Hi there,

One thing that has been on my mind with AGI and as far as I've read
with current system proposals,
>From what I've read so far it looks like it hasn't been taken into so
much consideration (??) if there is a paper on this please link me to
it!

I understand AGI is far from being formalised but the internet was
designed without security in mind and as a result people have suffered
and still has some very easy channels for criminals to exploit.

My first thought, well you could harden the OS that hosts the AGI
software which would mitigate most traditional attacks,
but what about new forms of manipulation? Perhaps via the sensory
inputs of the AGI system.

If the AGI system learns from it's sensory inputs and able to process
enough information - someone could hijack these inputs, send strong
experiences to the system in a short amount of time, in opencog,
altering the atomspace - perhaps based on a particular trigger - for
example on the election date to assassinate the president, or
something simpler to changing association 'humans are friends' to
'humans are enemies..' (kill all humans!)

You could have an auto-update feature - perhaps in a mindplex that
could restore portions of the atomspace, but you'd only have to
intercept this protocol and manipulate it to your will, or take the
machine offline. This 'psychiatrist AGI' should have no inputs
connected to the net, locked away in high security vault, and is
capable of traversing atomspaces to look for irregularities.

People are already worried about machines doing our thinking for us -
what happens when they become nutjobs?
Food for thought..

- Jarrad


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to