Emergence of Belief Systems and the Future of Artificial Intelligence
In the Meaningful-Based Cognitive Architecture (MBCA), every evaluation cycle the input sensory vector is propagated through a hierarchy of Hopfield-like Network (HLN) functional groups, and subsymbolic processing of the input vector occurs. The processed sensory input vector is also propagated to causal groups of HLNs and/or logic/working memory groups of HLNs, where it is matched against the vector from the MBCA’s intuitive and learned logic, physics, psychology and goal planning—collectively forming its world model or belief system. The output of the logic/working memory can be propagated back to the start of the subsymbolic system where it can be the input for the next evaluation cycle, and allows multiple symbolic processing steps to occur on intermediate results. In this fashion, full symbolic causal processing of the input sensory vector occurs. In order to achieve causal and symbolic processing of sensory input, i.e., allowing the MBCA to not only recognize but interpret events in the environment and produce motor outputs in anticipation of future outcomes, every evaluation cycle the MBCA must not simply process the input sensory vector for subsymbolic recognition, but must process this vector in terms of the belief system activated. Artificial Neural Networks have excellent pattern recognition and reinforcement learning abilities, but perform poorly at causally and logically making sense of their environment. Future artificial intelligence systems that attempt to overcome the limitations of subsymbolic architectures by the integration of symbolic processing, may be at risk for developing faulty belief systems with unintended results.
KeywordsCognitive architecture Artificial general intelligence Cortical minicolumns Belief systems Neural-symbolic integration
Thanks for discussion to Dr. Aliye Keskin. This article builds upon work originally presented at BICA 2018 (reference 5).
- 6.Schneider H (2019, in press) Subsymbolic versus Symbolic data flow in the meaningful-based cognitive architecture. In: Samsonovich AV (ed) Biologically inspired cognitive architectures 2019, BICA 2019. Advances in intelligent systems and computing. SpringerGoogle Scholar
- 7.Schneider H (2019, in press) Schizophrenia and the future of artificial intelligence. In: Samsonovich AV (ed) Biologically inspired cognitive architectures 2019, BICA 2019. Advances in intelligent systems and computing. SpringerGoogle Scholar
- 8.Samsonovich AV (2010) Toward a unified catalog of implemented cognitive architectures. In: BICA, vol 221, pp 195–244Google Scholar
- 9.Bach J (2008) Seven principles of synthetic intelligence. In: Proceedings of 1st conference on artificial general intelligence, Memphis, pp 63–74Google Scholar
- 11.Collier M, Beel J (2018) Implementing neural Turing machines. In: ICANNGoogle Scholar
- 19.Bostrom N (2014) Superintelligence: paths, dangers, strategies. Oxford University Press, OxfordGoogle Scholar