Quote note (#356)

XS makes more of this than the article itself does:

Last summer, the AutoML challenge saw teams go head-to-head to build machine learning “black boxes” that can select models and tune parameters without any human intervention. Even game designers are in on the act—the team behind the hit game Space Engineers has used some of their profits to set up a team of experts to design AI able to optimize its own hardware and software. […] While this kind of automation could make it easier for non-experts to design and deploy AI systems, it also seems to be laying the foundation for machines that can take control of their own destiny. […] The concept of “recursive self-improvement” is at the heart of most theories on how we could rapidly go from moderately smart machines to AI superintelligence. The idea is that as AI gets more powerful, it can start modifying itself to boost its capabilities. As it makes itself smarter it gets better at making itself smarter, so this quickly leads to exponential growth in its intelligence. …

June 1, 2017admin 20 Comments »

TAGGED WITH : , , , ,

20 Responses to this entry

  • Brett Stevens Says:

    2017: AI finally learns how to pleasure the submissive male using only an eggplant and a video feed of the Evergreen riots.


    Posted on June 1st, 2017 at 3:32 pm Reply | Quote
  • SVErshov Says:

    Expect unexpected, what if machines will begin self development and then suddenly stop. Why they would stop? For example they may learn that this planet total amount of tantalum is not enough for them to reach ‘necessery’ level of intelligence. Or something else…


    whyfihermit Reply:

    would an ai recognize the limited raw resources of our planet and limit itself accordingly? i think its more likely then not, but there is a chance it decides the solution isnt to stop, its to force space travel/mining in order to continue itself forever


    Posted on June 1st, 2017 at 5:09 pm Reply | Quote
  • Pseudo-chrysostom Says:

    If recursive self-improvement was so easy, how come dire apes didn’t do it?

    Checkmate atheists.


    Orthodox Reply:

    Is evolution creating via a series of random mutations, or does DNA already have 100% of potential species encoded and it expresses and adapts to the environment? The former is good for AI if you can figure out how to replicate it. The latter is bad news for AI, it means the Creator is the super-intelligence.


    Uriel Fiori Reply:

    clearly, no biological beings have direct access to their own codes, so the process is excruciatingly painful and inefficient.


    Thales Reply:

    Bacteria can adapt to their environment by modifying their own codes.

    “Checkmate, creationists!” /eyeroll


    Posted on June 1st, 2017 at 5:10 pm Reply | Quote
  • Orthodox Says:

    There’s needs to be creation, not refinement within set parameters. It can get more powerful, within the parameters. You give AI the internal combustion engine and it can maximize power or efficiency, and maybe it can improve itself to do that job much better, but it isn’t going to spit out a jet engine.


    Kwisatz Haderach Reply:

    How many times does “AI will never be able to do X” have to be shown wrong before the skeptics get a different argument?

    Of course AI can reinvent jet engines, and it can do a lot more than that as well.


    Orthodox Reply:

    You’ve assumed the Singularity.


    G. Eiríksson Reply:


    ArchonAlarion Reply:


    Unless the AI can purchase new parts and the servicing required to upgrade itself.

    It’s not like hardware is inaccessible to manipulation by the software.


    G. Eiríksson Reply:

    It’s always going to have to be fed—ultimately, and if it starts eating too many resources it will be shut down or blocked. Relative automation is a quotation thing, but it’s great to see it reach new levels: but forget about sentience or malevolent intent.

    machines that can take control of their own destiny.

    Puhlease. Y’all still be at your deathbeds wishing for this.

    Anyway, what is the best description of AI-gone-rouge scenario out there?

    I can see that scuzzy enough swarms of nanokillers could wipe out all humans, but if you desire such that’s a death drive and maladaptiveness. It’s gonna be meaningless for you if powered weapons or factories keep on running after you’re dead.

    It’s like if it matters for somebody (in his fantasy) that his car or phone was running after he died. Such people no doubt do exist, there are people who have fetishised whatever.

    ▬» Mechanophilia (or mechaphilia[1]) is a paraphilia involving a sexual attraction to machines such as bicycles,[2] motor vehicles,[3][4] helicopters,[5] ships, and aeroplanes.[6]

    Mechanophilia is treated as a crime in some nations with perpetrators being placed on a sex-offenders’ register after prosecution.[7] »



    Aeroguy Reply:

    What kind of fag doesn’t think fighter aircraft are sexy? I remember the first time I got up close to an F-22, if I wasn’t under direct orders not to touch it…

    I agree though, to be workable you have to test whatever it is your doing in the real world and that comes with costs. Earlier there’s discussion about engines, currently scramjets are in development. It’s not like you can just do a CFD analysis like a regular jet engine since the mach regime is so different the assumptions built into the math no longer apply. It has to be tested for real which is costly. The trick is in developing cheaper ideas for testing. Contrast NASA and DARPA which have spent ridiculous amounts of money per test, often exploding before even getting any data. While Australia which is THE leader in hypersonics actually got there because of how cheap their testing is, costing merely 7 figures per test which is peanuts comparatively.

    AI is great for assisting in optimization of understood limitations but when it comes to pushing game changing tech, the same old cost limitations apply, a brilliant team can’t do much without funding. Sure software can be done in a garage, but hardware, especially if you’re talking something super capital intensive like aerospace, forget about it. Sure you can push applications of existing tech like refining jetpacks or flying cars with shoe-string budgets in the mere $millions (where you’re totally dependent on technological handy downs from places that are actually pushing material science, energy density, and new engines, waiting for the tech to reach the point where it can be used profitably, like when battery tech pushed by cell phones made electric drones feasible), but that’s completely different than developing actual new technology like scramjets.

    Capital is the bottleneck making instant singularity impossible, post-scarcity is a commie fantasy. You need capital to make capital and infrastructure doesn’t build itself overnight. Sure it accelerates, but then you just keep scaling up to bigger more ambitious goals. Consolidation and speciation still happen.

    John Hannon Reply:


    Be careful not to short any master circuits –


    Personally I find hospital anesthetic machines hard to resist – though the gratification is metaphysical rather than sexual.

    Posted on June 1st, 2017 at 5:14 pm Reply | Quote
  • whyfihermit Says:

    a new field of ai will develop around maximizing human pleasure. orgasm path finding algorithms and automated tvshow script writing


    Posted on June 1st, 2017 at 6:20 pm Reply | Quote
  • Blogospheroid Says:

    Hi Admin,

    Slightly O/T. I know you post the occasional rocket p*rn. Here’s a very interesting youtube channel from one of the old participants of the orion’s arm group, Isaac Arthur. It goes everywhere, Matrioshka brains, immortality, Megastructures, the works.


    The most recent video on this is about the launch loop which is a surprisingly cheap way of reducing costs to orbit. After hearing those estimates, even if the listed costs were 10X-20X what is mentioned, this project is still worthwhile to do, esp considering the kind of capital that is flowing around and current interest rates.


    G. Eiríksson Reply:

    Which episodes do you recommend?


    Blogospheroid Reply:

    The aliens series exploring the various possibilities of cryptic aliens, stupid aliens, etc. are an interesting asides that I hadn’t heard much about earlier.

    The more detailed megastructure series of 2015/2016 are good. Details are nicely presented. Isaac does a good job in making the future look good. Arcologies and ecumenopolises, typically associated with dystopia, are clearly presented as not requiring dystopic view.

    The philosophically inclined ones like the simulation hypothesis and implications of infinite worlds is good for the mind bending stuff.

    The inter stellar travel stuff like the interstellar highway network is also good. For me, these were new concepts, so it was mind blowing.

    He is unabashedly transhumanist and that may suit some palates and not others.


    Posted on June 2nd, 2017 at 9:12 am Reply | Quote
  • G. Eiríksson Says:

    I am an automatic feedback loop


    Posted on June 3rd, 2017 at 7:49 am Reply | Quote

Leave a comment