How companies and startups can develop AI products - recommendations from the Ministry of Digital Transformation

The Ministry of Digital Transformation, together with experts, has developed guidelines for the responsible development of technological systems with artificial intelligence. The guidelines will help companies and start-ups implement AI in their products and services without violating human rights, copyright and private property.

"There is a growing demand for responsible AI in the global market. Users of innovative products expect companies to provide breakthrough AI solutions that are safe for them at the same time. These recommendations are a practical guide for Ukrainian developers on how to maintain the degree of innovation in products and at the same time design solutions that respect human rights," said Oleksandr Borniakov, Deputy Minister of Digital Transformation for IT Development.

The guidelines are based on the best international standards, including the Council of Europe's HUDERIA methodology, which helps to assess the human rights impact of AI systems.

The recommendations in the document can be applied to the development of technological solutions in various fields, including finance, medicine, education, employment, marketing, etc.

What you will learn

The document provides practical tips on how companies can choose a large language model and adapt it to their needs through training and fine-tuning. In particular, it is recommended to pay attention not only to the number of parameters or the popularity of the model, but also to the quality of training data, transparency, adaptability, and vendor reputation.

The guidelines explain what data should be used to train a model. It also explains how to process them before training so that the model works without hallucinations, discrimination, and biases.

In addition, experts have compiled a list of key human rights and examples of violations of these rights in AI products. They cover cases in medicine, employment, education, finance, etc. These examples will help companies avoid common mistakes in product development.

A separate section provides tips on how to avoid copyright and property rights infringement, especially when using generative AI.

The appendix contains more than 30 tools for developing AI solutions, detecting code errors, auditing bias and fairness in systems, etc.

Here are the key tips:

• Only use data sets to which you have legal rights to train your model, and check that the data is balanced, complete, anonymised, and normalised.

• Involve users in the creation and improvement of the product. Use an iterative approach where you regularly test the model, get feedback, and adapt the product. This helps to make the system fairer, more useful, and more accurate - especially in sensitive areas.

• Ensure professional human oversight, especially in decisions about access to services, loans, employment or education, and in the medical field.

• Be mindful of copyright. Always check the terms of use of models and services, and whether the generated content does not contain other people's copyrighted material.

• Design privacy by default and design: minimise data collection, provide encryption, access control, logging, and the right to delete data. Users should be in control of their information.

Read the full list of recommendations here.

The recommendations were developed by the Ministry of Digital Transformation together with the Expert Advisory Committee on AI Development under the Ministry of Digital Transformation, IT Ukraine Association, Ukrainian Bar Association, CEDEM, the Department of Intellectual Property and Information Law of the Educational and Research Institute of Law of the Taras Shevchenko National University of Kyiv and the Research Institute of Intellectual Property of the National Academy of Legal Sciences of Ukraine.

The document was developed with the assistance of the UK Digitalisation for Growth, Integrity and Transparency (UK DIGIT) project, implemented by the Eurasia Foundation and funded by UK Dev.