China
Africa
Explore chapters and articles related to this topic
An Unexpected Renaissance Age
Published in Alessio Plebe, Pietro Perconti, The Future of the Artificial Mind, 2021
Alessio Plebe, Pietro Perconti
We saw in §4.4.1 that the so-called ‘Bayesian epistemology’ – the idea that our reasoning is probabilistic in accordance to Bayes’ theory – was influential in philosophy at the beginning of the last century. It had a reception in AI too, mainly in the work of Pearl (1986, 1988). In the last decade, Bayesian epistemology has come to be a fundamental reference in cognitive science, thanks to the predictive brain proposal of Karl Friston (2009, 2010, 2012). At the heart of his proposal is a formal expression of free energy, derived from Bayesian variational inference (Friston and Stephan, 2007; Friston and Kiebel, 2009). Variational autoencoders in DL are a precise correlate of Friston’s free-energy principle in the brain and the mathematical formulations are almost the same. Curiously, Kingma & Welling glaringly neglect the connection between their new architecture and its cognitive counterpart, as do Rezende and co-authors. This striking connection is ignored in all further refinement on the variational autoencoder in the DL community, and it is first acknowledged only by Ofner and Stober (2018). This is symptomatic of the difference in attitude of the DL community towards cognitive science, compared to the first generation of artificial neural networks (Perconti and Plebe, 2020). While the primary motivation for the development of the early neural networks was the study of cognition, as described in §4.1.3, the scope of DL has drastically shifted towards engineering goals. Even if several of the protagonists of DL are the same scientists associated with earlier artificial neural networks – Hinton included – the majority of the DL community is totally indifferent to cognitive studies.
The maturing science of consciousness
Published in Journal of the Royal Society of New Zealand, 2023
The British neuroscientist Karl Friston has proposed a controversial answer. According to Friston, when living organisms interact with their environment a quantity known as free energy is minimised (Friston 2010). To take a simple example, a bacterium minimises free energy by moving towards a source of nutrition. Remarkably, it turns out that minimising free energy is mathematically equivalent to minimising prediction error, using the principle discovered by Thomas Bayes. This identity enables Anil Seth to build a bridge between his theory and Friston’s ideas. According to Friston’s free energy principle, minimising free energy is a key characteristic of basic processes, such as growth and metabolism, that are shared by all living organisms. If minimising prediction error and minimising free energy amount to the same thing, then free energy is also minimised by the brain processes that create consciousness. According to Anil Seth and like-minded theorists (Clark 2013), when neural models are created, and updated in response to sensory information, prediction error and free energy are both minimised.
Analyzing trajectories of learning processes through behaviour-based entropy
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2020
Shinsuke Seino, Kenji Kimura, Satoshi Kawamura, Yoshifumi Sasaki, Akira Maruoka
Understanding how the human brain functions is a challenging and difficult problem because the human brain is not only the most complex object; it is a massy product developed through evolution over a long period of time. Friston launched an ambitious and yet controversial research effort to reveal a computational principle underlying brain functions (Friston, 2010). His theory was an attempt to capture how the brain functions, making predictions and then updating them repeatedly based on perceived signals from the environment. In doing so, the brain works to minimize the free energy, which is roughly the quantity that upperbounds the ‘prediction error.’ The unifying principle of minimizing the free energy is called the ‘free energy principle.’ Behavior-based entropy of our study is based on the notion of entropy, whereas the free energy principle also adopts the notion of entropy: behaviour-based entropy is the quantity such that, if a participant ultimately can have full knowledge of the function , then its behaviour-based entropy tends to be zero, whereas the free energy principle indicates to us that the human brain works to minimize the free energy to maintain its states within certain bounds.