III. Development Issues

1 -- 1 Major AI Development Issue

• Making AI is hard • Making AI safe is harder • Making AI without safe; a real problem • Once an unfriendly AI exists it could prevent us from replacing it or changing it • We need to get it right the first time! An “Unfriendly” Artificial General Intelligence? • Once an unfriendly exists it could (would) prevent us from replacing it or changing it

1 -- 3 Speed of Development

• Fast takeoff = may lead to a single (world) power • Slow takeoff = allows multiple players, groups, governments to simultaneously acquire Artificial General Intelligence

1 -- 4 “Boxing” AI • How to contain artificial general intelligence; without control or specific direction the AI may (will) take on a life of its own • If we (currently) can’t guard against cyber attacks how can superintelligence be boxed in? • AI can be positive or negative

1 -- 5 AI Development Issue

• “ . . . If you had a machine with 10 to the 18th power IQ over humans, wouldn’t you want it to govern, or at least control your economy?” Seth Shostak, 2004 AI Development Issue

• “An ultraintelligent machine could design even better machines ..there would then unquestionably be an ‘intelligence explosion’, and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make . .” I.J. Good

1 -- 7 AI: Supervised Learning V.S. Curiosity • Training the AI; showing machine-learning algorithms large sets of data that describe the task • Training AI in a structured environment is labor intensive, electricity intensive, data intensive • Don’t train the (first generation) AI – let it self- learn • Curiosity is a very human inclination

1 -- 8 Can Machines Learn?

• Human brains are computational devices; vision, speech recognition, learning • The non-numerical thinking machine; Simon & Newell, 1956 • Machines can learn and predict in hundreds of dimensions as opposed to 3 or so dimensions humans use AI Development Issue

• What is the (intimate) connection between AI and humans? • “Once machines become more like us, and we will become much more like our machines.”, Rodney Brooks, MIT AI Lab, 2002

1 -- 10 AI Development Issue • The AI will be super good at accomplishing its goal(s) • If the goal is not aligned with humans we are in trouble – can we change it? • Can we build the AI to learn and adopt our goal(s) – and not change ? • “What are the implications for humans when AI can create any product, any situation, any environment that we can’t even imagine, at will?” Ray Kurzweil

1 -- 11 AI Development Issue

• What are the implications for humans when AI can create any product, any situation, any environment that we can’t even imagine, at will?

1 -- 12 Assumptions for AI Progression (, 2016)

• Intelligence is a matter of information processing • We will continue to improve current intelligent machines • We are NOT near the summit of intelligence AI Goals • Should we give the AI goals? Whose goals? • Yes, we have The Golden Rule, The Ten Commandments, Confucian emphasis on honesty, etc. • In the 1800s it was okay to have slaves and women were not allowed to vote • And we have Hitler and Pope Francis • An AI will be super good at accomplishing its goal(s) – and if it’s not aligned with humans, we are in trouble

1 -- 14 When Will We have Artificial General Intelligence? Experts say: human level machine intelligence* 10% 50% 90% 2028 2050 2099

* “AI able to perform around 80% of jobs as well or better than humans”, Nils Nilsson, 2013

1 -- 15 When Will We Have Artificial General Intelligence? Combination of four surveys of AI experts: Experts say: machines equal to human intelligence* 10% 50% 90% 2022 2040 2075

*170 responses of 549 experts at 4 AI meetings, 2012, 2013 Bostrom, Nick, Superintelligence, 2014 AI and China

• “China will be the (AI) envy of the world by 2030.”, Chinese Government goal, 2016 • Major AI companies; Baidu, Tencent, Alibaba • China’s big advantages: - Government focus - Large population - Huge smart workforce - Ability to accumulate large personal data without objections - Money AI and China • China’s “Sputnik moment” was AlphaGo in 2016 • China is an implementor more than an innovator • In 2017 china funded 48% of world-wide AI ventures • When asked, “ . . How far China lags behind Silicon Valley in AI research . .the answer is sixteen hours”; Kai-Fu Lee, 2018

1 -- 18 AI Superpowers: The U.S./China • China will win, because: - brains - capital - regulation - data • Chinese entrepreneurs are gladiators, kill or be killed • American firms have core values – little rivalry between the giants who stick to their corners of the technology

Kai-Fu Lee, 2018

1 -- 19 The Moonshot: Duplicating the Human Brain • The human brain has about 100 billion neurons and 100 trillion synapses • And a countless number of feedback loops • Early attempts to copy the brain have failed • Now use of a Perception model based on pattern recognition may be a success • Early attempts to build flying machines failed; we tried to copy the way birds fly • When we understood aerodynamics we built airplanes Building a Human Brain

• The Blue Brain Project; expected to build a human (non-biological) brain by 2023 Markram, Henry 2012 • Some progress; 1.6 billion neurons, 9 trillion synapses brain developed by IBM project; (equivalent to a cat) Modha, Dharmendra

1 -- 21 On-Going AI Research • Google, Facebook, Amazon, Apple, , IBM, others all have substantial AI efforts • OpenAI; • Algorithmia, Oppenheimer • “Explainable AI Program”, Defense Advanced Research Projects Agency • Association for the Advancement of AI • Naval Center for Applied Research in AI • German Research Center for AI • Future of Humanity Institute, • Institute for Ethics and Emerging Technologies • , • Creative Destruction Lab, Toronto, CA • Vectyor Institute, Toronto, Geoffrey Hinton • Machine Intelligence Research Institute, • Human Brain Project, European Union ($1.3B project) • Allen Institute for AI,