Causal and Non-Causal Explanations of ArtificialIntelligence

Deep neural networks (DNNs), a particularly effective type of artificial intelligence, currently lack a scientific explanation. The philosophy of science is uniquely equipped to handle this problem. Computer science has attempted, unsuccessfully, to explain DNNs. I review these contributions, then identify shortcomings in their approaches. The complexity of DNNs prohibits the articulation of relevant causal relationships between their parts, and as a result causal explanations fail. I show that many non-causal accounts, though more promising, also fail to explain AI. This highlights a problem with existing accounts of scientific explanation rather than with AI or DNNs.

[1]  Paul Goldsmith-Pinkham,et al.  Predictably Unequal? The Effects of Machine Learning on Credit Markets , 2017, The Journal of Finance.

[2]  Searching for Non-Causal Explanations in a Sea of Causes , 2018, Oxford Scholarship Online.

[3]  Margaret Morrison,et al.  Reconstructing Reality: Models, Mathematics, and Simulations , 2015 .

[4]  Cameron Buckner.  Deep learning: A philosophical introduction , 2019, Philosophy Compass.

[5]  Noah A. Smith,et al.  Is Attention Interpretable? , 2019, ACL.

[6]  Cynthia L. Bennett,et al.  What is the point of fairness? , 2019, ACM SIGACCESS Access. Comput..

[7]  Angela Potochnik,et al.  Idealization and the Aims of Science , 2017 .

[8]  Byron C. Wallace,et al.  Attention is not Explanation , 2019, NAACL.

[9]  Mark O. Riedl,et al.  Rationalization: A Neural Machine Translation Approach to Generating Natural Language Explanations , 2017, AIES.

[10]  Cynthia Rudin,et al.  Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead , 2018, Nature Machine Intelligence.

[11]  Chris Russell,et al.  Counterfactual Explanations Without Opening the Black Box: Automated Decisions and the GDPR , 2017, ArXiv.

[12]  Robert W. Batterman,et al.  Minimal Model Explanations , 2014, Philosophy of Science.

[13]  Aaron Halfaker,et al.  Who Did What: Editor Role Identification in Wikipedia , 2021, ICWSM.

[14]  Vasant Honavar,et al.  Algorithmic Bias in Recidivism Prediction: A Causal Perspective , 2019, AAAI.

[15]  Yoshua Bengio,et al.  Neural Machine Translation by Jointly Learning to Align and Translate , 2014, ICLR.

[16]  Kelvin J. McQueen,et al.  Two Flavours of Mathematical Explanation , 2018, Oxford Scholarship Online.

[17]  Keith A. Markus,et al.  Making Things Happen: A Theory of Causal Explanation , 2007 .

[18]  Collin Rice,et al.  Idealized models, holistic distortions, and universality , 2017, Synthese.

[19]  Jette Henderson,et al.  CERTIFAI: A Common Framework to Provide Explanations and Analyse the Fairness and Robustness of Black-box Models , 2020, AIES.