HORVITZ, Eric, the AI 100 Project, MSR, April 20, 2016
<p class="MsoNormal" >HORVITZ, Eric, the AI 100 Project, MSR, April 20, 2016
This is in EE 380 in Gates CSD. Eric is head of MS Research.</p>
<p>
great slides</p>
<p>
shows Jean=Marc Cole painting from 1899..</p>
<p>
shows students wired up with professor throwing books into a hopper...</p>
<p>
and a torpedo plane...</p>
<p>
but shows an iPhone like devise from 1950s with buttons...</p>
<p>
analogy to 1903 Wright brothers...</p>
<p>
shows plane...</p>
<p>
50 years later we had 707s...</p>
<p>
great sign</p>
<p>
this room is equipped with Edison Electric lIght</p>
<p>
do not attempt to light with match...</p>
<p>
(and a reassure...in no way harmful to health)</p>
<p>
100 year study of AI..</p>
<p>
endowed to continue for as long as Stfd exists...</p>
<p>
motivation difficulty of anticipating the opportunities and issues with the advancementof AI</p>
<p>
here's the Charge...</p>
<p>
July 2014...</p>
<p>
to support a longit study of influence of AI advances on people and society...</p>
<p>
John Hennessy quote:</p>
<p>
we feel obliged and qualified to host a conversaiton abt how AI will affect our kids and our kids kids...</p>
<p>
AI is one of the most profound tasks in science and will affect every aspect of human life...</p>
<p>
roots .. the AAAI presid panel on long-term AI futures at Asilomar in 2008 and 2009..</p>
<p>
the AAAI Asilomar Study...</p>
<p>
his theme was open world AI when he was AAAI Pres then.</p>
<p>
in photo Sebas Thrun, Doug Lenat, Andr Ng, etc.</p>
<p>
how abt runaway AI; singularity...</p>
<p>
subgroups</p>
<p>
potential disruption in short term</p>
<p>
and longer timer</p>
<p>
and ethical legal (by David Waltz chaired it)</p>
<p>
short-term: tech surprise: jumps in capabilties (eg Tom Dean and Andr Ng gave talks)</p>
<p>
feb 2009.. they gave Deep Learning talks...</p>
<p>
and criminal AI... new attacks surface, privacy... malware and cybersec to next level...</p>
<p>
reason across a person's devices... have best minds... try to do white hat AI hacks...</p>
<p>
and Stuxnet was then being dev...
similar to criminal AI..</p>
<p>
and human-macine collab... eg airlines that had crashed... china airlines... bad pass off from mahcine to humans...</p>
<p>
LT: loss of control...</p>
<p>
big diff of optinion on time frame of that...</p>
<p>
resch guidelines and relev of containment...</p>
<p>
and logal and ethical issue...</p>
<p>
liability for auton and semi-auton systems
repres , ealrningand infer abt human preferences..</p>
<p>
and psychosocial issues with human-like experiences.</p>
<p>
AI comm... huge divergenc eof opinion...</p>
<p>
shows John Markoff in July 25 2009 in NYT (John is in our aud)</p>
<p>
ai100. stanford.edu...</p>
<p>
standing comm..</p>
<p>
Barb Grosz is chair...
and Russ Altman, Tom Mitchell, and a few others...</p>
<p>
red dot is Asilomar study.. in 2009..</p>
<p>
2015 study panel... this study will be out in June 2016...</p>
<p>
and next report in 2020...</p>
<p>
standing com. interacts with study panel of many experts...</p>
<p>
study for abt 9 to 12 months...</p>
<p>
new project called AI index...</p>
<p>
want to capture # of dimensions of the march of AI...</p>
<p>
including sentiment analysis...</p>
<p>
and convey results to AI res comm and genl public and industry and policy makers...</p>
<p>
Framing Memo...</p>
<p>
ai100.stfd.edu/reflections and framing..</p>
<p>
tech trends and surprise
key opportunies... etc.</p>
<p>
expand on that...</p>
<p>
low-hanging fruit for translation...</p>
<p>
avoidable deaths and injuries...</p>
<p>
what are the bottlenecs to that kind of dev...</p>
<p>
eg CARS; worldwide > 1M deaths per year 1.25 deaths/ year a death every 25 seconds.</p>
<p>
US 32,675 pr year in 2014 in USA 90 deathsper day...</p>
<p>
shows pic of his 16 year old boy when he got license...</p>
<p>
but big incr linked to distracted driving...</p>
<p>
and preventable deaths of hospital pts
44K to 98K per year from Inst of Med in 2000.
and now up to 440K per yea in 2013 (J of Patient Safety)
now called Natl Acad of medicine (name change this year)</p>
<p>
*
AI and Ethics...</p>
<p>
unethical ... with no human in the loop..</p>
<p>
aud comment.. what you call autonomy is actually decisions preprogrammed by programmer...</p>
<p>
shows kids skating in front of a car...</p>
<p>
(proxy intelligence)</p>
<p>
know when to ask and turn decisons over to people...</p>
<p>
Human collab with machines eg in ISRG...</p>
<p>
shows mixed initiative with suturing under robotic control... mixed initiative...</p>
<p>
encdoing rpefernces abt tradeoffs...</p>
<p>
eg controling true positive rat eand false positive rate control ROC curver..</p>
<p>
AI democracy and freedom...</p>
<p>
how abt a study that assists the election of Don Trump...</p>
<p>
AI and warfare...</p>
<p>
more precise, humane, fewer casualties...</p>
<p>
concern abt unintended conseques...</p>
<p>
how abt pulling people out of the moreal loop..</p>
<p>
compet. rpessure to build system that cna act first (and take human out of the loop)</p>
<p>
Ash Carter... US DoD direc 3000.09..</p>
<p>
auton and semi-auto weapons sytem shall be designed to allowed comanders approp level of response...</p>
<p>
see John Markoff in Feb 28 2016... NYT... dangers of human in the loop or not...</p>
<p>
*
might we lose control theintel we build?</p>
<p>
is their tech rational for the rise of superintell...</p>
<p>
are these Qs relevantnow or in 100 years from now.</p>
<p>
Jack Good from 1965... Ultraintell machine... Eric says... a friend of his...</p>
<p>
passed away shortly after 2009 meeting...</p>
<p>
shows transcript fro Alan Turin in 1951 in BBC broadcast... in his handrwriting...</p>
<p>
similar to Jack good quote...</p>
<p>
Samuel Butle's Erewho... late 19th centruy...</p>
<p>
can we backtrack from imagined poor outcomes.. (eg from Forbin Project outcome)</p>
<p>
eg large scale totalitarian regime...</p>
<p>
EH video in Jan 2015.... when he won Feigenbaum prize...</p>
<p>
we will get incred benfits from this...</p>
<p>
E O Wilson AI is essential to the surivval of life on our planet
Oct 2014...</p>
<p>
Study Panel 2015 in 2016</p>
<p>
Peter Stone is chair of it UT Austin</p>
<p>
perhaps 1 to 2 dozen people</p>
<p>
focus now</p>
<p>
AI and Life in 2030..</p>
<p>
what are the poss advances over the next 15yers
proactive efforts on tech, design, and policy...</p>
<p>
what are advances that would have greatest potential...</p>
<p>
and public policy...</p>
<p>
will get report in June 2016...</p>
<p>
John Markoff: McCarthy coined term AI to AVOID the controversy abt whether its actually intel...</p>
<p>
Turing used term machine intelligence...</p>
<p>
was it Manchester or Geniac...</p>
<p>
</p>