The Quest for Artificial Consciousness: Can Machines Truly Think and Feel?

Artificial consciousness (AC), the ability of machines to think and feel like humans, is a highly debated topic. Rapid progress in the field of AI fuels discussions about its potential reality. Researchers, inspired by human consciousness, design AI with ever-growing capabilities. The study „Consciousness in Artificial Intelligence“ states that while consciousness is not present in current AI, it is not ruled out either. Some neuroscientific models of consciousness, such as the global workspace theory, come closer to this idea. This shift has the potential to fundamentally change the use of AI, including its role in the world of AI art (AIART).

Artificial consciousness is a hypothetical state in which a machine could perceive, think, and feel just as a human does. It’s a complex and controversial topic that scientists and philosophers have been discussing for decades. The possibility of realizing artificial consciousness or a computational system that exhibits certain consciousness-like characteristics in technologies utilizing artificial intelligence is gaining importance due to rapid progress in this field. Researchers are inspired by characteristics typical of human consciousness when developing AI, indicating potential for further development of AI capabilities and possibilities. Artistic strategies that operate with AI as a partner to some extent respond to these tendencies and integrate them into practice. Meanwhile, the increasing number of AI systems capable of convincingly imitating human conversation, evaluating probable solutions, or generating various forms of artistic or cultural artifacts, previously only human domains (Manovich & Arielli, 2019), increases the likelihood that people will consider these systems as conscious entities.

Drawing on the study „Consciousness in Artificial Intelligence: Insights from the Science of Consciousness“ [2], which suggests methods for evaluating CAI through systematic research, it is possible to state in some respects that artificial consciousness in artificial intelligence is more real than unreal. The study introduced a methodology for investigating the consciousness of artificial intelligence CAI, based on three key principles. The first principle is based on the adoption of computational functionalism (CF) as a fundamental working hypothesis. The central thesis admits that CAI is fundamentally possible, which also means that by studying the functioning of AI systems, it can be determined whether the examined systems are likely conscious. The second principle involves the use of neuroscientific theories of consciousness, which provide significant empirical support and can be applied to evaluating CAI. Such theories focus on identifying functions sufficient for human consciousness, on which computational functionalism relies. The third principle is based on examining whether AI systems perform functions similar to those that scientific theories associate with consciousness, and then assigning a credibility rating based on (a) similarity of functions, (b) strength of evidence for the subject theories, and (c) belief in computational functionalism.

The cited study took a grounded approach to CAI in evaluating existing AI systems with respect to neuroscientific theories of consciousness, such as: the theory of recurrent processing (RPT) [Lamme, 2006, 2010, 2020], the global workspace theory (GWT) [Baars, 1988; Dehaene et al., 2001; Dehaene & Changeux, 2011; 9; 10], the higher-order theory of consciousness (HOT) [Rosenthal & Weisberg, 2008], the predictive processing theory (PP) [12–17], and the attention schema theory (AST) [Chalmers, 2018; Graziano & Webb, 2015]. From these theories also stem the indicators of evaluation, „indicative properties“ of consciousness, elucidated in computational terms, which allow for assessing AI systems. The analysis demonstrated that CAI is unlikely to be found in current AI systems, but on the other hand, there are no apparent technical obstacles to building AI systems that meet these indicators, thus CAI as well.

Google Gemini can be understood as a family of multimodal language models introduced in 2023. Gemini is designed to process and understand information from various sources, including text, images, and video. This capability of multimodality makes them suitable for a wide range of tasks. The computational principles on which Gemini is based closely correlate with models in the cited study. Opening the discourse on large language models and the Perceiver architecture [Jaegle et al. 2021a, b], in relation to the global workspace theory and systems like PaLM-E [20], „virtual rodent“ [21], and AdA (DeepMind Adaptive Agents Team 2023), „embodied agents“, may suggest that CAI is still only in the early stages of development in current and forthcoming AI systems. While some systems, such as PaLM-E and AdA, exhibit certain features of consciousness, such as the ability to plan, make decisions, and respond to their environment, they remain very primitive elements of consciousness. Moreover, it appears that these systems do not actually have any deeper self-reflection or understanding of themselves.

However, this does not mean that it will not be possible to develop states in Gemini systems that exhibit certain characteristics of consciousness in the future. The advent of Gemini as a multimedia language model already represents, or can represent, a fundamental shift in the use of this tool even in the world of AIART. Gemini is known for its ability to generate accurate and informative text. It can, for example, respond to your questions about factual topics or generate guides. However, this does not compare older systems like PaLM-E (also known as PaLM embodied), but rather focuses on the use of Gemini as a potential partner for AIART.

Author: Tomáš Marušiak, 2024


1.            MANOVICH, Lev and ARIELLI, Emanuele. Artificial Aesthetics: A Critical Guide to AI, Media and Design. 2019.

2.            BUTLIN, Patrick, LONG, Robert, ELMOZNINO, Eric, BENGIO, Yoshua, BIRCH, Jonathan, CONSTANT, Axel, DEANE, George, FLEMING, Stephen M., FRITH, Chris, JI, Xu, KANAI, Ryota, KLEIN, Colin, LINDSAY, Grace, MICHEL, Matthias, MUDRIK, Liad, PETERS, Megan A. K., SCHWITZGEBEL, Eric, SIMON, Jonathan and VANRULLEN, Rufin. Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. . 16 August 2023. Whether current or near-term AI systems could be conscious is a topic of scientific interest and increasing public concern. This report argues for, and exemplifies, a rigorous and empirically grounded approach to AI consciousness: assessing existing AI systems in detail, in light of our best-supported neuroscientific theories of consciousness. We survey several prominent scientific theories of consciousness, including recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory. From these theories we derive “indicator properties” of consciousness, elucidated in computational terms that allow us to assess AI systems for these properties. We use these indicator properties to assess several recent AI systems, and we discuss how future systems might implement them. Our analysis suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.

3.            LAMME, Victor A.F. Towards a true neural stance on consciousness. Trends in Cognitive Sciences. November 2006. Vol. 10, no. 11, p. 494–501. DOI 10.1016/j.tics.2006.09.001.

4.            LAMME, Victor A. F. How neuroscience will change our view on consciousness. Cognitive Neuroscience. 18 August 2010. Vol. 1, no. 3, p. 204–220. DOI 10.1080/17588921003731586.

5.            LAMME, Victor A. F. Visual Functions Generating Conscious Seeing. Frontiers in Psychology. 14 February 2020. Vol. 11. DOI 10.3389/fpsyg.2020.00083.

6.            BAARS, Bernard J. A Cognitive Theory of Consciousness. . Cambridge University Press, 1988.

7.            DEHAENE, Stanislas, KERSZBERG, Michel and CHANGEUX, Jean-Pierre. A neuronal model of a global workspace in effortful cognitive tasks. Pnas. 2001. Vol. 95, no. 24.

8.            DEHAENE, Stanislas and CHANGEUX, Jean-Pierre. Experimental and Theoretical Approaches to Conscious Processing. Neuron. April 2011. Vol. 70, no. 2, p. 200–227. DOI 10.1016/j.neuron.2011.03.018.

9.            DEHAENE, Stanislas, KERSZBERG, Michel and CHANGEUX, Jean-pierre. A Neuronal Model of a Global Workspace in Effortful Cognitive Tasks. Proceedings of the National Academy of Sciences of the United States of America. 1998. Vol. 95, no. 24, p. 14529–14534.

10.         MASHOUR, George A., ROELFSEMA, Pieter, CHANGEUX, Jean-Pierre and DEHAENE, Stanislas. Conscious Processing and the Global Neuronal Workspace Hypothesis. Neuron. March 2020. Vol. 105, no. 5, p. 776–798. DOI 10.1016/j.neuron.2020.01.026.

11.         ROSENTHAL, David and WEISBERG, Josh. Higher-order theories of consciousness. Scholarpedia. 2008. Vol. 3, no. 5, p. 4407. DOI 10.4249/scholarpedia.4407.

12.         SETH, Anil. Being you. . London, England : Dutton, 2021. ISBN 9781524742874.

13.         SETH, Anil K. and BAYNE, Tim. Theories of consciousness. Nature Reviews Neuroscience. 3 July 2022. Vol. 23, no. 7, p. 439–452. DOI 10.1038/s41583-022-00587-4.

14.         SETH, Anil K and HOHWY, Jakob. Predictive processing as an empirical theory for consciousness science. Cognitive Neuroscience. 3 April 2021. Vol. 12, no. 2, p. 89–90. DOI 10.1080/17588928.2020.1838467.

15.         DEANE, George. Consciousness in active inference: Deep self-models, other minds, and the challenge of psychedelic-induced ego-dissolution. Neuroscience of Consciousness. 1 September 2021. Vol. 2021, no. 2. DOI 10.1093/nc/niab024.

16.         HOHWY, Jakob. Conscious Self-Evidencing. Review of Philosophy and Psychology. 5 December 2022. Vol. 13, no. 4, p. 809–828. DOI 10.1007/s13164-021-00578-x.

17.         NAVE, Kathryn, DEANE, George, MILLER, Mark and CLARK, Andy. Expecting some action: Predictive Processing and the construction of conscious experience. Review of Philosophy and Psychology. 10 December 2022. Vol. 13, no. 4, p. 1019–1037. DOI 10.1007/s13164-022-00644-y.

18.         GRAZIANO, Michael S. A. and WEBB, Taylor W. The attention schema theory: a mechanistic account of subjective awareness. Frontiers in Psychology. 23 April 2015. Vol. 06. DOI 10.3389/fpsyg.2015.00500.

19.         CHALMERS, David. The Meta-Problem of Consciousness. Journal of Consciousness Studies. 2018. Vol. 25, no. 9–10.

20.         DRIESS, Danny, XIA, Fei, SAJJADI, Mehdi S. M., LYNCH, Corey, CHOWDHERY, Aakanksha, ICHTER, Brian, WAHID, Ayzaan, TOMPSON, Jonathan, VUONG, Quan, YU, Tianhe, HUANG, Wenlong, CHEBOTAR, Yevgen, SERMANET, Pierre, DUCKWORTH, Daniel, LEVINE, Sergey, VANHOUCKE, Vincent, HAUSMAN, Karol, TOUSSAINT, Marc, GREFF, Klaus, ZENG, Andy, MORDATCH, Igor and FLORENCE, Pete. PaLM-E: An Embodied Multimodal Language Model. . 6 March 2023. Large language models excel at a wide range of complex tasks. However, enabling general inference in the real world, e.g., for robotics problems, raises the challenge of grounding. We propose embodied language models to directly incorporate real-world continuous sensor modalities into language models and thereby establish the link between words and percepts. Input to our embodied language model are multi-modal sentences that interleave visual, continuous state estimation, and textual input encodings. We train these encodings end-to-end, in conjunction with a pre-trained large language model, for multiple embodied tasks including sequential robotic manipulation planning, visual question answering, and captioning. Our evaluations show that PaLM-E, a single large embodied multimodal model, can address a variety of embodied reasoning tasks, from a variety of observation modalities, on multiple embodiments, and further, exhibits positive transfer: the model benefits from diverse joint training across internet-scale language, vision, and visual-language domains. Our largest model, PaLM-E-562B with 562B parameters, in addition to being trained on robotics tasks, is a visual-language generalist with state-of-the-art performance on OK-VQA, and retains generalist language capabilities with increasing scale.

21.         MEREL, Josh, ALDARONDO, Diego, MARSHALL, Jesse, TASSA, Yuval, WAYNE, Greg and ÖLVECZKY, Bence. Deep neuroethology of a virtual rodent. . 21 November 2019. Parallel developments in neuroscience and deep learning have led to mutually productive exchanges, pushing our understanding of real and artificial neural networks in sensory and cognitive systems. However, this interaction between fields is less developed in the study of motor control. In this work, we develop a virtual rodent as a platform for the grounded study of motor activity in artificial models of embodied control. We then use this platform to study motor activity across contexts by training a model to solve four complex tasks. Using methods familiar to neuroscientists, we describe the behavioral representations and algorithms employed by different layers of the network using a neuroethological approach to characterize motor activity relative to the rodent’s behavior and goals. We find that the model uses two classes of representations which respectively encode the task-specific behavioral strategies and task-invariant behavioral kinematics. These representations are reflected in the sequential activity and population dynamics of neural subpopulations. Overall, the virtual rodent facilitates grounded collaborations between deep reinforcement learning and motor neuroscience.