summary: New research focuses on examining the possibility of consciousness in artificial systems and ruling out scenarios where an AI appears conscious without actually being conscious. Using the free energy principle, the study highlights that although some biological information processing can be simulated by computers, differences in the causal structure of the brain and computers may be important for consciousness. This approach aims to prevent the accidental creation of artificial consciousness and to mitigate deception by AI that appears conscious.
Key Facts:
- Wanja Wiese's research uses the free energy principle to explore consciousness in AI.
- He suggests that differences in the causal structure between brains and computers may have crucial implications for consciousness.
- The goal is to prevent artificial consciousness and mitigate deception by AI that appears conscious.
sauce: Rub
When considering the possibility of consciousness in artificial systems, there are at least two different approaches.
One approach asks how likely it is that current AI systems are to become conscious, and what needs to be added to existing systems to increase the chances of them becoming conscious.
Another approach raises the question of what types of AI systems are unlikely to be conscious, and how we can rule out the possibility that certain types of systems might be conscious.
Wanja Wiese pursues the second approach in her research: “My aim is to contribute to two goals: first, to reduce the risk of inadvertently creating an artificial consciousness. This is a desirable outcome, since at present it is unclear under what conditions it would be morally permissible to create an artificial consciousness.”
“Second, this approach should help rule out deception by ostensibly conscious AI systems that only appear to be conscious,” he explains.
This is especially important because there are already indications that many people who frequently interact with chatbots believe these systems are conscious, even as the expert consensus is that current AI systems are not conscious.
Free Energy Principle
“For example, how can we test whether there are necessary conditions for consciousness that are not satisfied by classical computers?” Wiese asks in his essay. The common feature of all conscious animals is that they are alive.
However, being alive is such a demanding condition that many people don't think of it as a reasonable candidate as a necessary condition for consciousness, but perhaps some of the conditions necessary for being alive are also necessary for consciousness.
In her article, Wanja Wiese refers to British neuroscientist Carl Friston's free energy principle, which suggests that the processes that ensure the continued existence of self-organizing systems like living organisms can be described as a type of information processing.
In humans, these include processes that regulate important parameters such as body temperature, oxygen content in the blood, blood sugar levels, etc. The same type of information processing can also be achieved by a computer, except that the computer does not regulate body temperature or blood sugar levels, it only simulates these processes.
Most differences have nothing to do with consciousness
Researchers suggest that the same may be true for consciousness. If we assume that consciousness contributes to the survival of a conscious organism, then the free energy principle dictates that the physiological processes that contribute to the maintenance of the organism should bear the traces left by conscious experience and be explainable as information-processing processes. This can be called the “computational correlate of consciousness.”
This too can be achieved by a computer, although additional conditions may need to be met within the computer in order for it to reproduce, rather than just simulate, conscious experience.
So in her article, Wanja Wiese analyses the difference between how a conscious organism realizes the computational counterpart of consciousness and how a computer achieves it in a simulation.
He argues that most of these differences are irrelevant to consciousness – for example, unlike electronic computers, our brains are extremely energy efficient – but it is hard to believe that this is a requirement for consciousness.
But another difference lies in the causal structure of computers and the brain: in a classical computer, data is always first loaded from memory, then processed by the central processing unit, and finally stored back into memory.
In the brain, no such separation exists, meaning that the causal connections in different brain regions take different forms, and Wanja Wiese argues that this may be a relevant difference between the brain and conventional computers for consciousness.
“In my view, the perspective provided by the free energy principle is particularly interesting because it allows us to describe the properties of conscious organisms in a way that is in principle realizable in artificial systems but does not exist in many types of artificial systems (e.g. computer simulations),” explains Wanja Wiese.
“This means we can capture the prerequisites for consciousness in artificial systems in much greater detail and precision.”
About this consciousness and AI research news
author: Mike Driessen
sauce: Rub
contact: Mike Driessen – RUB
image: Image courtesy of Neuroscience News
Original Research: Open access.
“Artificial Consciousness: A Free Energy Perspective” by Wanja Wiese. Philosophical Studies
Abstract
Artificial Consciousness: A Free Energy Perspective
Does the assumption of a weak form of computational functionalism, that a right form of neural computation is sufficient for consciousness, mean that a digital computational simulation of such neural computation could be conscious, or does this computational simulation need to be implemented in the right way in order to replicate consciousness?
From the perspective of Carl Friston's free energy principle, self-organizing systems (such as living organisms) share a set of properties that can be realized in artificial systems but not in computers with classical (von Neumann) architecture.
I argue that at least one of these properties, namely some kind of causal flow, can be used to distinguish systems that merely simulate from those that actually replicate consciousness.
