Neural networks
We cannot know the strategy or rules it develops, whichmakes the use of neural networks in life-critical situations a moral dilemma. This also means we don’t really know why some models work and others don’t.
There have been attempts to try and understand what goes on in a deep neural network by organizations such as Google Deep-Mind, whose Deep-Dream experiments used a convolutional neural network to generate psychedelic images that copied real photos (Alexander Mordvintsev, 2015). We can observe how the convolutional neural network (CNN) creates objects in the image which are from the database it was trained on. The CNN identifies patterns in the image and then accentuates the patterns throughout. This gives us an idea of the way in which CNNs actually process images. We can see below how after multiple iterations of the algorithm an image of jellyfish turns into what looks like a herd of animals. After enough iterations the original image will be completely lost, leaving a psychedelic image from inside the neural network.
Conclusion
Despite the black box nature of neural networks, advancements in neural networks will likely change the future of computing and of society at large. By better emulating the structure of the human brain we will be able to use artificial intelligence to aid the design and engineering of computer systems. Ethical problems aside, the productivity and cost-effectiveness of neural network-based AI systems will revolutionize many industries in the near future. Future advancements in computing power can only help speed up this process.
Bibliography
Apple's credit card is being investigated for discriminating against women. The Verge, 2019.[Online] Available at: https://www.theverge.com/2019/11/11/20958953/apple-credit-card-genderdiscrimination- algorithms-black-box-investigation [Accessed 18 August 2020] Architectures, E. S.-B. B. f. T. D. N. N., 2019. Chankyu Lee, Syed Shakib Sarwar. arXiv. Joy Buolamwini, T. G., 2018. Gender Shades: Intersectional Accuracy Disparities in Commercial Judith E. Dayhoff, J. M. D., 1999. Artificial Neural Networks: Opening the Black Box. Arlington, Virginia, Conference on Prognostic Factors and Staging in Cancer Management: Contributions of Artificial Neural Networks and Other Statistical Methods. Gender Classification. Proceedings of Machine Learning Research, Volume 81, pp. 1-15. Khosravani, M. R., 2012. Application of Neural Network on Flight Control. International Journal of Machine Learning and Computing 2 (6). Maass, W., 2016. Energy-efficient neural network chips approach human recognition capabilities. Proceedings of the National Academy of Sciences of the USA 113(41), pp. 11387-11389. Martin Abadi, D. G. A., 2016. Learning to Protect Communications with Adversarial Neural Cryptography. Google Brain.
195
Made with FlippingBook Digital Publishing Software