Work

Causal and Non-Causal Explanations of ArtificialIntelligence

Year: 2020
Type: conference-paper
Abstract: 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.
Language: English
Open Access status: green
OpenAlex ID: W3043302035
FWCI: 1.984
Cites: 15
Cited by: 2
Related to: 10