Transcript Document
PRESENTATION INTRODUCTION BRAINS OF ANIMALS ARTIFICIAL INTELLIGENCE HOW DID WE GET HERE? CAN THEY TURN HOSTILE? WHERE ARE WE GOING? INTRODUCTION Computer systems that can: simulate intelligent behaviour and perform tasks normally requiring human intelligence Interact with the real world (do something?) Lower intelligence Control loops interfacing with sensors / actuators Similar to autonomic nervous systems Act repetitively / unconsciously In animals - regulate heart rate etc. Sympathetic = quick and mobilizing Parasympathetic = slower and activated Quickly controlling actuators More slowly monitoring sensors Need to be told what to do Can be left unattended BRAINS OF ANIMALS More intelligent =similar to animal brains More cognisant and less repeatable. Brain is higher control centre: Walking, thinking, intellectual activities Plans / decides how to do things Learns / remembers Originally = continuous Now = discrete (managed by clocks) Can a brain be simulated? Will simulated brains be intelligent? BRAINS OF ANIMALS Copy brain into hardware / software? Essentially identical? Computers have speed + memory Only do what their designers understood Some skills and talents there Some not What are human abilities? ARTIFICIAL INTELLIGENCE Intelligence exhibited by machines / software Can learn by observing General intelligence is long term goal Challenges include Reasoning, planning, learning, perception Philosophical issues Ethics / Character of the mind Brain is analogue NOT digital Human Intelligence may not be an algorithm HOW DID WE GET HERE? 1940s -Idea of child like machine 1950s - Turing 1956 - John McCarthy "Artificial Intelligence” "the science and engineering of making intelligent machines" 1960s - Programming computers "within a generation the problem of creating AI will be substantially solved“ Now… Machines exceeding human performance But … still no ``child machine'‘ We cannot agree what to call intelligent CAN THEY TURN HOSTILE? Is AI going to terminate us all ? Open letter from Stephen Hawkin note "Because of the great potential of AI, it is important to research how to reap its benefits while avoiding potential pitfalls." Worry is NOT eerie growing consciousness Worry is ability to make high-quality decisions aligned with our values something else - tricky to identify. CAN THEY TURN HOSTILE? Old legend of genie in the lamp; you get what you ask for - not what you want Bigger unease about hypothetical menaces We need useful intelligence. AI is a tool not a threat Massive difficulty of creating perceptive AI Machine learning can: - differentiate categories - fit data etc CAN THEY TURN HOSTILE? Cannot understand human intentions / desires. Any spiteful AI would require those abilities. They do not know that humans exist in any meaningful way It would be thrilling if AI was about to get frightening Parts that may cause real difficulties are not computers but actuators 1 or 2 year old infant AI may be frightening WHERE ARE WE GOING ? Cheap parallel computing - billions of neurons Big Data Massive computing power But what should they be allowed to do? Who to kill on the battlefield? Our brain is analogue AI may need analogue computers NOT digital But - Can our replicas ever surpass reality? WHERE ARE WE GOING ? AI may take over choices… e.g: driving cars, aiming and firing missiles. Life & death decisions / ethical problems If AI makes a decision we regret - change software We can modify principles that govern them Human beings make mistakes - intelligent machines will too WHERE ARE WE GOING ? Difference between intelligence / decision-making Intelligent machines can be useful Stupid machines are scary Algorithms can make appropriate decisions then suddenly make a crazy one AI has no emotion - not subject to natural selection No metric for intelligence - are we improving? Humans use a virtual model Long way from replicating: anger, fear, aggression Even Deep Blue - anyone can pull the plug