can’t we all just share and get along? But in practice it’s about as useful as the Communist Manifesto. Every one should code what they can, and take what they need, right? It doesn’t take a genius to figure out how that will turn out. And you have the gall to call me intellectually lazy? I suspect Ray Kurzweil thinks the same as me, which would explain his lukewarm response to your breathless open letter.
But i believe you already know this, which explains your outright hypocrisy. If it’s all so good for me to give away the code that i spend man-months writing, why can’t i download your book for free? It’s a social entreaty, is it not? It provides the “scientific” raison d’etre for FOSS. Why should i have to buy it? If you actually believed that your book had any chance at changing the world for the better, you’d give it away. You’re just out to make a buck, which is why you’re trolling blogs like mine looking for any reason to post a linkback to your own site.
Matthew:
I don’t mind hearing your feedback on anything.
Regarding “I thought about it a lot”: I was just making the point that you don’t need to write software to understand the concepts behind it. You seem to believe that, and I was explaining why I think that is erroneous. I was just telling you what I had done because you asked.
I never said “I don’t need to have done anything”. I just said I don’t need to have done X to say Y. You seem to believe that doing is the only way of learning, and I have personally found it not true. I’ve worked with teams who spend years reading English and software, and writing only English, before writing the actual code. I see analysis as top down, and writing code as a later stage.
Thinking (and researching) is doing. I only told you what I’ve done because you wanted to know. If you find my ego a problem, I’ll just say that you are the one who has introduced these personal topics and now you don’t approve of my answers. If the topic is irrelevant, don’t judge me based on my answers.
You happen to think that thinking is not doing. Okay, whatever you think
I guess we are at a standoff then. In such a case, the default behavior is to just ignore my resume and consider the ideas strictly on their merit.
Okay, on to the rest:
I’ve discovered that free software has nothing to do with Marxism. I have put a proof in a book. The biggest difference is that free software is about bits of science, and Marxism is about atoms. I have discovered free software works better for the free market than proprietary software. Marxism is actually more like proprietary software because both have too much control and strangle the innovation in the edge.
I don’t make my book freely available right now because it is not science. Some have described it as ramblings. I don’t think every 1 and 0 should be licensed the same way. I’m part of the free software movement, not the free music & ramblings movement. Anyway, this is another issue about me, and not about the topic at hand so I will just go on to your next point.
Your next point is that I’m writing all of this to make money. Another irrelevant personal discussion. I think that is laughable as making money on books today is incredibly long odds. I write this here because I expect most programmers do not want to read my book, and so I try to summarize the points to save them time.
I don’t believe my book will change the world. I think the emails I send around to programmers will do the same. I don’t expect today’s programmers to read the book. You can read it for free. I can charge and still be available for anyone who really wants to read it.
It is silly to think that I would go to the trouble to write these posts to get linkbacks to my blog from this tiny little blog. I can get 1,000s of hits for any post I write. I write here because I think you and the few others might be open to new ideas. I’ll make the trouble to talk to a handful of people although I’m realizing to focus more on larger audiences.
Fortunately, I’m using a few of the points I’m making in this mail in another mail to the LKML. I will assume that the mistakes you are making are ones will make as well. I also just added a sentence to my book’s page saying that every penny will be invested in free software.
I only said that saying Strong AI is like a Manhattan project is an analogy from very very high up. I was using lazy to apply to the analogy, not to you. It is paranoid to think I would conclude you are lazy from a random wrong analogy you have made.
My motivation is of no concern to you. I might even have multiple motivations. That is such an anti-scientific perspective. Finally, you don’t consider the obvious one: that I believe I know how to get robotic butlers faster, so it is just a matter of getting others to do some things. Why isn’t that a perfectly good justification? If you had a choice between a million dollars and pervasive robotics for everyone, which would you choose?
Furthermore, what about the idea that every penny I earn on the book will go into free software development? That giving me this bucks is just giving me some money to invest in making things happen faster. Does that fact change your opinion much? I didn’t tell you that before, but you never asked and it isn’t really any of your business. There are many ways to improve the world.
I am interested to know how much you judge me because you perceive arrogance and greed and that I might be a hypocrite. I think I’m none of that, but it is interesting to know as you are a potential customer. Your weirdnesses in your way of thinking, the wrong things you have concluded about me, is something I should think about.
Anyway, I diligently responded to all of your points. You again may not like my answers, but there is nothing I can do about it. I’ll just end by saying that I’m happy to keep going if you can focus more on technical things. Notice, I never ask about your qualifications or motivations. I’m doing my part to keep this focused.
Rambling pretty much nails it, yeah. Whatever, Polly. Good luck.