Lake and his colleagues have chosen to build “machines that learn and think like people.” I beg to differ. Or perhaps it is a matter of what one means by learning and thinking like people. Permit me to explain. Early in the history of artificial intelligence (AI) and simulation research, investigators began following two different roads. The roads might potentially have converged, but it has become more and more apparent from recent events that they have actually diverged.

One road was initiated by pioneers like Newell et al. (Reference Newell, Shaw and Simon1957), Winograd (Reference Winograd1972), Minsky and Papert (Reference Minsky and Papert1987), Minsky (Reference Minsky2003), and Feigenbaum and Feldman (Reference Feigenbaum and Feldman1995). This road was based on understanding people's competencies in learning and thinking. Investigators taking this road studied causal reasoning, game playing, language acquisition, intuitive physics, and people's understanding of block worlds. Today, Anderson's ACT-R perhaps provides the most comprehensive simulation model (Anderson et al. Reference Anderson, Bothell, Byrne, Douglass, Lebiere and Qin2004).

A second road was taken by pioneers like Colby (Reference Colby1975) with PARRY, a computer program simulating a paranoid, Abelson and Carroll (Reference Abelson and Carroll1965) with their True Believer program, and Weizenbaum (Reference Weizenbaum1966) with his ELIZA non-directive psychotherapy program. The idea in this research was to understand people's often suboptimal performances in learning and thinking. These programs recognized that people are often emotional, a-rational, and function at levels well below their capabilities.

Many of these ideas have been formalized in recent psychological research. For example, Stanovich (Reference Stanovich2009) has shown that rationality and intelligence are largely distinct. Mayer and Salovey (Reference Mayer and Salovey1993) have shown the importance of emotional intelligence to people's thinking, and Sternberg (Reference Sternberg1997) has argued both for the importance of practical intelligence and for its relative independence from analytical or more academic aspects of intelligence.

The two roads of AI/simulation research might have converged with comprehensive models that comfortably incorporate aspects of both optimal and distinctly suboptimal performance. They haven't. At the time, Abelson, Colby, and others worked on their models of what was at best a-rational, and at worst wholly irrational, thinking. The work seemed a bit quirky and off the beaten track – perhaps a road not worth following very far. That was then

The 2016 presidential election has upended any assumption that everyday people think along the lines that Lake and his colleagues have pursued. Whether one is a Republican or a Democrat, it would be hard to accept this election process as representing anything other than seriously deficient and even defective thinking. The terms learning and thinking seem almost too complimentary to describe what went on. To some people the 2016 election was a frightening portent of a dystopia to come.

The first road, that of Lake et al., is of human cognition divorced from raw emotions, of often self-serving motivations and illogic that characterize much of people's everything thinking. On this view, people are more or less rational “machines.” One might think that it is only stupid people (Sternberg Reference Sternberg2002; 2004) who think and act foolishly. But smart people are as susceptible to foolish thinking as are not so smart people, or even more susceptible, because they do not realize they can think and act foolishly.

The United States, and indeed the world, seems to be entering a new and uncharted era of populism and appeals by politicians not to people's intellects, but to their basest emotions. Unless our models of learning and thinking help us understand how those appeals can succeed, and how we can counter them and help people become wiser (Sternberg & Jordan Reference Sternberg and Jordan2005), the models we create will be academic, incomplete, and, at worst, wrong-headed. The field came to a fork in the road and took it, but to where?