V.A. Shakhnov1, A.V. Proletarsky2, L.A. Zinchenko3, V.V. Kazakov4, V.V. Terekhov5
1–5 Bauman Moscow State Technical University (Moscow, Russia)
1 shakhnov@mail.ru, 2 pav_mipk@mail.ru, 3 lzinchenko@bmstu.ru,
4 samharm777@gmail.com, 5 vterekhov.bmstu@gmail.com
Artificial intelligence, including its machine learning subsection, is increasingly being used in various applications. The paper considers various approaches to machine learning, as well as some possible risks of its application from the point of view of ethics. A review of the latest systems has been carried out. Machine learning and artificial intelligence systems such as text systems ruGPT-3, FRED-T5, InstructGPT, ChatGPT (OpenAI), AI visual systems: ruCLIP, ruDALL-E, Kandinsky 2.1 from Sber, Shedevrum from Yandex and multi modal systems NeONKA (GigaChat), GPT-4 are described. There are also such basic algorithms as XGBoost, CatBoost, transformer architectures such as BERT, and recurrent architectures such as LSTM and GAN networks. The mechanism of Shapley values is described for analyzing machine learning and artificial intelligence systems to introduce greater transparency into the processes taking place in them and the principle of decision-making by such systems. The paper provides an overview of the ethical principles highlighted by the community in the field of artificial intelligence. The developed CognShield software, its tasks and the main structural elements of the user interface are described. An example of software analysis for compliance with the Code of Ethics in the field of Artificial Intelligence is given. In this paper, it is proposed to supplement the list of ethics criteria with the principle of adaptability, implying such behavior of the machine learning system, in which the system, depending on the “degree of confidence of the system in the forecast” (for example, deviations from a given probability threshold in the case of classification), the importance and degree of influence (determined based on the tasks of the machine learning system being developed) of the forecasting by various methods focuses the user on aspects of the interpretation of the result or recommendations for its application.
Shakhnov V.A., Proletarsky A.V., Zinchenko L.A., Kazakov V.V., Terekhov V.V. On the ethics of artificial intelligence. Neuro-computers. 2024. V. 26. № 5. Р. 17-29. DOI: https://doi.org/10.18127/j19998554-202405-02 (In Russian)
- Bush V. As We May Think. [Electronic resource] – Access mode: http://web.mit.edu/STS.035/www/PDFs/think.pdf, date of reference 25.10.2022.
- Anderson M., Anderson S.L. Machine Ethics. Cambridge: Cambridge University Press. 2011. DOI 10.1017/CBO9780511978036.
- Belousov P.A., Marukhina O.V., Skomorokhov A.O., Starkov S.O., Mirzeabasov O.A., Raspopov D.A. Machine learning and big data: textbook. stipend. St. Petersburg: GUAP. 2021. 119 p. ISBN 978-5-8088-1655-8. (In Russian)
- Rosstandart Order № 1732 dated 25.07.2019 On the establishment of the Technical Committee for Standardization "Artificial Intelligence". [Electronic resource] – Access mode: https://law.tks.ru/document/702234, date of reference 23.10.2022. (In Russian)
- XGBoost Documentation. [Electronic resource] – Access mode: https://xgboost.readthedocs.io/en/latest/, date of reference 25.10.2022.
- CatBoost – high-performance open source library for gradient boosting on decision trees. [Electronic resource] – Access mode: https://catboost.ai/, date of reference 25.10.2022.
- Devlin J., Chang M.-W., Lee K., Toutanova K. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. Proceedings of Conference of the North American Chapter of the Association for Computational Linguistics. 2019. V. 1. P. 4171–4186. DOI 10.18653/v1/N19-1423.
- Sinha K., Jia R., Hupkes D., Pineau J., Williams A., Kiela D. Masked Language Modeling and the Distributional Hypothesis: Order Word Matters Pre-training for Little. Proceedings of the Conference on Empirical Methods in Natural Language Processing. 2021. P. 2888–2913. DOI 10.18653/v1/2021.emnlp-main.230.
- Jabeur S.B., Mefteh-Wali S., Viviani J.L. Forecasting gold price with the XGBoost algorithm and SHAP interaction values. Annals of Operations Research. 2021. DOI 10.1007/s10479-021-04187-w.
- Understanding LSTM Networks. [Electronic resource] – Access mode: https://colah.github.io/posts/2015-08-Understanding-LSTMs/, date of reference 25.10.2022.
- Karatzoglou A., Jablonski A., Beigl M. A Seq2Seq learning approach for modeling semantic trajectories and predicting the next location. Proceedings of the 26th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 2018. P. 528–531.
- Declaration of montr´eal for a responsible development of AI. [Electronic resource] – Access mode: https://www.montrealdeclaration-responsibleai.com, date of reference 25.10.2022.
- The Institute for Ethical AI & Machine Learning. [Electronic resource] – Access mode: https://ethical.institute/index.html, date of reference 25.10.2022.
- Ienca M., Vayena E. AI Ethics Guidelines: European and Global Perspectives. [Electronic resource] – Access mode: https://rm.coe.int/cahai-2020-07-fin-en-report-ienca-vayena/16809eccac, date of reference 25.10.2022.
- Kashapova E.R., Ryzhkova M.V. Cognitive distortions and their influence on individual behavior. Bulletin of Tomsk State University. Economy. 2015. № 2(30). P. 15–26. DOI 10.17223/19988648/30/2. (In Russian)
- Russian DALL-E. [Electronic resource] – Access mode: https://rudalle.ru/, date of reference 25.10.2022.
- Code of Ethics in the field of artificial intelligence. [Electronic resource] – Access mode: https://a-ai.ru/wp-content/uploads/2021/10/ Кодекс_этики_в_сфере_ИИ_финальный.pdf, date of reference 25.10.2022. (In Russian)
- Zinchenko L.A., Kazakov V.V., Mironov A.A. Hierarchical descriptor for displaying information about multilayer radiation-resistant screens of electronic equipment protection. Automation. Modern technologies. 2021. V. 75. № 4. P. 171–176. (In Russian)