Artificial Intelligence And Human Accountability
Artificial Intelligence andHuman AccountabilityJoanna J. BrysonUniversity of Bath, United Kingdom@j2bryson
Intelligence is doing the right thing at the right time. Definitions A form of computation (not math)–transforms for communicatingperception into action. Requires time, space, andenergy. Agents are any vector of change, e.g. chemical agents. Moral agents are considered responsible for theiractions by a society. Moral patients are considered the responsibility of asociety’s agents. Ethics is the set of behaviours that creates andsustains a society, including by defining its identity. Artificial intelligence is an artefact, built intentionally.right now
Romanes, 1883 – AnimalIntelligence, a seminal monographin comparative psychology.
Intelligence is doing the right thing at the right time. A form of computation (not math)–transformsDefinitionsfor communicatingright nowsensing into action. Requires time, space, and energy. Agents are any vector of change, e.g. chemicalagents. Moral agents are considered responsible for their}actions by a society. Moral patients are considered the responsibility ofMoralsubjectsa society’s agents. Ethics is the set of behaviours that creates andsustains a society, including by defining its identity. Artificial intelligence is an artefact; builtintentionally.}Including but not limitedto general principles} Intent responsibility.
AI is built by humans, with orwithout machine learning.We’re responsible. How AI is built, how it is trained (and onwhat), how it is tested, monitored–all thingsfor which humans can be held to account. Every aspect of developing and operating AIcan be logged; we can demand evidence ofdue diligence.
AI Trained on Human LanguageReplicates Our Implicit Biases, RealityCaliskan, Bryson &Narayanan(Science, April2017)Our implicitbehaviour isnot our ideal.Ideals are forexplicitplanning andcooperation.2015 US labor statisticsρ 0.90
At Least Three Sources of AI Bias Implicit: Absorbed automatically using machinelearning on data from ordinary culture. Accidental: Introduced through ignorance byinsufficiently diverse or careful development teams. Deliberate: Introduced intentionally as a part of thedevelopment process (planning or implementation.)
How to deal with them Implicit–compensate with design, architecture (seealso accidental). Accidental–diversify work force, test, log, iterate,improve. Deliberate–audits, regulation.
Caio Machado and Marco Konopacki
Only Humans Can Be Accountable Law and Justice are more about dissuasion than recompense. Safe, secure, accountable software systems are modular –suffering* in such is incoherent. *e.g. systemic dysphoria ofisolation, loss of status or wealth. No penalty of law enacted directly against an artefact (including ashell company) can have efficacy.Bryson, Diamantis & Grant(AI & Law, September 2017)
People want to make AI theycan be friends with, fall inlove with, will their fortunesto – “equals” over which wehave complete dominion.This is (arguably) both sickand dangerous.Bryson & Kime 1998; 2011Kathleen Richardson 2016
1.119PolarizationIndex1r .67Is all thisbecause of AI?Income share of top 1%Voorheis, McCarty & 1929192119137911131519C inequality was perhapsdriven by then-new distancereducing technologies: news, oil, rail,telegraph; now bootstrapped by ICT? Great coupling – period of lowinequality where wages trackproductivity – probably due to policy.Implies we could fix it now too.Figure 1.2: Top One Percent Income Share and House Polarization Technological innovation maymandate regulatory innovation.9Polarization lagged 12 years r .9119 LatePercentageShare17Economically,we’ve beenhere before.Polarization and the Top 1%Polarization indexState Income Inequality and Political Polarization
Aylin Caliskan Arvind Narayanan@random walker@aylin cimThanks to my collaborators, andto you for your attention.Mihailis E.DiamantisRob Wortham@RobWorthamHolly Wilson@wilsh010 Andreas Theodorou@recklessCodingArtificial Models of Natural Intelligence,(AmonI, 2019) University of BathNolan McCarty@nolan mcAlexStewart@al cibiadesTom Dale GrantAmonI, 2016
Artificial intelligence is an artefact, built intentionally. Definitions for communicating right now. Romanes, 1883 – Animal Intelligence, a seminal monograph in comparative psychology. Intelligence is doing the right thing at the right time. A form of computation (not math)–transforms sensing into action. Requires time, space, and energy. Agents are any vector of change, e.g .
Arti cial Neural Networks Human and Arti cial Neurons Arti cial Neurons ANN Artchitecture A simple neuron Arti cial neuron is a device with many inputs and one output Two modes: Training Using Firing Ruledetermines when a neuron should re. Are very important in neural networks and accounts for their high exibility
CHAPTER 1. INTRODUCTION Arti cial neural networks (https://www.merriam-webster.com) de nes as arti cial intelligence: 1. A branch of computer science dealing with the simulation of intelligent behavior in computers;
Pour l’année 2018, la Conférence Nationale d’Intelligence Arti cielle et les Rencontres des Jeunes Chercheurs en Intelligence Arti cielle sont organisées conjointement. Elles ont lieu du 4 au 6 juillet 2018 à Nancy sur laPlate-Forme Intelligence Arti cielle (PFIA). Lors de la même semaine, la plate-
2. Chapter 3 reviews the state of the art of Arti cial Intelligence applications in wind energy. The prospects and challenges of applying Arti cial Intelligence in the wind energy sector are discussed in Chapter 4. Finally, some conclusions are given in Chapter 5. The funding from DTU Wind Energy is greatly appreciated. I also want to thank my .
Cho and Kasa, fiValidation.fl J. Bullard Discussion Parameters versus models Connections to escape dynamics Model validation Speci–cation testing Assignment of the PLM Nature of the validation dynamics Instability generated by rival models Arti–cial intelligence Comparison with arti–cial intelligence Statistical versus economic .
CS 188 Spring 2015 Introduction to Arti cial Intelligence Midterm 1 You have approximately 2 hours and 50 min
CS 188 Spring 2014 Introduction to Arti cial Intelligence Midterm 1 You have approximately 2 hours and 50 minutes. The exam is closed book, close
brain in processing data and creating patterns for use in decision making. Deep learning belongs to the eld of arti cial intelligence. The term \Arti cial Intelligence" was rst used at a workshop held on the campus of Darmouth College during the summer of 1956. The term \deep learning", however, would come much later, used for the rst time by
