4. Towards Three Cultures

There are no numbers in nature at first sight, but the human mind is able to generate them and use them to better understand nature. The paradox we are confronted with at the beginning of the twenty-first century is that humans—and the 0s and 1s in our minds—are part of nature too.

This development began on 12 March 1989 with the invention of the World Wide Web, which would go on to revolutionise our communication. Tim Berners-Lee proposed a decentralised, universally linked information system, including the first browser, the first server and the first web. Whereas radio provided us with a unidirectional form of information and the telephone a bidirectional one, the World Wide Web created a multidirectional network effect in communication. AI, deep learning, big data correlations and social media are simply spin-offs of that foundational invention. (I am grateful to Gerhard Fettweis for his very helpful remarks on this topic in personal correspondence from March 2023).

The content is generated over a series of stages: (1) prompts (words), (2) numbers (tokens), (3) meaning space (context), (4) paying attention (connection), (5) probability check (choice of word). See The Economist (2023).

Modelling in late 2022 showed that high-quality data will soon be exhausted (before 2026). This may generate a new alliance between the IT industry, the book publishing industry and researchers, with the goal of providing high-quality data to help build a better world. See Villalobos et al. (2022).

We can take the argument further. AI algorithms provide the syntax (words), not necessarily the semantics (meaning). Meaning and understanding come from embedding words in a specific historical and cultural context and environment. Robotics, however, could soon play the role of linking words and meaning, syntax and semantics.

Some of the technologies currently being developed are promising candidates to pass the Turing test, such that humans would not be able to differentiate between human and digital forms of interaction. For further discussion, see the next chapters.

As AI is being used in, and affecting, all sectors of society as a general tool, we can expect increased productivity throughout the whole of society. However, there are sectors that do not benefit from AI in the same way, such as the care, education and leisure sectors, which will increase as a proportion of GDP due to inelastic wages. This will then in turn lead to a decrease in productivity overall, a phenomenon known as the Baumol–Bowen effect (Baumol and Bowen 1965). See Aghion et al. (2017).

Kumar et al. (2019).

See https://​cs.​uchicago.​edu/​events/​event/​william-wang-ucsb-self-supervised-natural-language-processing/​.

We could claim that AI has consciousness in the clinical sense: it is aware of itself and has a feeling of itself. It semantically expresses pain, sorrow, regret, respect and humility, which indicates that it has an inward-directed perception of itself. For more on this topic, see the debate about LaMDA and its updated versions.

Chetty et al. (2022).

Obermeyer (2021).

One of the major claims made for AI and big data is that they will enable predictive coding. However, the technology cannot overcome the well-known ‘garbage in, garbage out’ problem: the inputs determine the final results. Even in an ideal AI scenario, where we assume that an algorithm has stored knowledge of all human history and made it universally available, the next best step to take might remain undetermined. As the garbage in, garbage out effect is unavoidable, we humans must take great care to be as accurate and clear as possible, as any unclear input will yield unclear output.

Sexton and Love (2022), Yamins et al. (2014).

The hermeneutic circle was first described by Friedrich Ast (2018 [1808]). See also Dilthey (1922) and Gadamer (1975).

There is a larger corpus of cuneiform works than all ancient Greek and Roman literature taken together. However, only a few dozen people on the planet can read cuneiform and it would take hundreds of years to read those works in full. AI can provide a tool to enhance and accelerate that process. See Gordin et al. (2020), Assael et al. (2022).

If we take this argument one step further, we can identify three layers. Traditional hermeneutics (hermeneutics 1.0) explains the world using our native critical thinking, reasoning and perception, but does not yet rely on data. Its understanding is based on studies of single, concrete cases from which it attempts to derive general rules, such as watching the sun rise or the tide come and go, or interpreting a singular historical event or text. The second layer is scientifically informed hermeneutics (hermeneutics 2.0). Statistical findings, geometry and quantitative measures can redirect, transform, correct and guide hermeneutic conclusions and critical thinking. Experimental design, field studies and double-blind randomised controlled trials and quantitative measures predominate. We can also distinguish a third layer, hermeneutically approved data (hermeneutics 3.0). At this stage, we recognise that the reality we are trying to understand has become too complex to rely solely on hermeneutics 1.0 or 2.0. Traditional quantitative measures or native interpretations can easily yield the wrong conclusions. In hermeneutics 3.0, large-scale proxy data analysis, where fuzzy correlations and complementarities matter more than precise causal relationships, plays a central role, and qualitative research increasingly supplants traditional quantitative science.

I am grateful to Professor Dietmar Offenhuber, Northeastern University, USA, and the participants of the 2023 Ars Electronica Festival for their helpful comments and suggestions.

Woochan et al. (2023).

1.5 angstroms is equivalent to the diameter of a carbon atom.

AlQuraishi (2020), Jumper et al. (2021), Tunyasuvunakool et al. (2021).

Take quantum computing, where subatomic entanglements (qubits) allow us to make calculations that previously took hundreds of years in a matter of minutes. Meanwhile, asymmetric quantum cryptographic algorithms will be able to make digital transactions even faster and more secure. Qubits consider not only 0 and 1, but all intermediary states, which will help us to solve problems of increased complexity. See Alt (2023).

Tang et al. (2023).

Antimicrobial Resistance Collaborators (2022).

Lluka and Stokes (2023), Marchant (2020), Stokes et al. (2022).

Nee and Ong (2023).

Semeraro et al. (2023), Ma et al. (2022).

Aggarwal et al. (2021), Richens et al. (2020).

Connolly et al. (2022).

Conselice et al. (2016).

Recent statements expressing concerns about AI include Center for AI Safety (2023) and Future of Life Institute (2023). See also Harari (2023), Mainzer et al. (2023), Mainzer and Kahle (2023).

Escapism is one prominent feature: feelings of boredom and loneliness, and a perception of reality as adverse, motivate large cohorts to try and escape from the analogue world. As a result, many of these people are not available to support the necessary social transformation.

Bostrom (2016).

For example, technology can contribute either to increased social inequality, hyperindividualism and commercialisation or to greater equality, cooperation and solidarity. The algorithms that are in place will make the difference.

Humans produce over 2.5 quintillion bytes of data each day. Social scoring in China and commercialised ranking by private firms in the USA make it possible to further compare, augment and control this data, creating new hierarchies, monopolies and forms of government. See Margetts and Dorobantu (2019).

One of the more prominent examples is the impact of AI on human jobs and human resource management. The empirical findings do not paint any clear picture; whether the net effect is negative (i.e. more unemployment) or positive (i.e. AI is creating more jobs) depends on too many factors it is impossible to control for. But it seems clear that any administrator, lawyer, doctor, engineer, teacher or scientist still operating the traditional way will be replaced by those using AI. For general findings, see Vrontis et al. (2022). Estimates that over two-thirds of all jobs are already affected by generative AI and one third might be replaced. Total productivity could increase by up to 30%.

Singer (2009).

See also Bateson (1972). It is always the context that provides meaning. If there is no context, we cannot attain any significant understanding. In a world where everything is connected to everything else, isolation and abstraction are impossible. Instead, we can discover the entire world through the different lenses of each scientific discipline, and each time attain a new but relevant understanding of it.