Are you AI literate?
How high is your AI literacy on a scale of 0 to 10?
If you’re hesitating to answer these questions – that’s wonderful! Because these questions don’t make any sense. I’d like to explain why in this article. It’s the first in a series of articles in which I’ll be unpacking the phenomenon of ‘AI literacy’.
AI literacy is on everyone’s lips and the term is becoming increasingly established (see Google Trends graph). There is a booming market for AI literacy courses. The common narrative behind the call for AI literacy: We will live in a world where we have to interact with AI. We are facing far-reaching changes in the labour market, and that is why we need an AI-skilled workforce;
AI literacy is regarded as so urgent and important that it is being anchored in law: in Europe, the AI Act requires companies to provide their employees with AI literacy training. In the US, the President has issued an ‘Executive Order’ declaring AI literacy among Americans a national priority. From 2029, the PISA study will explicitly measure AI skills. And March 5th can be marked in the calendar as ‘AI Literacy Day’.
Educating people about AI sounds, at first glance, like a thoroughly positive thing. And it’s something that is very close to my own heart: I studied AI (2010–2013), have been a lecturer at several universities, write blog posts and run workshops in which I explain AI to the general public. Demystifying AI and fighting the hype surrounding AI are among my main priorities. If I were allowed to choose just one word on LinkedIn to describe my professional focus, it would probably be ‘AI literacy’.
And yet I find the term ‘AI literacy’ problematic: in its appealing simplicity, it suggests that there is a standardised curriculum and a consensus on what there is to know about AI. This is false – for two reasons. Firstly, the ambiguity begins with the very term ‘artificial intelligence’: AI is not a single technology, but a group of highly heterogeneous technologies (check out my last article). Secondly, AI is a controversial, complex topic in which opinions, beliefs and ideologies play a major role. An “AI fanboy” (who celebrates every new foundation model release as the next breakthrough towards AGI) and an “AI refusal” advocate (who fundamentally rejects all kinds of AI) will hold very different views on what knowledge about AI and which skills are essential.
As important as reading and writing?
The English word ‘literacy’ has two meanings: the first is the ability to read and write. UNESCO states: “Literacy is a fundamental human right for all. It opens the door to the enjoyment of other human rights, greater freedoms, and global citizenship.”
The second meaning of ‘literacy’ refers to basic competence in a specific area: financial literacy denotes basic knowledge of finance – interest rates, investments, loans. Computer literacy refers to the confident use of digital tools.
AI literacy follows this second definition: a basic understanding of artificial intelligence and practical skills for everyday life. And yet the term ‘AI literacy’ still carries some of that first meaning. When the European Commission states that in a survey “63% of respondents agreed that everyone will need to be AI-literate by 2030” (Link), this wording evokes the existential significance of an essential cultural skill such as reading and writing. At the same time, it carries the unrealistic implication that AI literacy is a standardised skill that one either possesses or does not: literate or illiterate.
Who is AI literacy for?
The concept of ‘AI literacy’ is aimed at the general public – people from all backgrounds and levels of education – and usually includes children as well. AI literacy should be distinguished from specialist AI knowledge, such as that taught on a computer science degree course and hold by ‘AI experts’* such as AI researchers, ML engineers, AI engineers, AI agent architects (and whatever other technical roles the AI ecosystem might produce), AI ethicists or AI policymakers. Building your own computer vision model with TensorFlow or programming a gradient descent function in Python – these technical, low-level skills do not fall under AI literacy but rather form part of a machine learning engineer’s skill set. *However, the fact that ‘AI expert’ is not a protected term is evident from the many self-proclaimed ‘AI experts’ on LinkedIn (most of whom do not even have a technical background but rather a marketing one).
What does AI literacy involve?
What exactly is essential for AI literacy? Being able to rattle off the 10 “hottest AI tools” off the top of your head? Master prompt engineering in your sleep? Being aware that chatbots aren’t search engines? Knowing that the generative AI industry relies on exploitative labour practices? Understanding the links between AI development and ideologies such as longtermism and eugenics?
All of this may or may not be considered essential for AI literacy. What matters is how one views the complex topic of AI and what stance one takes on it:
Should AI be viewed as a set of tools or interfaces that are meant to be used as efficiently as possible? Or should AI be seen, in the bigger picture, as a complex socio-technical system based on the exploitation and extraction of human and environmental resources?
Should we view AI as a kind of force of nature that has come upon us and with which we now have to live? Or should we see the development of the kind of AI we have today (generative AI, particularly LLMs) as the result of a complex interplay between politics, the economy, research and hype?
Should AI be viewed as something to which one should, in principle, say ‘yes’ and overcome one’s reservations and scepticism? Or as a group of highly diverse technologies with varying risk profiles, the use of which should be assessed in a specific context, and where saying ‘no’ is always a legitimate decision?
To put it simply, the question is: Do you believe the narratives put forward by the AI industry and the AI hype, or not?
AI Literacy Frameworks
AI Literacy Frameworks promise a structured approach to AI literacy and provide guidelines from which specific curricula can be derived. There are several framework initiatives from around the world, including those from the OECD, UNESCO and the Digital Education Council.
Most of these frameworks agree that AI literacy has several dimensions: knowledge, skills and attitude.
The ‘Knowledge’ dimension includes, for example, facts about the fundamental technical basics of machine learning (‘How does AI work?’), the types of AI (generative, discriminative, predictive), different types of machine learning (supervised, unsupervised, reinforcement learning), as well as their capabilities and limitations.
The ‘Skills’ dimension refers to the ability to apply generative AI as a user and how to use specific tools as efficiently as possible (prompting), how to evaluate AI outputs, and which ethical considerations come into play when using it.
The third dimension, “Attitude”, describes the desired underlying attitude towards AI. In most cases, this is a positive rather than neutral attitude. For example, the OECD’s AILit Framework states: “believe AI can be a powerful tool for creating positive change in their own lives and the lives of others.”
Wild growth: “AI upskilling” courses – the AI literacy industrial complex?
Apart from the thoughtful AI literacy frameworks, there is a rapidly expanding and unregulated market of courses promising ‘AI upskilling’ and ‘AI fluency’. Here, there is a clear focus on the ‘Skills’ dimension – practical competence in using generative AI tools for daily tasks. In the world of these AI courses, knowing how to prompt effectively (“prompt engineering”) is regarded as an essential skill. Participants are encouraged to automate as many areas of their personal and professional lives as possible using generative AI. “You’re shit at AI” is the aggressive marketing slogan of a UK-based AI upskilling startup that has just made a name for itself on the British start-up show Dragon’s Den.
There are many companies like these. In Germany, the Employment Agency uses taxpayers’ money to fund AI training courses run by private companies. Many of these companies have sprung up overnight. It is reminiscent of the commercialisation of the ‘Agile’ concept in software development: a multi-billion-dollar ecosystem of consultancies, certifiers and framework vendors, for which the term ‘Agile Industrial Complex’ was coined at the time. Are we now witnessing the emergence of an ‘AI Education Industrial Complex’ amid the hype surrounding AI literacy?
Why this is problematic
The fundamental problem is that much of what falls under the umbrella of ‘AI literacy’ simply adopts the narrative of AI hype and Big Tech:
Firstly, the ‘AI inevitability’ narrative: it portrays AI – particularly generative AI and large language models (LLMs) – as an inevitable step in the evolution of technology or even of humanity, rather than as something driven by capitalist interests. This narrative is problematic because it disempowers the public, rather than showing them what agency they have to influence these developments.
Secondly, they are overselling the capabilities of AI: Generative AI is not intelligent, but is based on statistics and the imitation of intelligence – it is an error-prone, non-robust technology. In many AI courses, these errors and limitations are mentioned, but often phrased far too mildly. Statements such as “AI can make mistakes” or “AI can hallucinate” are common – but these are often followed by “but AI is getting better all the time” or “humans make mistakes too”. This is dangerously misleading: hallucinations are an intrinsic part of the probabilistic nature of LLMs; the errors made by LLMs differ fundamentally from human errors, as they lack common sense and operate without contextual knowledge. To make matters worse, people often fail to recognise misinformation: LLM output sounds eloquent and confident and people tend to consult them precisely when they themselves lack the knowledge to verify the answer.
Thirdly, the narrative is being promoted that artificial intelligence is ‘artificial’: whilst explaining how AI works, technical aspects are described (which are certainly useful to know), but the crucial role of human labour in the production of ‘AI’ is often not sufficiently emphasised: Anyone who talks about ‘supervised learning’ and ‘labeling’ without mentioning that these labels are created by people – data workers who often work under precarious conditions – is only telling half the story. Anyone who says that deep learning requires vast amounts of training data, but fails to mention that its acquisition is based on a breach of intellectual property rights relating to the creative work of countless people, is also not giving the full picture. AI education that conceals the exploitative practices of the AI industry and shifts ethical issues onto users (e.g. “If you prompt it this way, you’ll use less energy”) is highly problematic.
The overemphasis of practical skills is particularly irresponsible. Firstly, it implies tacit acceptance of the unethical production conditions in the AI industry. Furthermore, it imposes a ‘techno-solutionism’ mindset on the public: AI can solve many, if not all, problems, and more aspects of our lives should be automated;
What good are we actually doing for learners by teaching them how to use AI tools? So-called skills such as prompt engineering and tool fluency age like milk, given the extremely volatile market for consumer AI applications. And what’s even more problematic is that over-reliance on AI tools has been shown to lead to actual deskilling.
It is irresponsible to teach people tools and practices that flood the world with even more slop. What I find particularly concerning in this sphere is the trend to teach people ‘workflow automations’ for their job – often after just one day of ‘AI training’. Creating automation workflows used to be a complex IT task carried out by trained software developers, involving quality assessment and testing. Handing such tasks over to IT novices with absolutely no knowledge of software best practices or data protection is irresponsible – it is the democratisation of the credo ‘Move fast and break things’.
Some criticise and condemn the concept of ‘AI literacy’ as a whole because they equate it with that uncritical ‘AI upskilling’ approach. For example, Miriam Reynoldson, a digital learning specialist from Australia, writes: “‘AI literacy’ is a dangerous tool of neoliberal education and it deserves to be dismissed out of hand”. But is the author aware that, through her article, she herself is carrying out important educational work on AI and is thus contributing to critical AI literacy?
The urgent need for a Critical AI Literacy
There appear to be various ‘brands’ of AI literacy. I see a parallel here with the situation in AI ethics: there, too, are two camps with differing views on what the main concerns of AI ethics should be.
On the one hand, there are AI ethicists such as Timnit Gebru, Margaret Mitchell, Emily Bender, Cathy O’Neil who vocally point out the systemic issues within the AI industry and the real harm that AI systems are already causing today: from algorithmic discrimination and misinformation to mass surveillance. As well as the social and environmental costs of their production and the accumulation of wealth.
On the other hand, there is the ‘AI Safety’ camp – that is, those who warn of the potential dangers posed by a possible super-intelligent AI in the distant future, or who even attribute the status of a moral agent to AI. This strand of AI ethics is closely linked to ideologies such as effective altruism, longtermism and transhumanism. Terms that indicate affiliation with this camp include ‘AI alignment’ and ‘existential risks’. Many industry leaders – including OpenAI, Google and Anthropic – as well as many AI start-ups belong to this camp. Some argue that the focus on existential risks is a diversionary tactic designed to distract attention from the actual, real-world harm caused by the AI industry in the here and now, as regulation in this area could have a massive impact on these companies’ business models.
Just as with AI ethics, there are at least two ‘brands’ of AI literacy. So how does one distinguish one form of literacy from the other?
Over the past three years, the term ‘Critical AI Literacy’ – or the plural form ‘Critical AI Literacies’ – has emerged to distinguish itself and promote a more responsible approach to AI literacy. I find it very appealing because it ties in with the emerging field of ‘Critical AI studies’, which promotes a holistic, critical and hype-resistant view of AI.
There is no standardised curriculum for critical AI literacy either, but the various approaches share some common features:
it does not view AI as a purely technical subject, but as a socio-technical system
it promotes a realistic view of the capabilities of different types of AI, particularly generative AI; it does not downplay its shortcomings and fragility
it highlights the extractive, exploitative practices of the (generative) AI industry, and raises awareness of the negative impact that AI is already having on humans and the environment
it educates people about AI in order to enable them to make informed choices, including the choice to refuse the use of (generative) AI technology
Critical AI literacy faces several challenges: It must hold its ground against the pervasive and well-funded marketing narratives of the AI industry. Also, the question arises as to how one should raise awareness of the environmental and social costs of developing AI when key facts are deliberately withheld by companies.
But things are changing, we are at a turning point: Studies show evidence that the public is developing a more critical awareness of AI. And many grassroots movements that oppose the trajectory and narrative of the AI industry are becoming more visible, as shown in the recently launched AI Resist List.
In my next article, I will share my thoughts on developing a curriculum for responsible critical AI literacy. The aspects I would particularly like to focus on are:
Mental models for (generative) AI – How can we challenge the ‘intelligence’ narrative? What other metaphors are there to help us think about AI and evaluate its use?
Nuance – What type of AI are we talking about? Where can which type of AI be useful? In which areas of application do the benefits justify the high resource costs?
Use Case Literacy – How to recognise when and why (generative) AI isn’t the right tool for the job
Metacognition – How am I influenced by automation bias and sycophancy when using chatbots?
Hype Resilience – How to spot AI snake oil and debunk common marketing narratives and nonsense
Working with AI is not an essential skill, but the ability to make informed decisions about when to use it is indispensable. It is perfectly possible to be very knowledgeable about AI and yet refuse to use generative AI tools because you have weighed up the benefits they offer against the negative impacts they have on yourself, other people and the environment.
We must not leave education about AI to Big Tech!
Credits:
Cover image: Hanna Barakat & Cambridge Diversity Fund / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/