Computational Agents, Design and Innovative Behaviour: Hetero Oeconomicus
This paper is closed for comments.
Far too long economic stories treat of perfectly informed as well as fully rational optimization within a purely materialistic world. No wonder there is a lack of evidence and explanations consistent with regard to the subject of interest: human decision makers and entrepreneurs revolutionizing the decision space. Strands like game theory and institutional economics partly already take a more practical view. Evolutionary and behavioural economics finally establish the necessary link to other disciplines like psychology and informational science. This paper recaps selected parts of the literature in favour of a conceptional view on computational agents. The latter, first, invites economic modellers to question and argue the microfoundation of their assumptions with regard to the individual or aggregate level of human behaviour they truly refer to. Secondly, the design serves to illustrate the potential as well as the limitations computational agents exhibit with regard to the incorporation of creativity as the main source for innovative behaviour. Thirdly, the rather superficial collection of ideas serves as position paper for future approaches.
Comment: Scheuer on Computational Agents
John Davis, Marquette University and University of Amsterdam
Timon Scheuer offers us an interesting interpretation of Jason Potts’ Hetero economicus idea. Building on Herbert Simon’s characterization of how agents deliberate, he recommends we adopt a computational model of the human organism. Central to this is what he calls the agent’s “operating structure” that is diagramed in his Figure 1. Scheuer then explores the motivational foundations of such agents, and emphasizes that agents act in environments involving “unknown circumstances” and exhibiting “unforeseeable dynamics.” He emphasizes that central to this conception is that agents are creative: “With regard to computational agents creativity then seems to be an operation itself that alternates and especially extends the set of so far imaginable operations” (Scheuer, 2017, p. 10). How might we understand this emphasis on creativity?
I take the idea of “imaginable operations” to be the idea of possible operations, and that what Scheuer is recommending is that we conceptualize the agent’s “operating structure” in terms of possibilities. I believe that this points us toward the difference between classical logic and modal logic, where the former excludes and the latter includes modal adverbs, or operators as they are called in logic, such as ‘possibly’ and ‘necessarily’ – which get expressed in natural languages in the subjunctive rather than indicative mood with terms such as ‘might’, ‘could’, ‘should’, etc. Modal logic, a 20th century development is often associated with ‘possible worlds’ thinking – what would be true were such-and-such possibly the case – and ‘counterfactuals’ – states of affairs we can imagine that are contrary to fact.
The traditional Homo economicus with its specific axiomatic foundations regarding preferences, then, is formulated strictly in classical logic terms in which possibility does not exist, allowing agents’ behavior to be deductively derived as optimizing. That is, neoclassicism, logically speaking, avoids the domain of natural language (and the more complicated modal logics developed to explain it), leaving it with an artificially regimented agent conception of little value for the world we live in. In contrast, Scheuer’s creative computational agents must contend with multiple possible worlds (because the future is open under the sway of possibility), and deliberate in a counterfactual way about how they would act depending on what might possibly be the case.
Post-Keynesians have long argued that the world is open by emphasizing true uncertainty, but have not developed a conception of the agent as open. I suggested that Keynes’ agents should be treated as reflexive agents (Davis 2017), that is, always adapting to unmet goals, and offered a simple logical analysis of how a changing (non-ergodic) world continually changes the circumstances of agent behavior. Scheuer’s computational agent idea would expand on this were his agents’ “operating structures” framed in terms of the ‘possible worlds’ agents imagine they might encounter. This framework is an evolutionary one in that the pathways agents follow determine the arrays of ‘possible worlds’ they subsequently encounter. This would also be consistent with Brian Arthur’s (2009) combinatorial understanding of evolution and technology change, which makes human need central as does Scheuer.
Arthur, W. Brian (2009) The Nature of Technology, New York: Free Press.
Davis, John (2017) “The Continuing Relevance of Keynes’s Philosophical Thinking: Reflexivity, Complexity, and Uncertainty,” The Annals of the Fondazione Luigi Einaudi. An Interdisciplinary Journal of Economics, History and Political Science, LI 1 (2017) 55-75; also at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2881573
This paper discusses if and how agent based models can be used for studying innovative behaviour. According to the author, agents should be designed using insight from psychology and behavioural economics. He proposes to look beyond neoclassical models and tries to define some principles for building psychologically founded agent based models.
This is a useful idea, even if not so new. Herbert Simon, before others, did something similar, mixing computer simulation based on a deep understanding of human behaviour (Novarese, 2004 https://econpapers.repec.org/paper/wpawuwpmh/0412001.htm).
Innovation is a very general term and maybe it should be better defined. A model of innovation can have different aims and according to its aim, creativity should be treated in many possible ways. Even behavioural models are tools and cannot deal with all of reality.
I’m not sure if a model dealing with innovation should necessarily be able to innovate. One of the main problems and possible limitations of the proposed approach is, yet, the possibility of having artificial agents capable of creating something new. Can novelty arise from programmed algorithms? According to the line of research best identified by Amabile, quoted in the paper, innovation can hardly be modeled, as it is linked with the creation of something new and non-existent.
But this is only one side of the discussion. There is another stream of research that fits better with agent based models. The same Herbert Simon can again be recalled. In his view innovation is nothing but problem solving and can, therefore, be performed by artificial agent.
Even if we look at Arthur (2009, The Nature of Technology) we find a picture of innovation and creativity which could possibly be modelled. Technologies arise and evolve and have to be seen as emerging, combinatorial, complex, social, cumulative, and at least partly unplanned phenomena. Technologies arise and evolve combining different pieces and using new scientific development. They do not appear from nowhere and as such they can be, eventually, modelled.
Besides, agent based model allows emergent and unexpected solutions (Terna, 2013 http://www.spaziofilosofico.it/wp-content/uploads/2013/01/Terna-English.pdf). Something unplanned or unforeseeable, even if not completely new, can be seen as innovation.