The rapid development of artificial intelligence (AI) and machine learning has led to various parallels between human neural interaction and machine-machine interaction. This essay explores the similarities and differences between these two forms of interaction and highlights the implications for our understanding of intelligence and cognition.
1. Anatomical and physiological analogies
Although machines have no direct biological equivalent to human anatomy and physiology, interesting parallels can be drawn:
Intra-individual level:
LeCun et al (2015) discuss how different artificial neural network (ANN) architectures mimic aspects of human visual processing [1]. The structure of an ANN can be viewed as an analogue of brain anatomy, with different layers and connections representing different functions of the human brain.
Inter-individual level:
In distributed AI systems, the communication protocols between the systems can be seen as analogous to neural communication between brains. Foerster et al. (2016) show how multiple AI agents can cooperate through communication [2], which has similarities to interpersonal communication and cooperation.
2. The connectome: natural vs. artificial
The human connectome, the totality of neuronal connections in the brain, has no direct equivalent in machines. Nevertheless, there are remarkable similarities:
Sporns et al (2005) describe the human connectome as a complex network with small-world properties [3]. These properties enable efficient information processing and integration in the human brain.
In AI systems, the weights and connections in neural networks can be regarded as a kind of ‘artificial connectome’. Karpathy et al. (2015) visualise such networks and show how they form complex structures [4] that are similar in function to the human connectome.
3. Cortical sources and activation patterns
While machines do not produce EEG signals, parallels can be drawn in activation and signal processing:
Michel and Murray (2012) discuss how EEG signals arise from the activity of cortical sources in the human brain [5]. These complex patterns of neuronal activity are fundamental to human cognition.
In AI systems, the activation patterns in different layers of a neural network can be considered analogous to cortical sources. Zeiler and Fergus (2014) visualise these activations and show how they contribute to the outputs of the network [6]. These visualisations offer insights into the ‘thought processes’ of artificial neural networks.
4. Topological influences
The topology of neural networks has significant effects in both biological and artificial systems:
Bullmore and Sporns (2009) discuss how the topology of the human brain influences its function [7]. The specific structure and organisation of neuronal connections enables the complex cognitive abilities of humans.
In AI systems, He et al. (2016) demonstrate with their ResNet architecture how the topology of the network can drastically improve its performance [8]. This finding emphasises the importance of the network structure for the performance of artificial intelligence.
5. Forms of synchronisation
Synchronisation plays a central role in both human and machine interactions:
In human interactions, synchronisation manifests itself in the form of synchronised brain activity, movements and behaviour, as discussed in the previous sections.
In distributed AI systems, we can observe synchronisation in the form of consensus algorithms. Olfati-Saber et al (2007) discuss how distributed systems can reach consensus [9], which has similarities to consensus finding in human groups.
6. Measurement and evaluation of interactions
The measurement and evaluation of interactions differs significantly between human and machine systems, but there are also similarities:
In human interactions, physiological measurements (e.g. EEG, fMRI) and behavioural observations are often combined to assess the quality and effectiveness of the interaction.
Various aspects of machine-machine interactions can be measured, including data exchange, synchronisation, performance and emergence. Stone et al. (2016) discuss methods for measuring and evaluating AI systems that are also applicable to machine-machine interactions [10].
Conclusion and implications
Although there are fundamental differences between human brains and machines, there are fascinating parallels in the structure, function and interaction of these systems. The analogies between the human connectome and the structure of artificial neural networks, as well as between cortical activation patterns and the activations in AI systems, offer valuable insights into the functioning of both systems.
Exploring these parallels has important implications for the development of AI systems that can interact more effectively with each other and with humans. It could lead to new approaches in human-machine interaction and deepen our understanding of collective intelligence and emergence.
Furthermore, this research raises profound philosophical and ethical questions. If machines increasingly show human-like patterns of interaction, what does this mean for our understanding of consciousness, intelligence and even humanity? How can we ensure that AI systems that interact with each other act ethically and respect human values?
Future research should focus on further investigating the differences and similarities between human and machine interactions. This could lead to new insights into the nature of intelligence and cognition and potentially pave the way for a new generation of AI systems that are better able to interact and co-operate with humans.
References:
[1] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.
[2] Foerster, J., Assael, I. A., de Freitas, N., & Whiteson, S. (2016). Learning to communicate with deep multi-agent reinforcement learning. Advances in Neural Information Processing Systems, 29.
[3] Sporns, O., Tononi, G., & Kötter, R. (2005). The human connectome: A structural description of the human brain. PLoS Computational Biology, 1(4), e42.
[4] Karpathy, A., Johnson, J., & Fei-Fei, L. (2015). Visualizing and understanding recurrent networks. arXiv preprint arXiv:1506.02078.
[5] Michel, C. M., & Murray, M. M. (2012). Towards the utilization of EEG as a brain imaging tool. Neuroimage, 61(2), 371-385.
[6] Zeiler, M. D., & Fergus, R. (2014). Visualizing and understanding convolutional networks. In European Conference on Computer Vision (pp. 818-833). Springer, Cham.
[7] Bullmore, E., & Sporns, O. (2009). Complex brain networks: graph theoretical analysis of structural and functional systems. Nature Reviews Neuroscience, 10(3), 186-198.
[8] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 770-778).
[9] Olfati-Saber, R., Fax, J. A., & Murray, R. M. (2007). Consensus and cooperation in networked multi-agent systems. Proceedings of the IEEE, 95(1), 215-233.
[10] Stone, P., Brooks, R., Brynjolfsson, E., Calo, R., Etzioni, O., Hager, G., … & Teller, A. (2016). Artificial intelligence and life in 2030. One Hundred Year Study on Artificial Intelligence: Report of the 2015-2016 Study Panel, 52.