Applying Deutsch’s concept of good explanations to artificial intelligence and neuroscience – An initial exploration

Artificial intelligence has made great strides since the deep learning revolution, but AI systems remain incapable of learning principles and rules which allow them to extrapolate outside of their training data to new situations. For inspiration we look to the domain of science, where scientists have been able to develop theories which show remarkable ability to extrapolate and sometimes even predict the existence of phenomena which have never been observed before. According to David Deutsch, this type of extrapolation, which he calls “reach”, is due to scientific theories being hard to vary. In this work we investigate Deutsch’s hard-to-vary principle and how it relates to more formalized principles in deep learning such as the bias-variance trade-off and Occam’s razor. We distinguish internal variability, how much a model/theory can be varied internally while still yielding the same predictions, with external variability, which is how much a model must be varied to predict new, out-of-distribution data. We discuss how to measure internal variability using the notion of the Rashomon set and how to measure external variability using Kolmogorov complexity. We explore what role hard-to-vary explanations play in intelligence by looking at the human brain, the only example of highly general purpose intelligence known. We distinguish two learning systems in the brain – the first operates similar to deep learning and likely underlies most of perception while the second is a more creative system capable of generating hard-to-vary models and explanations of the world. We make contact with Popperian epistemology which suggests that the generation of scientific theories is a not an inductive process but rather an evolutionary process which proceeds through conjecture and refutation. We argue that figuring out how replicate this second system, which is capable of generating hard-to-vary explanations, is a key challenge which needs to be solved in order to realize artificial general intelligence.     

Cognitive and motor deficits in older adults with HIV infection: Comparison with normal ageing and Parkinson’s disease

Despite the life-extending success of antiretroviral pharmacotherapy in HIV infection (HIV), the prevalence of mild cognitive impairment in HIV remains high. Near-normal life expectancy invokes an emerging role for age–infection interaction and a potential synergy between immunosenescence and HIV-related health factors, increasing risk of cognitive and motor impairment associated with degradation in corticostriatal circuits. These neural systems are also compromised in Parkinson’s disease (PD), which could help model the cognitive deficit pattern in HIV. This cross-sectional study examined three groups, age 45–79 years: 42 HIV, 41 PD, and 37 control (CTRL) participants, tested at Stanford University Medical School and SRI International. Neuropsychological tests assessed executive function (EF), information processing speed (IPS), episodic memory (MEM), visuospatial processing (VSP), and upper motor (MOT) speed and dexterity. The HIV and PD deficit profiles were similar for EF, MEM, and VSP. Although only the PD group was impaired on MOT compared with CTRL, MOT scores were related to cognitive scores in HIV but not PD. Performance was not related to depressive symptoms, socioeconomic status, or CD4⁺ T-cell counts. The overlap of HIV-PD cognitive deficits implicates frontostriatal disruption in both conditions. The motor-cognitive score relation in HIV provides further support for the hypothesis that these processes share similar underlying mechanisms in HIV infection possibly expressed with or exacerbated by ageing.     

Hick’s law equivalent for reaction time to individual stimuli

Hick’s law, one of the few law-like relationships involving human performance, expresses choice reaction time as a linear function of the mutual information between the stimulus and response events. However, since this law was first proposed in 1952, its validity has been challenged by the fact that it only holds for the overall reaction time (RT) across all the stimuli, and does not hold for the reaction time (RT_i) for each individual stimulus. This paper introduces a new formulation in which RT_i is a linear function of (1) the mutual information between the event that stimulus i occurs and the set of all potential response events and (2) the overall mutual information for all stimuli and responses. Then Hick’s law for RT follows as the weighted mean of each side of the RT_i equation using the stimulus probabilities as the weights. The new RT_i equation incorporates the important speed–accuracy trade-off characteristic. When the performance is error-free, RT_i becomes a linear function of two entropies as measures of stimulus uncertainty or unexpectancy. Reanalysis of empirical data from a variety of sources provide support for the new law-like relationship.