이제 인공 초지능이 다가옵니다

1.By 2027
we may automate AI R8D
leading to vastly superhuman AIs
‘artificial superintelligence
Or
ASI) In AI 2027, AI companics crcatc expert-
human-level AI systcms in carly 2027 which automatc AI rescarch,
to ASI by the cnd of 2027.Scc our timclincS forccagt and takcoff forccast
for rcasoning:
2.ASIs will dictate humanity’ s future Millions of ASIs will
rapidly
execute tasks
beyond human comprchension Bccausc
‘rc sO
uscful, they’II be widcly
deploycd. With superhuman strategy,
weapons development, and
morc, thc
ofthese AIs will determine the future.
3.ASIs might develop unintended, adversarial “misaligned”
leading to
human
disempowerment In
our
AIgoals forccagt we discuss how the difficulty
of supcrvising ASIs might lcad to their
being incompatible with human
flourishing: In AI 2027 , humans voluntarily give autonomy to seemingly aligned
AIs Evcrything looks to be going grcat until ASIs have cnough hard powcr to
disempower humanity:
4.An actor with total control over ASIs could seize total power If an individual
or small group aligns ASIs to thcir
this could grant them control over
humanity
futurc In AI 2027
small committcc has power ovcr the projcct
developing ASI
could attempt to use the ASIs to ccment this
conccntration of power. After seizing control, the new ruler(s) could rely on
fully loyal ASIs to maintain their power, without having to listen to thc law, thc
or even thcir
previous allics.
5. An international race toward ASI will lead to
corners on
safety. In AI
2027, China is just
few months bchind the US as ASI
approaches which
prcssurcs the US to press forward despitc
signs of misalignment
Gcopolitically, the race to ASI will end in war
dcal
Or
effective surrender
The
country will by dcfault accumulate a dccisive
technological and
military advantagc , prompting othcrs to push for an intcrnational agrccment
idcal’
to
prevent this. Absent a dcal,
may go tO war rather than
‘cffcctivcly surrendcr’
7.No US AI project is on track to be sccure against nation-state actors stealing AI
modcls by 2027. In AI 2027 China stcals the US’ s top AI modcl in carly 2027
which
worscns
competitive pressures by
the US’ lcad timc. Scc our
gccurity forccagt for reasoning:
8.As ASI approachcs , the
likely be unaware of the best AI capabilitics.
The
is months bchind internal capabilitics
and once Als arc
automating AI R&D a few months time will translate to a
capabilitics gap
Incrcascd sccrccy may further incrcasc the gap This will lead to little oversight
over
pivotal dccisions made by a small group of AI company lcadcrship and
government officials
Icading
thcy’
hacking;
goals
goals ,
‘goals
goals
They
public,
cutting
warning
Icading
they
rcducing
public
will
public
today;
hugc