• Aucun résultat trouvé

Good AI and Law

N/A
N/A
Info
Télécharger
Protected

Academic year: 2022

Partager "Good AI and Law"

Copied!
1
0
0

Chargement.... (Voir le texte intégral maintenant)

Texte intégral

(1)

Good AI & Law

Bart VERHEIJ

Bernoulli Institute of Mathematics, Computer Science and Artificial Intelligence, University of Groningen

AI's successes are these days so prominent that—if we believe reports in the news—the times seem near that machines perform better at any human task than humans themselves. At the same time the prominent AI technique of neural networks—today typically called deep learning—is often considered to lead to black box results, hindering transparency, explainability and responsibility, values that are central in the domain of law. So in that specific sense, the distance between neural network AI and the needs of the law is vast. In this talk, it is claimed that for good AI & Law we need an AI that can provide good answers to our questions, has good reasons for them and makes good choices. It is argued that the path towards good AI & Law requires the integration of data-driven and knowledge-based AI, and that argumentation as it occurs in the law can show the way to such integration.

Reference

Verheij, B. (2018). Arguments for Good Artificial Intelligence. Groningen: University of Groningen. www.ai.rug.nl/~verheij/oratie.

Références

Télécharger maintenant ( PDF - 1 Page - 117.41 KB )

Documents relatifs

Are Deep Neural Networks good for blind image watermarking?

Are Deep Neural Networks good for blind image watermarking?.. Vedran Vukotić, Vivien Chappelier,

Random Neural Networks and Deep Learning for Attack Detection at the Edge

We apply the random neural networks (RNN) [15], [16] developed for deep learning recently [17]–[19] to detecting network attacks using network metrics extract- ed from the

The Black Box

The Black Box Romain Michon CCRMA Department of Music Stanford University, CA 94305-8180, USA rmichon@ccrma.stanford.edu Myles Borins CCRMA Department of Music Stanford University,

A combinatorial theory of random matrices III: random walks on $\mathfrak{S}(N)$, ramified coverings and the $\mathfrak{S}(\infty)$ Yang-Mills measure

In the article [8], the author and his co-authors show how to prove asymptotics of Yang-Mills measures driven by L´evy processes on the unitary and orthogonal groups without using

Directional features of the downshifted peak observed in HF-induced stimulated electromagnetic emission spectra obtained using an interferometer

Note also the approximately 1.5 ◦ eastward offset of the weighted DP source region position for pump pointing near magnetic field-aligned (14 ◦ south) as compared to the other

Using Formal Concept Analysis to Explain Black Box Deep Learning Classification Models (short paper)

We propose a general concept lattice theory based framework for explanation generation, where given an outcome O of a deep learning model and a domain ontology, the goal is to

On the Susceptibility of Deep Neural Networks to Natural Perturbations

In this paper, we propose natural attacks using visibly foggy images to generate input that causes incorrect classi- fication by the Inception deep learning model [22].. Apart from

Landcover Classification Using Texture-encoded Convolutional Neural Networks: Peeking Inside the Black Box

Our approach to probing the performance of the models in the classification of the various landcover categories surprisingly reveals that the texture-based model