IBM’s Francesca Rossi on AI Ethics: Insights for Engineers

As a computer scientist who was immersed in the ethics of artificial intelligence about a decade ago, I saw directly how the field developed. Today, an increasing number of engineers themselves develop artificial intelligence solutions while moving into complex moral considerations. Besides technical experience, the deployment of artificial intelligence requires a precise understanding of the moral effects.

In my league as an IBM AI Ethics global leader, I noticed a major shift in how artificial intelligence engineers work. Not only no longer speak to other artificial intelligence engineers on how to build technology. Now they need to deal with those who understand how their creativity will affect societies that use these services. Several years ago in IBM, we realized that artificial intelligence engineers need to integrate additional steps into their development, technical and administrative development process. We have created the Playbook book that provides the appropriate tools to test problems such as bias and privacy. But understanding how to use these tools properly is very important. For example, there are many different definitions of fairness in artificial intelligence. Determining the definition that applies to consultation with the affected community, customers and ultimate users requires.

In its role in IBM, Francesca Rossi Cochshies, the company’s artificial intelligence ethics council, assists in defining its basic principles and internal processes. Francesca Rossi

Education plays a vital role in this process. When trying the AI ​​Ethics game book with artificial intelligence engineering teams, one team believed that their project was free of concerns about prejudice because it did not include protected variables such as race or sex. They did not realize that other features, such as the postal code, could serve as agents associated with protected variables. Engineers sometimes believe that technological problems can be solved through technological solutions. Although software tools are useful, they are just a start. The biggest challenge lies in learning to communicate and cooperate effectively with the diverse stakeholders.

The pressure may be created for the release of new AI products and tools quickly tension with a comprehensive moral evaluation. For this reason, we have created the governance of central artificial intelligence ethics through the IBM Ethics Council. Often, individual project teams face the final dates and quarterly results, making it difficult for them to consider fully in broader effects on reputation or customer confidence. Principles and internal operations should be central. Our customers – other companies – request solutions that do not respect certain values. In addition, regulations in some areas now state ethical considerations. Even major artificial intelligence conferences require papers to discuss the ethical effects of research, and pushed artificial intelligence researchers to consider the impact of their work.

In IBM, we started developing tools that focus on major issues such as privacy, ability to clarify, fairness and transparency. For every source of concern, we created an open source tool group with instructions and educational lessons from programming instructions to help engineers to implement them effectively. But with the development of technology, as well as ethical challenges. With gym, for example, we face new concerns about creating potential or violent content, as well as hallucinations. As part of the IBM family models, we have developed protection models that evaluate both entry and output claims for issues such as realism and harmful content. These typical capabilities serve both of our internal needs and meet the needs of our customers.

Although software tools are useful, they are just a start. The biggest challenge lies in learning to communicate and cooperate effectively.

The company’s governance structures must remain graceful enough to adapt to technological development. We constantly evaluate how new developments such as artificial intelligence and AICEC AI can inflate or reduce certain risks. When making models as an open source, we assess whether this provides new risks and what are the required guarantees.

For artificial intelligence solutions that raise ethical red flags, we have an internal review process that may lead to adjustments. Our evaluation extends beyond technology characteristics (fairness, explanation, and privacy) to how it is published. Publishing can respect or undermine human dignity and undermine it. We make risk assessments for each technique use, with the realization that understanding risk requires knowledge of the context in which technology will work. This approach is in line with the framework of the European Artificial Intelligence law – it is not that obstetric artificial intelligence or automatic learning is risky, but some scenarios may be high or low. Disturbed use cases require additional scrutiny.

In this sophisticated scene quickly, artificial intelligence engineering requires continuous vigilance, ability to adapt, and adhere to moral principles that put human luxury in the center of technological innovation.