Your AI-aware future

#10.1Critiquing AI

You will notice that there is a great deal of public criticism of AI and generative AI, or, conversely, that it is praised to the skies. In this chapter and the previous chapters, we have already discussed various points. Critiquing the development and use of generative AI is ‘easy’: there is always some topic to mention where AI has a negative influence—on learners, people, the environment, inequality in the world, exploitation, increasing polarization, or the rise of fascism. But at the same time, it is also ‘easy’ to see only the positive sides of AI: increased productivity, breakthroughs in conducting research, unprecedented access to information presented in an understandable way, and more.

From an academic and informed perspective, you should at least know both sides. It is precisely an academic skill to see the whole picture and speak and judge about it in a nuanced way. Common limitations or errors in the discussion are based on typical logical fallacies, such as these:

• Incorrect information, where you immediately assume that information you read about generative AI is true, even though you do not check the sources and may not understand it yourself or may confuse it. (We hope this Companion helps with that; see also the example in section 9.4.4).
• Confirmation bias, where you selectively pay attention to information or evidence that supports your existing beliefs and ignore contradictory information.
• Cherry picking, where you deliberately select only those data or examples that support your own position, while ignoring the broader evidence that could complicate or contradict that position.
• Ad hominem reasoning, where arguments are rejected based on the character of the person presenting them rather than on their content.
• False dichotomies, where complex issues are reduced to two opposing options, even though more nuanced positions exist.
• Hasty generalizations, where you draw broad conclusions based on limited examples.
• Appeal to authority, where you accept claims simply because an authority endorses them, without examining the underlying evidence.
• Post hoc reasoning, where you assume that if one event follows another, the first must have caused the second.

These errors occur so often precisely because they often intuitively feel correct and can be socially reinforced within groups that share the same assumptions.

An interesting article in this regard was written by Barend Last and Jack Esselink (2025). They argue for distinguishing five levels of analysis surrounding generative AI. These are five fairly independent characteristics of AI that are often confused or lumped together in discussions, causing the logical fallacies mentioned above to occur frequently and discussions to bog down or become polarized. They distinguish the following characteristics:

• Level 1: The technology itself - how LLMs work and chat environments (in Companion chapters 2, 4, and 6)
• Level 2: How the technology is developed - training and ‘baking’ LLMs (chapters 4 and 5)
• Level 3: How AI is presented - chatbots, anthropomorphism (chapter 5)
• Level 4: How AI is used - what you can use it for and how (chapters 2, 3, 7, 8, and 10)
• Level 5: The societal embedding - impact on learning, education, society, work, the environment, etc. (chapters 1, 2, 6, 8, 9)

Based on that, they examine frequently voiced critical arguments against AI, such as:

• AI does not really think; it is all fake.
• You cannot use AI as a source.
• AI always gives an ‘approximate’ answer, and that is not good enough.
• AI is bad for your brain.
• We live in a ‘post-truth’ era because of AI.
• Everything used to be better.
• AI destroys the university / journalism / democracy / the environment.
• AI discriminates and is biased.

On the LinkedIn page where the article appears, you can read exactly how they view this criticism. But that, too, has been criticized. For example, one objection is that the article assumes people already possess enough knowledge about these levels, whereas that is often not the case. And at one point it uses the rhetorical trick of the so-called straw man. This is a fallacy in which someone distorts, exaggerates, or caricatures an opponent’s argument, and then attacks it instead of the real, original position. For example, the argument that people are afraid of change and are therefore opposed to AI. It is also argued, for example, that there is no longer any question of individual choice in the use of systems (Level 4), once the societal embedding (Level 5) and the commercial design choices (Levels 2 and 3) are in place and you are excluded from an important part of social interactions or news sources that offer other perspectives.

#10.2Scope for action - dealing with AI dilemmas

When you look across all the chapters of this Companion, it can make your head spin. Generative AI offers opportunities, from faster analyses to broader access to knowledge, but the same technology is tied to major problems: environmental impact, labor exploitation in data labeling, concentration of power among a few tech companies, and risks to privacy and democracy (see chapters 6 and 9). That tension can be paralyzing.

This is where the concept of scope for action helps: you focus your attention and energy on what you can actually influence yourself, instead of on what lies beyond your reach. What is within your reach are your own choices: whether you publicly speak out for or against the use of AI, when you do use AI and when you consciously do not, how you draw attention to the data, and how honest you are about your use.

The environment chapter clearly shows how scope for action works in practice, and also how difficult it is. Masley (2025a) uses comparisons to show that the environmental effect of your AI use pales in comparison to choices such as not flying, leaving the car at home, or eating less meat. His advice is therefore to focus your attention on the choices with the greatest impact, and that is a genuine form of scope for action. Ritchie (2024) adds the right attitude to that: look soberly at the data, but stay engaged—urgent and not self-satisfied. At the same time, there is a tension you should not paper over. Maniates (2001) warns that a strong emphasis on individual lifestyle choices can distract attention from structural solutions, even though companies and governments control the biggest levers. Scope for action therefore does not mean choosing between these two, but holding on to both: use the influence you do have over your own choices, and at the same time put the bigger questions on the agenda in the places where they belong, such as the policy of your program, the company where you work, or by advocating for government measures.

This does call for a warning. The emphasis on personal responsibility can also have a downside, because attention to individual behavior can distract from the structural collective solutions and rules that ultimately matter most. Anyone who reduces the environmental debate about AI to "just use fewer prompts" runs the same risk that Maniates (2001) described for the broader environmental debate: when responsibility is individualized, there is less room to think about institutions and the distribution of power, even though that is precisely where the biggest choices are made.

Scope for action therefore works only if you set honest limits to it. Your individual freedom of choice is not unlimited. That choice can become an illusion once AI is so deeply embedded in social interactions, news provision, and work processes that you can hardly avoid it. Scope for action therefore does not mean that everything is your individual responsibility, nor is it an escape into small gestures. It is a way to stay engaged without feeling powerless, by distinguishing between what you can influence yourself, what you can put on the agenda together with others, and what mainly requires policy and regulation. That distinction aligns with the work of Bandura (2006) who shows that people exert influence not only individually, but also through others and through collective, coordinated action. At the individual level, you can develop sustainable habits and use sources responsibly, but you are also part of larger contexts in which your voice carries more weight: as a student, you can help think about the AI policy of your program; as a future professional, you can help determine within your field how AI is used responsibly, and as a citizen, you can participate in the public debate. In this way, the question shifts from "can I change the world?" to the more useful question: "what is my next sensible step here, in my situation, and which problems should I refer to the level where they belong?"

#10.3Ethical self-reflection

The way you use AI says something about who you are and who you want to be. Precisely because AI technology is becoming increasingly intertwined with your studies, work, and daily life, it is important to pause and consider the ethical dimension of your choices.

#10.4Future outlook

AI is continuously developing and will continue to influence your professional field - whether you work in healthcare, education, the judiciary, or the arts. That is why it is important to teach yourself to adapt flexibly and critically to those changes. Not by following everything blindly, but by staying curious, thinking ethically, addressing issues higher up the hierarchy when necessary, and continuing to learn.

A personal roadmap helps you stay on course. Think about where you want to go, what you need to get there, and be prepared for change. AI literacy is not a destination, but an ongoing process that starts with your own choices.

#10.5.1Review questions

What are, according to the chapter, the four components that together form the basis of AI literacy?

What is the 'Best Available Human' concept as Mollick describes it?

Why does the chapter distinguish between learning and being productive with AI systems?