The Nobel Prize in Physics 2024
This year’s Nobel Prize in Physics was aimed at recognizing the great strides made in artificial intelligence (AI), honoring two of its most distinguished creators. Their merits are highlighted in the official information released by the Royal Swedish Academy of Sciences: “This year’s two Nobel Laureates in Physics have used tools from physics to develop methods that are the foundation of today’s powerful machine learning. John Hopfield created an associative memory that can store and reconstruct images and other types of patterns in data. Geoffrey Hinton invented a method that can autonomously find properties in data, and so perform tasks such as identifying specific elements in pictures.”[1]
These discoveries have led to developments in technology that are changing the way we live. Every time we get a product recommendation on our electronic device, perform a Google search, or book a trip online, AI is behind it. It has also infiltrated our homes, where small robots that assist us in the kitchen or clean the floor are already popular; in our work and our factories, where robots have long been carrying out complex and, above all, repetitive tasks; in our hospitals, with numerous diagnostic applications; in the financial world; and in market research. It would be a never-ending task to explain all the practical applications of AI present in our everyday life.
Comments focus on the ethical aspects
From the outset, Hinton’s thoughts on the potential dangers of AI took on particular relevance in the information field.[2] He himself was responsible for highlighting these in his interview by the Royal Academy’s organization: “In particular with respect to the existential threat of these things getting out of control and taking over, I think we’re a kind of bifurcation point in history where in the next few years we need to figure out if there’s a way to deal with that threat. I think it’s very important right now for people to be working on the issue of how will we keep control? We need to put a lot of research effort into it. I think one thing governments can do is force the big companies to spend a lot more of their resources on safety research.”[3] We have already seen this idea in media headlines such as Fortune: “Nobel laureate Geoffrey Hinton is both AI pioneer and front man of alarm”[4]; and CNN: “With AI warning, Nobel winner joins ranks of laureates who’ve cautioned about the risks of their own work.”[5]
The ethical concerns of AI
AI is undoubtedly the most transformative technology of our time and raises deep ethical concerns in relation to its use. Three major threats are inherent in the very essence of this technology:
- Privacy;
- The reliability of the data with which the systems are trained; and
- The responsibility for decisions made by AI.
Privacy needs strict protection against its potential invasion, since the effectiveness of AI depends on the availability of large volumes of personal data subject to processes for collection, storage and use.
AI systems are only as good as the data with which they are trained, so objective data selection is essential. This is tremendously complicated, if not impossible, in open systems like ChatGPT, so developers and researchers need to prioritize and standardize rigorous testing and continuous monitoring.
AI systems are increasingly making decisions that affect our lives with the introduction of self-driving cars and medical diagnostic systems, and it is therefore critical to establish clear lines of responsibility.
Society is unprotected against all these, since technology is changing so fast that even the most informed legislators cannot keep up, and there are no realistic investments to do so. This was affirmed by Joseph Fuller, a professor at Harvard Business School: “The regulatory bodies are not equipped with the expertise in artificial intelligence to engage in [oversight] without some real focus and investment.”[6]
To these ethical problems inherent to the operation of AI must be added others derived from its inappropriate or malicious use, such as cyberattacks, disinformation that allows public opinion to be manipulated and social divisions to be amplified, and the development of autonomous weapon systems. However, what some fear most on seeing the rapid rate of AI development is the possibility that AI systems will surpass human intelligence, demanding that action be taken to ensure their control and alignment with human values.
Geoffrey Hinton’s warning
The fact is that Hinton turned away from AI development work to be able to freely express himself about the dangers of AI and actively participate in shaping the regulations for its development, collaborating with international organizations and research institutions, and providing ideas on how to address the ethical challenges posed by advances in AI.
After leaving his position at Google, he gave a lecture in Cambridge entitled “Two Paths to Intelligence,”[9] in which he compared the way biological intelligence and AI operated, and explained the reason for his fears that it could surpass human intelligence and take control.
In his presentation, he distinguishes between digital computation, in which the software is independent of the hardware, so that programs can be transferred from one computer to another, making knowledge “immortal”; and biological computation, which takes place in the human brain, exploits the “analog” properties of its “hardware” (neurons, synapses) and in which the results are stored in the brain itself, so that knowledge cannot be transferred, defining it as “mortal” knowledge.
This leads to key differences in two aspects: knowledge sharing and learning mechanisms.
Digital computation makes it possible to instantly merge learned information between the different copies of a neural network, which gives digital intelligence greater speed in knowledge acquisition and dissemination; biological systems, on the other hand, are based on a slower process of learning by observing and replicating a teacher.
Learning in digital computation uses backpropagation, which, although apparently a “dumb” algorithm, turns out to be a powerful algorithm for adjusting the connections in a neural network based on the errors it makes, and that when combined with the computational power and precision of digital systems, would allow more effective learning than biological computation.
Moreover, access to data and the computational power of digital intelligence is virtually unlimited compared to the limited knowledge capacity of biological brains.
The controversy over the superiority of human intelligence
Scientists and philosophers have been arguing for decades about the possibilities of AI surpassing human intelligence. Humans acquire knowledge organically, by having experiences and understanding their relationship with reality and with ourselves. However, this mechanism is something extremely complex that we do not understand, and that includes perception, communication, memorization and information with qualities such as emotion, subjectivity, intentionality and attention. Furthermore, human intelligence allows us to comprehend abstractions that are not perceptible to the senses. In contrast, artificial neural networks learn in the abstract, processing huge information repositories through the use of algorithms without relating it to reality, i.e., they lack genuine emotional experiences.
Although neural networks were originally developed in an attempt to mimic human intelligence, the reality is that we are looking at two very different ways of acquiring knowledge. What Hinton is saying is that the intelligence shown by AI systems transcends their artificial origins and could be better than the human brain. Moreover, he argues that if AI becomes much smarter than humans, it will be very good at manipulating us without us realizing it. Equally, we might not realize that we are being manipulated by AI because it would be much smarter than us.[10]
Nevertheless, there are other ways of thinking: that of those who do not see that AI could become a threat; and most radically, that of those who believe it is impossible for AI to reach the level of human intelligence.
Yann LeCun, a former student and collaborator of Hinton’s, currently vice president and chief AI scientist at Meta, believes that AI is currently running into the limits of the physical, and expresses it by stating that “[a]ny cat can jump on a series of pieces of furniture and get to the top of some shelf. We don’t have any A.I. systems coming anywhere close to doing these things today, except self-driving cars,” and they are overengineered, requiring “mapping the whole city, hundreds of engineers, hundreds of thousands of hours of training.”[11] Still, he does have faith that this will be overcome by a simple basic idea: if neurons can do it, then so can neural networks. He says, however, that it will never pose a threat: “AI assistants will eventually be smarter than us, but we should not feel threatened by that. We should feel empowered by it. It’s kind of like everybody would walk around with a staff of smart people working for them. There is nothing better than working with people smarter than you, right?”[12]
There are many who defend the superiority of human thinking compared to AI based on different arguments. The mathematical logic perspective bases its argument — as does Nobel laureate Roger Penrose — on Gödel’s incompleteness theorem, according to which the capacity of any algorithm-based system is limited. Moreover, in human nature, there are noncomputable features; our brain works according to the laws of quantum mechanics.[13],[14]
Neuroscience also argues that artificial algorithms attempt to mimic only the conscious function of parts of the cerebral cortex, ignoring the fact that not only is all conscious experience preceded by an unconscious process, but also the passage from the unconscious to consciousness is accompanied by loss of information.[15]
A still open debate
Although the arguments for the supremacy of human intelligence over AI may seem robust, our limited knowledge of the nature of the universe and the mechanisms of the brain does not allow the opposite possibility to be completely ruled out. The sensible thing will be to consider the recommendations of Geoffrey Hinton and prioritize transparency and security in the development of AI. His recent appointment as a Nobel laureate will undoubtedly lend greater visibility to his proposals.
Manuel Ribes
Bioethics Observatory – Institute of Life Sciences
Catholic University of Valencia
References:
[1] Press release: The Nobel Prize in Physics 2024 – NobelPrize.org 8 October 2024
[2] M. Ribes La inteligencia artificial como problema UCV Bioethics Observatory 20 July 2021
[3] Geoffrey Hinton – Interview – NobelPrize.org October 2024
[4] Paolo Confino Nobel laureate Geoffrey Hinton is both AI pioneer and front man of alarm Fortune 10 October 2024
[5] Meg Tirrell With AI warning, Nobel winner joins ranks of laureates who’ve cautioned about the risks of their own work CNN 13 October 2024
[6] Christina Pazzanese Great promise but potential for peril The Harvard Gazette 26 October 2020
[7] Katyanna Quach Top Google boffin Geoffrey Hinton quits, warns of AI danger The Register 1 May 2023
[8] Bobby Allyn ‘The godfather of AI’ sounds alarm about potential dangers of AI NPR 28 May 2023
[9] Geoffrey Hinton – Two Paths to Intelligence (25 May 2023, Public Lecture, University of Cambridge) https://blog.biocomm.ai/2023/06/06/cser-cambridge-geoffrey-hinton-two-paths-to-intelligence-06-june-2023/
[10] Yunzhe Wang A Synopsis of Geoffrey Hinton’s Warning of Humanity’s Existential Threat from AI Medium May 7, 2023
[11] Joshua Rothman Why the Godfather of A.I. Fears What He’s Built The New Yorker 13 November 2023
[12] Business Standard AI will amplify human intelligence, not replace it, says Meta’s Yann LeCu 23 October 2024
[13] F. Gelgi Implications of Gödel’s Incompleteness Theorem on A.I. vs. Mind NeuroQuantology, Issue 3, 186-189, 2004
[14] D. Heredia Penrose and his position against the possibility of the computability of the human mind and consciousness Dialnet Depósito de Investigación Universidad de Sevilla 4 July 2024
[15] Athanassios S Fokas Can artificial intelligence reach human thought? PNAS Nexus, Volume 2, Issue 12, December 2023, pgad409 https://doi.org/10.1093/pnasnexus/pgad409