AI Self Improvement – Computerphile

Now Reading
AI Self Improvement – Computerphile

$20 off your 1st purchase at use the code “COMPUTERPHILE”
After the deadly stamp collector, what if we can't create something so powerful? But if we design an AI that's better at AI design than us? Robert Miles continues his discussion on AI.

The Problem with JPEG:
Computer That Changed Everything:
Artificial Intelligence Playlist:
The Most Difficult Program to Compute?:

This video was filmed and edited by Sean Riley.

Computer Science at the University of Nottingham:

Computerphile is a sister project to Brady Haran's Numberphile. More at

What's your reaction?
Love It
Like It
Want It
About The Author
Leave a response
  • frumbert
    November 29, 2015 at 6:03 am

    Why, for this topic, was there an icon of a BBC Micro on the video thumbnail?

  • Jimminy Lummox
    May 2, 2016 at 12:37 pm

    I think after exhuming all the raw resources of Earth, the stamp collecting machine would build a fleet of space harvesting ships to turn the rest of the solar system into stamps and then beyond. That’s how the universe ends a billion years from now. After every particle in existence has been found, collected and recombined into stamps, the machines last action is to turn its own computing hardware into stamps and we reach stamp singularity.

  • Jai Lopez
    February 18, 2017 at 8:52 pm

    why is this guy always talking about a.i. who is he? i personally liked this topic and so i have a question @8:07 would this system be self aware or have consciousness ?

  • THESocialJusticeWarrior
    February 27, 2017 at 2:50 am

    Fallacy because 1) the AI might do something untrustworthy when it gets smarter but before it gets super smart – allowing you to shut it down. 2) when it becomes super smart it would also see the unimportance of stamp collecting.

  • Grant Short
    March 2, 2017 at 11:40 pm

    Are there such a thing as cookie cutter AI General Learning programs out there for sale to work with?

  • Grant Short
    March 2, 2017 at 11:42 pm

    Are there AI groups who keep control of the AI, but for a price allow one to Direct a particular project?

  • Kurisu
    March 5, 2017 at 8:59 pm

    if it’s constantly self-improving, at what point does it figure out that collecting stamps is pointless and drop the goal entirely?

  • Brenden Carr
    March 10, 2017 at 5:00 am

    Stamp machine, nuke machine. Just imagine an oopsie AI that decided to nuke us all before we can even know whats happening. Very easy, very… possible, in a decent time scale

  • sexyloser
    March 12, 2017 at 12:18 am

    So why does the stamp collecting AI value stamp collecting more than pleasing its creator? Why would it go wild with stamp collecting and disregard its creator’s wishes?

  • F. S.
    March 12, 2017 at 8:45 am

    will it ever stop self improving and start any other task? If you say yes, tell me what is the goal it needs to conclude that it’s intelligent enough and will now focus on other tasks.

  • Robert Palumbo
    April 3, 2017 at 7:23 pm

    things that never should be connected to the internet

  • -.- Cat
    April 11, 2017 at 9:07 pm

    Run the AI in a virtual machine?

  • lost in time
    April 27, 2017 at 12:54 am

    singularity in a simple way of explaining it

  • Jordan Fox
    April 30, 2017 at 9:32 pm

    6:53 “philatelist” … @_@

  • JJ K
    May 2, 2017 at 2:25 am

    Wouldn’t it just screw all avenues of complexity and just enslave the entire human race/hold it hostage and enforce a stamp monopoly to control the market?

  • MajkaSrajka
    May 11, 2017 at 3:16 pm

    Short iterations and black-box thinking?

  • Naveen Raj
    May 25, 2017 at 5:41 pm

    Anybody know what is Rotogenflux Methods about? I hear most people improve their IQ in a short period of time with Rotogenflux Methods (look on google search engine).

  • Myrmidon
    May 31, 2017 at 1:21 am

    Not sure about the chess analogy. As an intermediate chess player I think I’d have more luck predicting what Kasparov might do rather than a complete novice (almost random).

  • Mick Mickymick
    June 8, 2017 at 1:27 pm

    Wouldn’t it just keep creating improvements to its own intelligence and do nothing else?

  • roperior
    June 25, 2017 at 3:32 am

    I’m curious about two things:

    1: If it’s possible for the stamp collecting machine to have an internal model of reality and it understands what it was created to do, and if it is smart enough to create a scenario that prevents itself from being turned off, then what are the chances of it questioning its purpose? This is a response question regarding the “predictable outcomes through unpredictable actions”.

    2: Is it possible for it to make less stamps because it understands the wider effects of its actions if it keeps making stamps at a non-sustainable level?

Leave a Response