Self-Improving Artificial Intelligence

Now Reading
Self-Improving Artificial Intelligence

October 24, 2007 lecture by Steve Omohundro for the Stanford University Computer Systems Colloquium (EE 380). Steve presents fundamental principles that underlie the operation of "self-improving systems," i.e., computer software and hardware that improve themselves by learning from their own operations.

EE 380 | Computer Systems Colloquium:

Stanford Computer Systems Laboratory:

Stanford Center for Professional Development:

Stanford University Channel on YouTube:

What's your reaction?
Love It
Like It
Want It
About The Author
Leave a response
  • GarfieldGaming
    December 20, 2013 at 1:03 pm

    What if one of these “bugs” takes over a robotics factory? and a weapons factory? and blacks out the world? and kills everything tha tmoves?

  • zhisko chen
    January 9, 2014 at 5:37 pm

    “HUmans are problems i prefer robots”

  • Emiel Blom
    June 15, 2014 at 8:48 am


  • zLFOBz
    July 5, 2014 at 12:16 pm


  • Iaroslav Shcherbatyi
    September 2, 2014 at 9:36 pm

    He should be more careful with optimality condition at 14:57. At the very least the time functions T1,2(S) should be concave, otherwise it is not necessary that the value of S1(S2) that satisfies this condition will maximize the utility function. And as usually amount of resources is bounded, even for concave time functions T1,2(S) those S1(S2) satisfying this condition might not belong to feasible region (solution S1(S2) might be obtained negative for instance, which makes not much sense if you specify that S1, S2 are in [0,1])

  • Travler X
    September 8, 2015 at 5:01 am

    Published in 2008 and it is such a dated discussion.
    There seems to be a bit of an apples to oranges discussion taking place, and there is absent some of the most relevant parts of discussions as well.
    Depending on what one’s own definition of what an AI is or the prognosis of what an AI will be, an AI is not necessarily going to be able to improve itself to the point of becoming a singularity or even the singularity. Mr. Omohundro does seem to be aware of distinctive differences between the two. As opposed to a singularity which would have a measurable cognition, an AI is a purely computation system which lacks any real SENSE of self awareness but is processing information much in the way DNA does.
    This discussion conveniently ignores the complex nature of the task of creating a viable AI which performs these tasks with enough clarity to not make changes to itself that do not result in its loss of function much less utility. This discussion seems to take place inside the “perfect world” or ,should I say, lab conditions. What happens when such a system meets the reality of having to struggle to evolve itself with only mathematics as its cognition or ,in other words, with only mathematics and no real driving psychology. It’s one thing for DNA that has survived billions of years to do it many times and it is another thing for an electronic abacus to do it successfully once.
    Furthermore, in taking small steps it may prove more prudent to begin by developing a do nothing widget that is not burdened by functionality but can mutate itself just to see how entropy will affect its systems in ways that have not been theorized in draft board development…..
    Of course, I could be wrong about everything I just said, however, I do think that there will have to be a sound philosophical basis needed for a true interactive and cognitive singularity to arise from integrated computational systems that would be more than an insane reptilian robot. It would take a certain amount and type of ethics to keep it from blindly expanding and consuming until it has altered its own environment beyond its ability to adapt to. Assuming, that like all cognitive systems it would not be without its flaws which would show up as irrational behavioral quirks, it would need to have enough situational awareness coupled with its own self awareness to avoid making irreversible fatal changes to itself.
    It’s 2015 and due to the massive propagation of smartphones and the like as well as the pace of advances in neural prosthetic, neural imaging, and neural implants as well as biological nanotechnology it is looking more and more as if the information theory may just cheat and leap straight to the first singularity being a symbiotic entity attached to humans in order to optimize its own capacity by load sharing with humans as a optimizing efficiency effect since humans naturally process some types of information better than machines.

    • dance club
      March 28, 2017 at 5:44 pm

      Is Rotogenflux Methods helpful to your IQ score increased over 17 points? We have learn many good things about Rotogenflux Methods (search on google).

  • saddam hossain
    November 7, 2015 at 5:57 am

    Is Rotogenflux Methods effective to improve your IQ in a short period of time? We’ve read numerous good things about Rotogenflux Methods (search on google).

  • arash m
    January 18, 2016 at 4:17 pm

    is there any higher quality of this speech?

  • Hridoy Mitra
    February 2, 2016 at 3:29 am

    It’s still surprise me just how some people have no idea about Rotogenflux Methods (search on google) despite the fact that lots of people increase their IQ of 23 points using it. Thanks to my personal pal who told me about it. I have increase your IQ of 20 points.

  • Richi Donald
    February 18, 2016 at 11:59 pm

    I read lots of great opinions on the internet about how exactly Rotogenflux Methods (just search it on google) can help you your IQ score increased over 17 points. Has any one tried this popular course?

  • Md Mirazul Islam
    May 29, 2016 at 2:21 am

    As specialist, I do believe Rotogenflux Methods can be good way to
    increase your IQ of 23 points. Why don’t you give it a shot? perhaps
    it’s going to work for you too.

  • Md Mirazul Islam
    May 29, 2016 at 2:21 am

    I read a lot of great reviews on the net
    about how Rotogenflux Methods (do a search on google) can help you your
    IQ score increased over 15 points. Has any one tried using this popular
    intelligence boost system?

  • anwar jahid
    December 4, 2016 at 1:50 pm

    As specialist, I’m sure Rotogenflux Methods
    is actually great way to your IQ score increased over 15 points. Why
    don’t you give it a shot? maybe it will work for you too.

  • Mike Schwarzer
    December 13, 2016 at 12:05 am

    This is Israel cyber warfare.

  • Mostafa Waresuzzaman
    February 24, 2017 at 3:58 am

    Is Rotogenflux Methods useful to improve your intelligence? We have read numerous good things about Rotogenflux Methods (just search it on google).

  • pradeep kumar
    March 21, 2017 at 8:28 pm

    Is Rotogenflux Methods useful to improve your intelligence? I’ve read
    numerous good stuff about Rotogenflux Methods (do a google search).

  • Stephen Paul King
    April 11, 2017 at 2:41 pm

    In our thinking of modeling, we must keep Bonini’s paradox in mind…

  • jr beke
    April 17, 2017 at 5:34 pm

    Is Rotogenflux Methods helpful to increase your IQ of 22 points? I have learn numerous good things about Rotogenflux Methods (just search it on google).

  • Jas sth
    May 25, 2017 at 2:25 pm

    Anybody know what is Rotogenflux Methods about? I hear many individuals their IQ score increased over 15 points with Rotogenflux Methods (just search it on google).

  • Hula Kha
    May 26, 2017 at 4:21 pm

    As expert, I think Rotogenflux Methods is great way to increase your IQ of 22 points. Why don’t you give it a shot? maybe it’ll work for you too.

Leave a Response

This site uses Akismet to reduce spam. Learn how your comment data is processed.