From Artificial Intelligence to Superintelligence: Nick Bostrom on AI & The Future of Humanity

i would like to introduce you to my present and  the rest of the world's future i call it stem some people think that some of these things are  sort of science fictiony far out there crazy but   i like to say okay let's look at the modern human  condition if we think about it we are actually   recently arrived guests on this planet think about  if the world like was created earth was created   one year ago the human species then would be 10  minutes old the industrial era started two seconds   ago there have already been 250 000 generations  since our last common ancestor and we know that   complicated mechanisms take a long time to evolve  so a bunch of relatively minor changes take us   from broken off three branches to intercontinental  ballistic missiles so this then seems pretty   obvious that everything we've achieved pretty much  and everything we care about depends crucially on   some relatively minor changes that made the human  mind and the corollary of course is that any   further changes that could significantly change  the substrate of thinking could have potentially   enormous consequences some of my colleagues  think we are on the verge of something that   could cause a profound change in that substrate  and that is machine superintelligence [Music]   artificial intelligence is a rapidly growing  field of technology with the potential to make   huge improvements in human well-being however  the development of machines with intelligence   vastly superior to humans will pose  special perhaps even unique risks   the united states have identified artificial  intelligence as one of the key technologies   that will ensure the united states will be  able to fight and win the wars of the future   potential international rivals in the field of ai  are also creating pressure for the united states   to compete for innovative military ai applications  china is a leading competitor in this regard   in 2017 the chinese government released a strategy  detailing its plan to take the lead in ai by 2030   less than two months later vladimir putin  publicly announced russia's intent to pursue   ai technologies stating whoever becomes the  leader in this field will rule the world   most surveyed ai researchers expect machines to  eventually be able to rival human intelligence   though there is little consensus on how this will  likely happen artificial intelligence used to be   about putting commands in a box you would have  human programmers that would painstakingly hand   craft knowledge items since then a paradigm  shift has taken place in the field of artificial   intelligence today the action is really around  machine learning so rather than hand crafting   knowledge representations and features we create  algorithms that learn often from raw perceptual   data basically the same thing that the human  infant does the result is ai that is not limited   to one domain now of course ai is still nowhere  near having the same powerful cross-domain ability   to learn and plan as a human being has cortex  still has some algorithmic tricks that we don't   yet know how to match in machines whichever  government or company succeed in inventing   the first artificial super intelligence will  obtain a potentially world dominating technology   via super intelligence we mean an intellect that  is much smarter than the best human brains in   practically every field including scientific  creativity general wisdom and social skills   there's this illusion that time horizon matters  if you feel that this is 50 or 100 years away   that is totally consoling but there's an implicit  assumption there the assumption is that you know   how long it will take to build this safely and  that 50 or 100 years is enough time sam harris   is a neuroscientist philosopher and new york  times best-selling author he explains that one   only needs to accept three basic assumptions to  recognize the inevitability of super-intelligent   ai intelligence is a product of information  processing in physical systems we will continue to   improve our intelligent machines we do not stand  on the peak of intelligence or anywhere near it   the problem with the race for ai is that it may  result in poor quality asi that does not take   the welfare of humanity into consideration sam  harris worries that the power of superintelligent   ai will be abused if governments and companies  perceive themselves to be in an arms race against   one another they could develop strong incentives  to create super-intelligent ai first or attack   whoever is on the brink of creating it the  chinese russians indians israelis koreans   japanese and europeans are at least as  motivated as americans to develop advanced ai   currently china is primarily focused on using  ai to make faster and more well-informed   decisions as well as on developing a  variety of autonomous military vehicles   russia is also active in military ai development  with a primary focus on robotics an arms race for   an asi could lead to authoritarianism it can  make it easier for political groups to shut   down negative speech against them it can make  it easier for businesses to assert their rights   over interests of workers consumers in the general  public philosopher nick bostrom expressed concern   about what values superintelligence should  be designed to have a biological neuron fires   maybe at 200 hertz 200 times a second but even a  present-day transistor operates at the gigahertz   neurons propagate slowly in axons 100 meters per  second tops but in computers signals can travel at   the speed of light there are also size limitations  like a human brain has fits inside a cranium but   a computer can be the size of a warehouse or  larger so the potential for super intelligence   kind of lies dormant in matter much like the  power of the atom like dormant throughout human   history patiently waiting there until 1945. in  this century scientists may learn to awaken the   power of artificial intelligence and i think we  might then see an intelligence explosion [Music]   any type of ai superintelligence could proceed  rapidly to its programmed goals with little or   no distribution of power to others it may not take  its designers into account at all the logic of its   goals may not be reconcilable with human needs the  ai's power might lie in making humans its servants   rather than vice versa if it were to succeed  in this it would rule without competition   under a dictatorship of one there is a need to  transition to more democratic forms of political   decision-making it cannot be assumed that ai  will not work to strengthen the power of those   who control it elon musk warned that the global  race towards ai could result in a third world   war to avoid the worst mistake in history it is  necessary to understand the nature of an ai race   as well as escape the development that could  lead to unfriendly artificial superintelligence   researchers in the field who believe that  asis will be friendly want to help creating   an environment that will make it possible for the  asis to be positive many experts however suggested   limiting the power of asis to determine whether it  would be acceptable to limit the power of an asi   it is important to understand what  it means to limit a superintelligence   it could keep any ai system permanently at a human  level of cognition to limit the super intelligence   is not to try to make the superintelligence less  capable it means the superintelligence stays below   a certain level of capability permanently  machine intelligence is the last invention   that humanity will ever need to make the machines  will then be better at inventing than we are   and they'll be doing so on digital time scales  now a super intelligence with such technological   maturity would be extremely powerful and at  least in some scenarios it would be able to get   what it was we would then have a future that  would be shaped by the preferences of this ai there are several technological capabilities that  could form the basic elements of a workable asi   the cognitive element to the ai  will be much like our own brains   these elements would not only have  the ability to learn and comprehend   but also the memory to store all of that knowledge  the ability to think could include morality ethics   rational thought artistic creation scientific  experimentation and especially the ability to   work with logic another important element is  the ability of human-like emotional responses   reason and to communicate in terms meaningful  for human beings whoever succeeds in creating   the first super-intelligent entity they need  to make sure that this new type of intelligence   is democratized understands people and is able to  communicate with them in some way bostrom argues   that we might make a mistake and give this new  entity goals that lead it to annihilate humankind   assuming its enormous intellectual advantage  gives it the power to do so the way to avoid this   mistake is to create an open system and help it  develop artificial intelligence is undoubtedly the   future and the possibility to create a friendly  asi is important as it will have a serious effect   on our future to ensure the friendly nature  of artificial super intelligence world leaders   should work to ensure that this asi is beneficial  to the entire humanity in addition it is important   to test all logical goals before development  of such ai the ai programming would have to be   open to access and limited to only positive uses  such as curing diseases or producing resources thanks for watching did you like  this video then show your support   by subscribing ringing the bell and enabling  notifications to never miss videos like this [Music] you

Channel: Science Time

Views: 368,364

Keywords: Artificial Intelligence, AI, Artificial Superintelligence, ASI, Superintelligence, Superintelligent AI, Digital Superintelligence, Singularity, Nick Bostrom, Elon Musk, Nick Bostrom Superintelligence, Nick Bostrom AI, AGI, Technology, Robots, Science, Science Time

Id: Kktn6BPg1sI

Length: 10min 2sec (602 seconds)

Published: Sat Nov 21 2020