Just for reference youtube would need 18000 humans in seats watching youtube 24/7 to have human screening of youtube. Say around 72,000 employees give or take. If my maths is right (and it could well be out by an order of magnitude) that's is a nice round billion dollars in wages cost at US minimum wage. That's presuming they are watching in real time of course not some kind of clockwork orange torture chamber with 10 videos simultaneously at warp speed then just firing them when they miss something.

On 10/4/19 11:54 am, Scott Wilson wrote:
I feel like legislation will compel tech companies to implement human screening in some capacity, and there will be huge downsides to that - I mean, which is more likely:

a) screening team members are offered abundant mental health support resources, given follow-through on reporting (that video you flagged last year resulted in a conviction and a jail sentence, congratulations!) and are limited to short periods...

or:

b) screening team members are a minimum wage disposable/contractor/gig economy workforce, desperate for any income, performance tracked to the extreme (we require 55 minutes of video content viewed per hour) and discarded when they inevitably burn out?

On Wed, 10 Apr 2019 at 11:45, Nick Stallman <[email protected] <mailto:[email protected]>> wrote:

    I didn't know Tineye could tell if an image was violent or not.

    The existing systems work for copyright purposes, finding a
    similar match.
    This works to some extent currently, and can handle recompression,
    scaling, etc...
    It falls apart when an adversary wants to get around it however.

    But for the case that this legislation is targeting, i.e. taking down
    violent video, fingerprinting is useless.
    It's brand new content - completely impossible to detect in advance.
    You can only remove the content after it's been distributed for quite
    some time, not pre-emptively which is what the politicians want.

    On 10/4/19 11:16 am, Paul Wilkins wrote:
    > https://tineye.com/search/f274c3b49edcca9a6d83994a43629445a5ea5a23/
    >
    > On Wed, 10 Apr 2019 at 11:12, Matt Palmer <[email protected]
    <mailto:[email protected]>
    > <mailto:[email protected] <mailto:[email protected]>>> wrote:
    >
    >     On Wed, Apr 10, 2019 at 10:56:12AM +1000, Paul Wilkins wrote:
    >     > Now I would say that for instance, if the eSecurity Director
    >     posts the CRC
    >     > of a file as being "abhorrent violent" content, and your
    company
    >     doesn't
    >     > expeditiously take down that material, expect problems
    down the
    >     pike. I
    >     > doubt a CRC check alone is sufficient.
    >
    >     Given that a CRC changes if you modify any bit of the file, and
    >     common CRC
    >     implementations have a space of either 16 or 32 bits (65,536 and
    >     ~4 billion
    >     possible values, respectively), "insufficient" doesn't even
    begin to
    >     describe such a scheme.
    >
    >     > I'd say a fingerprinting system to
    >     > match altered copies of the subject file should be
    implemented.
    >
    >     Once again with this magical "figerprinting" scheme. Nothing
    like
    >     what
    >     you're describing actually exists.  Further, there's no
    point in each
    >     company coming up with their own scheme for calculating this
    magical
    >     fingerprint, because if the eSecurity Director wants to say
    "take down
    >     everything like this fingerprint" they have to use the *same*
    >     scheme to come
    >     up with the same fingerprint.
    >
    >     > It doesn't have to work in all cases.
    >
    >     It won't work in *any* case.
    >
    >     > I am not a lawyer. This is not expert advice.
    >
    >     Yes, I think that is quite evident.
    >
    >     - Matt
    >
    >     _______________________________________________
    >     AusNOG mailing list
    > [email protected] <mailto:[email protected]>
    <mailto:[email protected] <mailto:[email protected]>>
    > http://lists.ausnog.net/mailman/listinfo/ausnog
    >
    >
    > _______________________________________________
    > AusNOG mailing list
    > [email protected] <mailto:[email protected]>
    > http://lists.ausnog.net/mailman/listinfo/ausnog
-- Nick Stallman
    Technical Director
    Email [email protected] <mailto:[email protected]>
    <mailto:[email protected] <mailto:[email protected]>>
    Phone   02 8039 6820 <tel:0280396820>
    Website www.agentpoint.com.au <http://www.agentpoint.com.au>
    <https://www.agentpoint.com.au/>


    Agentpoint <https://www.agentpoint.com.au/>
    Netpoint <https://netpoint.group/>

    Level 3, 100 Harris Street, Pyrmont NSW 2009    Facebook
    <https://www.facebook.com/agentpoint/> Twitter
    <https://twitter.com/agentpoint> Instagram
    <https://www.instagram.com/Agentpoint/> Linkedin
    <https://www.linkedin.com/company/agentpoint-pty-ltd>

    _______________________________________________
    AusNOG mailing list
    [email protected] <mailto:[email protected]>
    http://lists.ausnog.net/mailman/listinfo/ausnog


_______________________________________________
AusNOG mailing list
[email protected]
http://lists.ausnog.net/mailman/listinfo/ausnog


_______________________________________________
AusNOG mailing list
[email protected]
http://lists.ausnog.net/mailman/listinfo/ausnog

Reply via email to