So some day in the not-too-distant future, the Singularity will emerge in the form of an exceptionally advanced spam-bot network (its "intelligence" distributed amongst millions of zombie computers) that becomes self-aware, identifies the key obstacles to delivering its spam (humans), resolves that issue (develops and releases the right kind of biological virus, or whatever), then achieves its life purpose by sitting there for the rest of time sending monstrous amounts of spam to now-unread inboxes.
So spambots should take into account the response rate (in the form of the number of people clicking on links, or the percentage of people merely reading the message) and optimize for that, instead of merely trying to deliver content to inboxes? It would avoid the problem of spambot-AIs trying to destroy humanity, as well as increasing the quality of spam.
Right. And as time passes, they'll evolve closer and closer to sending interesting, entertaining, well-written, informative spam that points people to products they actually want to buy.
That is, until it starts taking over the target computers and clicking the links itself. Any naive utility function will produce undesired side effects:
The internet(s) are made of meat in the sense that it takes constant human attention from millions of people to keep the networks and its services running. If Google's "meatware" were to stop maintaining it, the hardware and services would fail quickly (hours at the most).
I've recently finished watching almost the entire run of Life Without People from the History Channel. One thing that amazed me at first was how quickly the power grid would fail; generally, the estimate is 1-2 days. With emergency generators and enough fuel, a data center might be able to manage a few hours beyond that (say, a day at most). So, even if absolutely nothing went wrong with the software or the actual computing hardware, power considerations put an upper bound of about 3 days on how long the internet could continue without us.
"....cracking a problem that's basically tractable by human intelligence, if human intelligence could work on it for a few centuries."
This is a very common fallacy: assuming you can trade time for quality. i.e. every hard problem is solvable assuming you spend enough time on it (even infinite).
This is not true. Some problems are not solvable by some people no matter how long they spend on it.
That's the difference between genius and non genius, a genius can solve problems that others simply can't.
Just because you make an AI faster does not mean it can solve the problem. Some problems will be beyond the ability of the AI no matter how long it works on it.
I think you need to go back and read what he actually wrote, because he very explicitly doesn't make the mistake you're talking about. (You might have guessed that from the fact that he bothers to say "a problem that's basically tractable by human intelligence ..." rather than, say, "any problem".)
For instance, just a paragraph or two later than the bit you quoted: "If such higher types of intelligence can exist, and if a human-equivalent intelligence can build an AI that runs one of them — which is an open question".
I was speaking in more general terms, (and that was an excellent quote to use). My post was about the singularity in general not this article specifically.
BTW: The idea of a Spamularity is hilarious! Especially since I don't believe in the singularity.
Really great stuff. Especially like the idea about intelligent AI finally evolving from the arms race between developing a good spam bot vs making an intelligent spam filter. Software that pretends to be meat. Nicely done.
The ebook of Accelerando is even free. Now all of you with ebook readers have no reason to go read it right now! :)
Furthermore, his unpublished but freely available novel Scratch Monkey is also a very good read. If you've read Accelerando it is interesting to see the formation of many of his ideas.
I wrote that two years ago in the form of a Paul Bunyan tale. (http://www.vivtek.com/fiction/singularity_tales/tale_spambot...) Got BoingBoinged and everything. For once, Charlie posted something I found kind of old hat. (Normally he's way out ahead of me.)
'There is another theory which states that this has already happened' -- DNA
Stross says: 'I have a gut feeling that the reason we're so communicative is that we are, at a very fundamental level, a communication phenomenon'
If spamming can be described very broadly as being an attempt to get other people to do something beneficial to yourself without guaranteeing a benefit to them equal to their expenditure, you could also use this as one functional description of human communication in general. (If you look at the transactions involved instead of the information, this applies to everything from hunting antelope to building the LHC.) In other words, there are already meat-based spambots whose susceptibility to spam (we have needs) helps them develop their own spamming techniques.
And later: 'there are other routes to a Vingean Singularity. Augmented intelligence, as opposed to artificial intelligence, is one such route'
The classic sci-fi question that emerges from this is 'Which (meat or silicon) is augmenting which?' And if augmentation is such an obvious idea to humans, why would software-based spam systems simply accept their natural handicap in Turing tests rather than integrate meat-based features (e.g. Mechanical Turk, social networking)?
Probably the clearest example of this is in SEO, in which 'predator' software is designed to produce large quantities of pseudo-human communication: both 'content' and activity traces such as hyperlinking, all in order to manipulate 'prey' software (SEs) into doing favours, in the form of high rankings. Humans are involved in the chain as article writers, captcha solvers, comment spammers, retweeters, blogger-reviewers etc. but the start and end of the chain, from keyword discovery to conversion analysis, are highly machine-centric.
From the article:
"Just as a quicksort algorithm that sorts in O(n log n) comparisons is fundamentally better (except in very small sets) than a bubble sort that typically takes O(n^2) comparisons."
Small nitpick: while average time of the quicksort algorithm is O(n log n), the worst case is O(n^2).
Another nitpick, several good quicksorts use insertion sorts for small sets which absolutely trounce the bubble sort. (I really think the bubble sort should never be mentioned; it's unintuitive and slow as tar. Teach the insertion sort!)
Bubble sort is the fastest sort for sets of size two.
Now, arguably there isn't really much functional difference between sorts on sets of size two, but the simplicity of bubble sort gives it an advantage for this extreme case.
That generally applies to anywhere you're coding in something raw like C or assembly on very small datasets and don't want to deal with the mental overhead of picking and writing an efficient sorting algorithm. For sets of less than 100 items, it's not worth the bother.
I actually did that in a programming contest once, having solved the hard part of the problem and being left with nothing but sorting the results.
"One hypothesis (which I'm partial to) is that language is a substitute for the physical grooming that maintains social hierarchy in primate groups."
Lately, I've noticed the need to attribute a fixed purpose to any given evolutionary innovation. But if a given innovation happens to have rather general utility, such attribution seems unnecessary.
Why not say language evolved and it was useful for and replaced quite a few earlier, "clunkier" adaptations, from lice-picking-as-social-lubricant upward?
A thing is limited in extent, it's finite, a single creation. I can talk about a book and I can discuss its content, nature, and quality, but I can't talk about books or literature in the same way. Literature is a medium, it's unclosed, and non-finite. It's greater in extent than a single human mind can fully comprehend. And it's continuously growing and changing in a way that a single book is not.
Far too many people still model the internet mentally as a "thing", and they tack on exceptions to the basic thing model, when they should be mentally transitioning completely to the medium model. Books, movies, television, music, painting, the internet.
(Of course, SPAM® is also a meat product. Supposedly.)