Epistemology of AI
Understanding chatGPT and its descendants - Media Ecology Newsletter #27
Data Ethics and AI Ethics are increasingly important disciplines for deciding what’s right to do with artificial intelligence. But the ways in which ChatGPT, Bard and similar technologies have captured the attention of many also compel reflection on the significance of the knowledge emerging from the use of these technologies. There is an AI Ethics, but there is an AI Epistemology, too.
In this issue of "Media Ecology" we start with two readings. And then we will simply try to list the issues to be addressed. Because this matter is going to be a long journey.
Understanding AI
Oops! How Google bombed, while doing pretty much exactly the same thing as Microsoft did, with similar results, by Gary Marcus
The bestselling author writes about the meaning of all those mistakes that are made by generative artificial intelligences.
Marcus had written about the reasons why ChatGPT and similar products make mistakes also a couple of months ago.
How come GPT can seem so brilliant one minute and so breathtakingly dumb the next? also by Gary Marcus
There is a specific philosophical research about the epistemology of AI and I’ll link some examples later in this issue.
AI understanding
How to talk about books you have not read? For the educated person, it is possible and legitimate, according to Pierre Bayard, a psychoanalyst and author of a book with that question as its title, a real bestseller some 15 years ago. So what's the harm if ChatGPT writes about things it doesn't understand and knows nothing about? OpenAI's generative artificial intelligence that could write instead of humans, for now, has only prompted many humans to write about artificial intelligence. Actually, on a cognitive level, that artificial intelligence is rather novice: those who solicit it to talk about matters they do not know are usually surprised by its eloquence, but those who question it on matters they know well find that it is not very clear in its ideas and fills its answers with errors. Well: could the announcements of these days change the scenario? (Please read this article on the Oecd Forum: Knowledge ecology of the new search engines, by Luca De biase
What is the problem
The epistemology of AI should be about the value of AI in terms of knowledge creation.
Here are three questions:
When the AI chat responds to us by simulating an understanding of the problem, it can trick people who get drawn into anthropomorphism: the simulation of understanding may appear to be real understanding. How to avoid the most important consequences of this misunderstanding?
To correct this flaw we can rely on a modern version of philology, suggests philosopher and MIT affiliate Cosimo Accoto. But will this become operational through education or through a redesign of platforms, once they are held accountable for any harm they may generate by spreading false or falsely perceived information?
To define how artificial intelligence will contribute positively to knowledge we will have to reframe the discussion to take into account not just the machines, but the social and technological construct that actually generates the consequences we are talking about.
More about epistemology of AI
A good strategy seems to be the one that divides the problem in small more defined pieces had written computer scientist John McCarty in 1981.
Epistemological problems of artificial intelligence, by John McCarty (1981)
Jean-Gabriel Ganascia opens on the many contemporary applications of artificial intelligence that have already transformed—and will continue to transform—all our cultural activities and our world. And it places those perspectives in the context of the philosophy of information and more particularly it emphasizes the role played by the notions of context and level of abstraction in artificial intelligence. (2010)
Epistemology of AI Revisited in the Light of the Philosophy of Information, by Jean-Gabriel Ganascia (2010)
Gregory Wheeler proposes a different perspective.
Epistemology and artificial intelligence, by Gregory Wheeler (2004)
In this essay we advance the view that analytical epistemology and artificial intelligence are complementary disciplines. Both fields study epistemic relations, but whereas artificial intelligence approaches this subject from the perspective of understanding formal and computational properties of frameworks purporting to model some epistemic relation or other, traditional epistemology approaches the subject from the perspective of understanding the properties of epistemic relations in terms of their conceptual properties. We argue that these two practices should not be conducted in isolation.
Podcasts in Italian, by me
L’altra metà del verso. Rai Radio 3
Media Ecology. Intesa Sanpaolo on air
Eppur s’innova. Luiss University Press
Have you seen this?
Epistemic AI, coordinator Fabio Cuzzolin, in Oxford.
«Epistemic AI breaks entirely with the current state of artificial intelligence and with the most exciting ongoing efforts, such as continual learning (making the learning process a life-long endeavour), multi-task learning (aiming to distil knowledge from multiple tasks to solve a different problem) or meta-learning (learning to learn). As these are all still firmly rooted in AI’s conventional principles, they fail to recognise the foundational issue that the discipline has with the representation of uncertain knowledge.
Our proposal goes beyond ‘human-centric’ AI, the push to make artificial constructs more trustable by human beings and more capable of understanding humans, since it strives to model the uncertainty stemming not just from human behaviour, but from all sources of uncertainty present in complex environments.
Epistemic AI’s overall objective is to create a new paradigm for a next-generation artificial intelligence providing worst-case guarantees on its predictions, thanks to a proper modelling of real-world uncertainties. Firstly, a new mathematical framework for optimisation under epistemic uncertainty will be formulated, superseding existing probabilistic approaches. The new optimisation framework will lay the premises for the creation of new ‘epistemic’ learning paradigms. In Epistemic AI we will focus, in particular, on some of the most important areas of machine learning: unsupervised learning, supervised learning and reinforcement learning.
Last but not least, the goal of the project is to foster an ecosystem of academic, research, industry and societal partners throughout Europe able to drive and sustain the EU’s leadership ambition in the search for a next-generation AI.»
Ecology of screens
On the occasion of the International Conference Vivre par(mi) les écrans: entre passé et avenir, which was held in Lyon at the end of May, the newsletter of the International Research Group Vivre par(mi) les écrans and the Media Ecology newsletter agreed to signal, each to its recipients, the importance to them of the other's content, inviting them to subscribe to receive it and disseminate it among their contacts. So please visit Vivre par(mi) les écrans and subscribe to the newsletter.
This choice of collaboration stems from the common project of promoting, developing and sharing highly qualified knowledge aimed at creating tools for guidance, critique and intervention in the field of media ecology and our current and future living between(mite) screens, as well as fostering the social dissemination of the aforementioned knowledge and tools.