TLDR: A research paper argues that debates about AI’s long-term future are driven more by philosophical disagreements on technological change than by technical facts. It identifies “transformationalists” (who believe in inevitable, profound AI impact) and “skeptics” (who doubt high expectations), and outlines three core non-technical questions that divide them: the possibility of non-biological intelligence, appropriate prediction timeframes, and the trajectory of technological growth (exponential vs. S-curve/stagnation). The paper concludes that the transformationalist view carries a high argumentative burden and calls for broader expertise, including humanities and social sciences, in AI discussions.
A new research paper delves into the core of debates surrounding the future of artificial intelligence, suggesting that many prominent discussions are less about the technology itself and more about fundamental philosophical disagreements concerning the history and trajectory of technological change. Authored by Mark Fisher and John Severini, the study, titled “Making AI Inevitable: Historical Perspective and the Problems of Predicting Long-Term Technological Change,” argues that understanding these underlying perspectives is crucial for navigating the complex landscape of AI development. The paper identifies two main groups: “transformationalists,” who believe AI will inevitably have a profound societal impact, and “skeptics,” who doubt AI will meet such high expectations. These groups differ on key questions, including the possibility of non-biological intelligence, the appropriate timeframes for technological predictions, and the assumed path of technological development. You can read the full paper here: Making AI Inevitable.
Transformationalists: The Vision of Inevitable Progress
The transformationalist camp is united by a belief that Artificial General Intelligence (AGI) is on the horizon and will profoundly alter human society. This perspective is deeply rooted in the Singularitarian movement, which posits a “Singularity” – a point where technological progress becomes so rapid it’s beyond human comprehension. Figures like John von Neumann and Henry Adams are cited as early thinkers who conceptualized accelerating technological change. Vernor Vinge further developed this, identifying mechanisms like constant acceleration, intelligence explosion, and competitive advantage as drivers.
Ray Kurzweil, a prominent “AI Visionary,” is a key proponent of strong transformationalism. He introduced the “Law of Accelerating Returns,” which suggests that while individual technologies follow S-curves (periods of growth followed by plateaus), the overall pattern of technological evolution is a “cascade of S-curves,” leading to continuous exponential growth. Kurzweil extends this law to the entire universe, seeing it as a fundamental process of increasing order, culminating in the fusion of human and artificial intelligence. While some critics view his ideas as bordering on religiosity, his influence is undeniable.
Weak transformationalists, like Nick Bostrom, acknowledge the possibility of AGI and its transformative effects but are more epistemically modest, viewing it as a likely possibility rather than an absolute certainty. However, the paper suggests that even these probabilistic assessments are often influenced by the underlying assumptions of strong Singularitarian thinking, sometimes hidden behind expert surveys. More recent strong transformationalists, such as Sam Altman and Mustafa Suleyman, continue to articulate visions of unstoppable, exponential technological progress, often attributing an autonomous logic to “technology” itself.
Skeptics: Questioning the Exponential Trajectory
Opposing the transformationalists are the skeptics, a diverse group united by their reservations about AI’s grand promises. Early philosophical arguments from thinkers like John Searle and Colin McGinn questioned the very possibility of non-biological intelligence and the understanding of consciousness. Margaret Boden, a contemporary philosopher, criticizes Singularitarians for misunderstanding the significant obstacles to achieving AGI.
Weak skeptics, such as Theodore Modis, engage with long-term technological predictions but dispute the specific rates of growth. Modis, whose work on complexity in large systems was even used by Kurzweil, emphasizes that phenomena subject to competitive pressures often follow S-curves, indicating natural limits or diminishing returns, rather than endless exponential growth. The debate between Kurzweil and Modis often centers on model specification – whether progress fits an exponential or sigmoid curve. Modis’s own predictions, like the stabilization of internet adoption rates, have shown the challenges of applying such models to complex historical processes.
Contemporary macroeconomists also contribute to the skeptical viewpoint, focusing on AI as a general-purpose technology. While some acknowledge AI’s potential to boost productivity, much research highlights a recent slowdown in productivity growth across industrialized nations. This slowdown, potentially caused by increasing complexity and “innovation bottlenecks,” suggests that “easy” innovations may have already been found, and research productivity might even be halving, not doubling. Concerns about AI’s reliance on training data, leading to “model collapse” where AI merely replicates its own generated data rather than learning novel information, further fuel skepticism. Economists like Daron Acemoglu express deep uncertainty about long-term AI forecasts, warning against “AI hype” and limiting their analyses to shorter timeframes.
Three Fundamental Questions Shaping the AI Debate
The paper distills the core disagreements into three fundamental, non-technical questions that shape perspectives on AI’s future:
1. Is non-biological intelligence possible, and can we recognize it? This question probes whether genuine intelligence can emerge from non-biological systems. Strong skeptics argue intelligence is inherently biological, making artificial intelligence a misnomer. Transformationalists, however, point to the vastly different speeds of biological neurons versus transistors, suggesting artificial superintelligence could be a distinct and more powerful category. The challenge lies in defining and recognizing intelligence beyond quantitative measures, especially when AI struggles with tasks requiring meaningful reasoning and contextual understanding, like the “frame problem.”
2. What are appropriate time frames for technological predictions? This question highlights the stark difference in scope. Transformationalists often look back to the Big Bang to derive long-term trends, a methodology that mainstream academia, particularly in humanities and social sciences, has historically been indifferent or skeptical towards. This academic bias has led to an idiosyncratic methodological toolkit in “big history” and Singularitarian thinking, creating a tension with more context-specific academic scholarship.
3. Is exponential growth more likely than steady-state growth or stagnation? This question addresses the overall trajectory of technological development, often conceptualized as aggregate productivity growth. The paper proposes a tripartite model of possible futures:
- Stagnation: Where bottlenecks become more numerous and more difficult to deal with as the complexity of the world increases. This has a negative compounding effect on the creation of new innovations, which begin to require continually greater amounts of time and effort to successfully create and diffuse.
- Steady-state: Where bottlenecks and innovations remain roughly in equilibrium. Bottlenecks may become more numerous and more difficult to deal with, but new ideas keep pace, allowing for a relatively proportional rate of the creation and diffusion of new innovations.
- Singularity: Where bottlenecks are still likely to become more numerous and more difficult to deal with, but where new ideas and innovations enable relatively faster growth. This has a positive compounding effect on the creation of new innovations, which become easier to create and diffuse as a result of previous innovations continually lowering the costs of subsequent technological development.
Transformationalists align with the Singularity world, while skeptics like Modis foresee S-curves and potential plateaus. Contemporary macroeconomists often lean towards a steady-state or even stagnation, given current productivity slowdowns and the increasing difficulty of finding new ideas.
Also Read:
- Navigating AI’s Moral Compass: A Call for Dynamic Value Alignment
- Unpacking AI’s Influence: How Language Models Persuade and Amplify Bias
Conclusion: Beyond Technicalities
Ultimately, the research paper concludes that the debates surrounding AI’s long-term impact are fundamentally shaped by these non-technical, philosophical disagreements rather than purely scientific or technical ones. The transformationalist position, especially its strong variant, carries a significant argumentative burden, requiring belief in the feasibility of non-biological intelligence, the constancy of historical processes over long terms, and continuous exponential technological growth. The paper emphasizes that while these arguments are difficult, they are not impossible, particularly in their weaker forms.
The study highlights the self-reinforcing nature of transformationalist beliefs, noting that the competitive drive for “first-mover advantage” among those who believe AGI is inevitable can itself accelerate its pursuit. To foster a more nuanced and agency-driven approach to AI’s future, the authors advocate for broadening the concept of “expertise” in these discussions. This means incorporating more philosophers of history, science, and technology, epistemologists, and political theorists, as the core questions are often not amenable to purely scientific answers. The paper calls for academia to engage more directly with these long-term historical and technological questions, moving beyond dismissiveness to contribute to a more informed collective response to AI’s ongoing development.
