Natural and artificial intelligence – the psychotechnical agenda of the 21st century

The aim in this paper was to consider some of the fundamental and important differences between natural and artificial intelligence (AI). We are currently amid the third wave of AI technology, and for the first time in human history, it can be said that Large Language Models (LLMs) fulfil the original 1955 definition of AI, stipulating the simulation of human intelligence in a machine. Current AI systems are potentially very useful in many domains, and seem likely to revolutionise the way aspects of human work occur. However, considering their technical limitations, it seems unlikely that current AI will develop exponentially to the equivalent of human level intelligence and beyond. Alternatively, it also seems unlikely that we are heading towards another AI winter. Hence, the future would appear to contain many useful AI tools, but to fully understand and benefit from the third wave of AI we need to be aware that natural and artificial intelligence are in fact very different. Understanding these differences comprises the important psychotechnical agenda of our times. In terms of the Dual Process Theory of human cognition, LLMs are analogous, at best, to system-1 cognitive processing, and it remains unknown how anything approaching system-2 cognition could be incorporated in a machine. In addition, the machine learning that underlies third-wave AI comes with several down sides such as biases drawn from the training data, the propensity to hallucinate, and to produce AI systems that suffer from the black box problem. While many attempts to increase the transparency of AI systems are underway, all third wave AI systems remain black boxes, complicating efforts to establish their safety and reliability, and to fix performance problems when they are identified. Such limitations underscore the importance of retaining a human in the loop, at least in safety critical domains.