Ethics in the Age of AI: In Conversation With ITMO Expert Daria Chirva

What is AI ethics all about?

Ethics in the field of AI concerns all stages of the technology’s life cycle. These include development, design, testing, implementation, and use. In other words, certain ethical questions, ambiguities, and problems can arise at all of these stages.

How does ITMO tackle the subject of AI ethics?

First, we have several relevant courses. For example, within the course on AI ethics, we focus on the ethical use of AI-based services and show what kinds of issues and challenges can arise as a consequence of poor-quality development. For AI, not only the architecture design matters, but also the quality of the data. Data that initially contains unintended bias can lead the model to produce recommendations that discriminate against certain social groups defined by age, gender, or region of residence. That is why we offer courses that are aimed more at developers than at users. We work closely with Master’s programs delivered by the ITMO’s Institute of Applied Computer Science.

We are also working on a course for Bachelor’s students called AI Philosophy that will cover ethical issues related to the technology.

We have also developed modules on AI ethics for the role-based competency model for training of AI specialists. Generally, we actively collaborate with the Research Center “Strong AI in Industry,” where we offer consultations on an as-needed basis.

What other aspects apart from ethical development are covered in the courses?

In class, we discuss the key ethical principles and values in the field of AI and analyze the existing writing on the subject. We also explore many specific examples of ethically controversial applications of the technology. The main idea of our courses is to equip our students with a habit of integrating ethical questions into their work; we are talking about future developers, team leads, and those planning to market their own products.

In the course, students analyze existing services – chat bots, co-pilot services, mental health bots – and evaluate them from an ethics point of view. Then, we ask them to also come up with regulatory measures that could minimize the risks they have identified in a service.

These days, students are used to relying on AI tools. From your experience of teaching the course, would you say that students understand the importance of the ethical principles it discusses? What questions do students find most relevant?

As a rule, students find the course interesting because they get to look at technology from a new point of view, through the lens of social sciences and humanities. Sometimes, they join the course thinking that they are responsible solely for development, while ethical issues should be left to managers and team leads. However, the past four years of working in the field show us that more and more people are becomingaware of the notion that AI systems have a significant impact on individuals, on social groups, and on society as a whole. And we need to understand the value framework within which we are operating: what we consider acceptable and unacceptable, desirable and undesirable in terms of the development, implementation, and use of AI.

In terms of specific questions, students are highly interested in identifying the responsible party when an artificial agent is added to a process (for example, in traffic incidents with self-driving cars). Students are also worried about authorship attribution: how do we determine who originated the idea when we create something with AI? Another hot topic is education in the age of AI. We discuss how students and lecturers can structure the educational system in such a way that preserves the core of the educational process, while at the same time using the benefits offered by AI.

I would like to add that our classes include a great number of live discussions of various points of view. The course also includes written assignments that involve using generative AI services to support students. But, of course, this should not mean directly copying output from a generative service; rather, it should involve working through that output in light of the live discussions in the classroom.

Do you think the audience of AI ethics courses should be extended? Maybe with a professional development course or a course for school students?

The team of Prof. Boukhanovsky, the head of ITMO’s Research Center “Strong AI in Industry,” already offers professional development courses that include our modules on AI ethics. I believe such programs are truly needed now. A new element of professionalism should be an understanding of the capabilities and limitations of AI technologies, from the perspective of not only technical possibilities, but also the consequences of using AI and the risks and benefits for society that inevitably accompany this technology.

And as for school students, I would say that this topic should be a part of a more general course on the ethical aspects of education and science.

What tips on the ethical use of AI would you give to our readers?

1. Use AI as a partner or assistant, but not as a substitute for your own thinking. Build your own competence instead of depending on technology;
2. Always verify AI replies, especially if important decisions depend on them. Verification is the basis of digital literacy;
3. Never share personal data or confidential information with AI;
4. Be honest about the way you use AI. Transparency is an important condition of trust;
5. Remember: responsibility is shared between people. AI is not morally responsible for the consequences of its applications.

Let’s consider a couple of cases of interacting with AI. The first case is evolutionary biologist Richard Dawkins’ recent claim that the AI model Claude is “conscious.” We know that many users also often see chat bots as “living” interlocutors. How can we avoid this illusion when interacting with models we use every day?

This is actually a big issue. We are prone to ascribing the properties of something alive, even human, to things that behave similarly to us. For instance, we attribute human traits to animals. A similar thing happens with chat bots that demonstrate mastery of language and offer semantically coherent replies. The danger here is in discounting the artificial nature of AI and emotionally engaging in this conversation. Here, there is no equal communication that happens between two people; AI is only imitating it. 

Currently, the possible solution to this problem as an ethical one is a demand for developers to introduce clear reminders about the artificial nature of the system into the interface of their products. A repeated reminder will pull users out of their involuntary emotional connection to an artificial service. However, at the moment, the responsibility for that lies fully on users.

As for Dawkins and whether or not modern AI is conscious, that’s a complex philosophical and neuroscientific question. By and large, we don’t have a straightforward, clear definition of consciousness and what criteria are enough to consider something conscious. So, when we look beyond humans and higher mammals, we lose the ability to give a clear answer to this question. In this sense, Dawkins’ statement rather opens a new line of research than gives a scientifically proven answer.

The next case is related to education. These days, we often hear the opinion that we don’t have to study anymore, that AI knows all the answers. How should the educational process transform in these new conditions?

AI is a definite threat to the translational education model, where the lecturer is the one who shares knowledge with their audience. It’s true that modern AI systems are capable of explaining complex topics on simple examples that students can relate to.

However, first of all, generative AI can make mistakes (and reminders of that are usually present in its interface). We still need an expert to avoid mistakes in information transfer. Second, contemporary education needs to center on building metaskills: reflection, critical thinking, and the ability to ask questions. Humans remain the ones that are curious and driven to change the world. For that, they need to ask questions and identify problems that haven’t been solved. AI is great at providing answers, but it can’t see new questions and turn them into scientific curiosity, research, or development.

It’s also important to understand that education has several functions and knowledge transfer and production are just two of them. Another important one is the communicative function – interacting with others in the educational process, sharing experience and life strategies. This social and communicative function remains crucial in the human-to-human educational system.

The third and final case deals with existing fears about the safety of AI development. For instance, organizations like Machine Intelligence Research Institute actively campaign for shutting down AI development to prevent a catastrophe. Is this an issue for AI ethics? How should we treat such “doomsday scenarios”?

This question is at the intersection of AI ethics and information security. Fear and a somewhat negative reaction is actually humanity’s typical reaction to a new technology, be that written language or the internet. When something changes, it causes anxiety. And we ourselves will have to change, too. In the case of written language, for instance, human brains changed when we learned to write things down.

Importantly, it’s these upcoming changes that are the reason we should treat doomsday scenarios critically. As a rule, they are built on extending existing tendencies and don’t account for the way we are going to change in the process.

However, I don’t want to sound techno-optimistic. I believe that we need to take technology seriously and monitor how it changes us; that way, we won’t lose the skills that are crucial for us as humans and humanity.