Lake et al. underrate both the promise and the limitations of contemporary deep learning techniques. The promise lies in combining those techniques with broad multisensory training as experienced by infants and children. The limitations lie in the need for such systems to possess functional subsystems that generate, monitor, and switch goals and strategies in the absence of human intervention.

Human reasoning is richer than Lake et al. acknowledge, and the emphasis on theories of how images and scenes are synthesized is misleading. For example, the world knowledge used in vision presumably involves a combination of geometric, physical, and other knowledge, rather than just a causal theory of how the image was produced. In physical reasoning, a model can be a set of constraints rather than a physics engine. In intuitive psychology, many inferences proceed without detailed causal generative models. How humans reliably perform such inferences, often in the face of radically incomplete information, remains a mystery.

We comment on ways in which Lake et al. advance our understanding of the machinery of intelligence and offer suggestions. The first set concerns animal-level versus human-level intelligence. The second concerns the urgent need to address ethical issues when evaluating the state of artificial intelligence.

We agree with Lake et al.'s trenchant analysis of deep learning systems, including that they are highly brittle and that they need vastly more examples than do people. We also agree that human cognition relies heavily on structured relational representations. However, we differ in our analysis of human cognitive processing. We argue that (1) analogical comparison processes are central to human cognition; and (2) intuitive physical knowledge is captured by qualitative representations, rather than quantitative simulations.

Lake et al. offer a timely critique on the recent accomplishments in artificial intelligence from the vantage point of human intelligence and provide insightful suggestions about research directions for building more human-like intelligence. Because we agree with most of the points they raised, here we offer a few points that are complementary.

Reverse engineering human cognitive processes may improve artificial intelligence, but this approach implies we have little to learn regarding brains from human-engineered systems. On the contrary, engineered technologies of dynamic network communication have many features that highlight analogous, poorly understood, or ignored aspects of brain and cognitive function, and mechanisms fundamental to these technologies can be usefully investigated in brains.

Lake et al. discuss building blocks of human intelligence that are quite different from those of artificial intelligence. We argue that a theory of human intelligence has to incorporate human motivations and emotions. The interaction of motivation, emotion, and cognition is the real strength of human intelligence and distinguishes it from artificial intelligence.

Lake et al. propose that people rely on “start-up software,” “causal models,” and “intuitive theories” built using compositional representations to learn new tasks more efficiently than some deep neural network models. We highlight the many drawbacks of a commitment to compositional representations and describe our continuing effort to explore how the ability to build on prior knowledge and to learn new tasks efficiently could arise through learning in deep neural networks.

Building machines that learn and think like humans is essential not only for cognitive science, but also for computational neuroscience, whose ultimate goal is to understand how cognition is implemented in biological brains. A new cognitive computational neuroscience should build cognitive-level and neural-level models, understand their relationships, and test both types of models with both brain and behavioral data.

Although we agree with Lake et al.'s central argument, there are numerous flaws in the way people use causal models. Our models are often incorrect, resistant to correction, and applied inappropriately to new situations. These deficiencies are pervasive and have real-world consequences. Developers of machines with similar capacities should proceed with caution.

Physical competence is acquired through animals' embodied interaction with their physical environments, and psychological competence is acquired through situated interaction with other agents. The acquired neural models essential to these competencies are implicit and permit more fluent and nuanced behavior than explicit models. The challenge is to understand how such models are acquired and used to control behavior.

Lake et al. suggest that current AI systems lack the inductive biases that enable human learning. However, Lake et al.'s proposed biases may not directly map onto mechanisms in the developing brain. A convergence of fields may soon create a correspondence between biological neural circuits and optimization in structured architectures, allowing us to systematically dissect how brains learn.

Lake et al. proposed a way to build machines that learn as fast as people do. This can be possible only if machines follow the human processes: the perception-action loop. People perceive and act to understand new objects or to promote specific behavior to their partners. In return, the object/person provides information that induces another reaction, and so on.

The argument by Lake et al. to create more human-like robots is, first, implausible and, second, undesirable. It seems implausible to me that a robot might have friends, fall in love, read Foucault, prefer Scotch to Bourbon, and so on. It seems undesirable because we already have 7 billion people on earth and don't really need more.

Autonomous lifelong development and learning are fundamental capabilities of humans, differentiating them from current deep learning systems. However, other branches of artificial intelligence have designed crucial ingredients towards autonomous learning: curiosity and intrinsic motivation, social learning and natural interaction with peers, and embodiment. These mechanisms guide exploration and autonomous choice of goals, and integrating them with deep learning opens stimulating perspectives.

Artificial intelligence algorithms seek inspiration from human cognitive systems in areas where humans outperform machines. But on what level should algorithms try to approximate human cognition? We argue that human-like machines should be designed to make decisions in transparent and comprehensible ways, which can be achieved by accurately mirroring human cognitive processes.

The search for a deep, multileveled understanding of human intelligence is perhaps the grand challenge for 21st-century science, with broad implications for technology. The project of building machines that think like humans is central to meeting this challenge and critical to efforts to craft new technologies for human benefit.

Machines that learn and think like people should simulate how people really think in their everyday lives. The field of artificial intelligence originally traveled down two roads, one of which emphasized abstract, idealized, rational thinking and the other, which emphasized the emotionally charged and motivationally complex situations in which people often find themselves. The roads should have converged but never did. That's too bad.

Machines that learn and think like people must be able to learn from others. Social learning speeds up the learning process and – in combination with language – is a gateway to abstract and unobservable information. Social learning also facilitates the accumulation of knowledge across generations, helping people and artificial intelligences learn things that no individual could learn in a lifetime.

Lake et al. point out that grounding learning in general principles of embodied perception and social cognition is the next step in advancing artificial intelligent machines. We suggest it is necessary to go further and consider lifelong learning, which includes developmental learning, focused on embodiment as applied in developmental robotics and neurorobotics, and crossmodal learning that facilitates integrating multiple senses.