02.11.2018 When and Where will a Breakthrough Come? • Technology breakthroughs often happens randomly and not linked to major initiatives INF3490/INF4490 Biologically inspired and projects. computing • AI breakthroughs depend on the invention of scalable learning methods. Future Perspectives on Artificial Intelligence – Need to understand more about how scalability and complexity arises in nature. – What to expect and should we worry? • Contributors to progress in AI: Jim Tørresen, October 31, 2018 – Researchers mimicking biological or medical phenomena. – Researchers solving engineering problems 2 What Methods are Best? CAM (Cellular Automata Machine) Brain 3 4 Autonomous Systems How Similar to Humans? • Systems that can operate with no or limited • Similarity depends on progress in a number human control of fields such as AI methods, computing • Degrees of human control: power, vision systems, speech recognition, – Full autonomy (human decides only the goal) speech synthesis, human-computer – Remotely monitored (autonomous planning but interaction, mechanics and actuators. human take over if the system is misbehaving) • Design and usability will be essential for – Remotely operated (autonomous steering using future robots but they should not necessarily human planning) be as similar as possible to humans (ref. – Remotely controlled (human take all decisions) u ncanny valley) . • We will see robots having human-like behavior , but not human consciousness . 5 6 1
02.11.2018 In what ways can AI be helpful/good Is Terminator Coming Close? and what are the potential challenges? Input collected from students attending the class: • helpful/good: – Eliminate repetitive tasks for human – Solve diff problems – Better NLP • potential challenges: – Legal consequences – Personal data sharing – Eliminating jobs – Aligning goals with humans 7 8 Ethical Risks of Developing AI Systems • People may become unemployed because of automation.? • We get too much free time.? • Artificial intelligence can be used for destructive and unwanted tasks.? (ref. “1984” George Orwell) • Successful KI can lead to the extinction of “Humans, limited by mankind? slow biological I am in the camp that evolution, couldn’t AI is our “biggest is concerned about compete and would be existential threat” super intelligence. 9 10 superseded by A.I.” Future Scenario with Autonomous Ethics for Artificial Intelligence Interacting AI Systems • Designers must be aware of possible ethical challenges that should be considered – eg. protect sensor data, avoid misuse, and allow human inspection of functionality • The systems should themselves be able to do ethical decision making to reduce the risk of unwanted behaviour • Optional paper: 11 12 https://www.frontiersin.org/articles/10.3389/frobt.2017.00075/full 2
02.11.2018 Self learning/Machine learning Ethics for Programmers • In the book “Moral Machines” it is argued that ethic Algorithm competence should be included during program development by e.g. Black box – machine learning methods based on examples of ethical and unethical behavior System to be • Software that will replace human evaluation and designed vs social function should adhere to criteria such as – accountability (no. ansvarlighet) Glass – inspectability box – manipulation robustness Data set/ Somewhat transparent – predictability (no. forutsigbarhet) specification and understandable behaviour 13 What would you require of a robot to be used in your home? 15 16 Ethical Recommendations for Ethical Guidelines for Robots and Commercial Robots (Euronet Roboethics Atelier) their Developers (Asimov 1942) • Safety: There must be mechanisms (or opportunities 1. A robot may not harm a human being , for an operator) to control and limit a robot's autonomy. or through inaction, allow a human to be • Security: There must be a password or other keys to injured. avoid inappropriate and illegal use of a robot. 2. A robot must obey orders given by • Traceability: Similarly as aircraft, robots should human beings except where such have a "black box" to record and document their own orders would conflict with the first law. behavior. 3. A robot must protect its own • Identifiability: Robots should have serial numbers and registration number similar cars. existence as long as such protection does not conflict with the first or second • Privacy policy: Software and hardware should be used to encrypt and password protect sensitive data law. 17 18 that the robot needs to save. 3
02.11.2018 This lecture is based on a translation of the last chapter in an introductory book to artificial intelligence Innhold: • Kunstig intelligens og intelligente systemer • Problemløsning med kunstig intelligens • Evolusjon, utvikling og læring • Sansing og oppfatning • Bevegelse og robotikk • Hvor intelligente kan og bør maskiner bli? 19 4
Recommend
More recommend
