Abstract
In the quest for artificial general intelligence (AGI), questions remain about what kinds of representations are needed for the kind of flexibility called for by complex environments like the physical world. A capacity for continued learning of many domains has yet to be realized, and proposals for how to achieve general performance improvement through continuous cumulative learning—while seemingly a necessary feature of any AGI—remain scarce.
In this paper we describe a cumulative learning mechanism that produces causal-relational models of its environment, to predict events and achieve goals. We show how such models, coupled with an appropriate modeling process, result in knowledge whose accuracy increases over time and can run continuously throughout the lifetime of an agent. The methods have been implemented, demonstrating learning of complex tasks and situated grammatically-correct natural language by observation. Here we focus on key theoretical principles of the modeling method and explain how effective cumulative learning is achieved.
This work was funded by Reykjavik U. and IIIM, Iceland.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
We mean any sub-division of E, \(e_n \subset E\), including sub-structures, component processes, whole-part relations, causal relations, etc.
- 2.
In any complex environment such as the physical world there will be innumerable ways of domain sub-divisions. The range of domains created from human-centric perspectives (e.g. transportation, electronics, home, commerce, clothing, etc.) demonstrate the utility of such sub-division.
- 3.
A controller is the process that dynamically couples knowledge and goals to obtain actions (or inaction) in an environment [14].
- 4.
Any reliable and repeatable regularity in a world is considered a causal relation, irrespective of whether it is observable or not, or truly deterministic or not [3].
- 5.
The speed of accurate model building is of course of critical importance for any real world implementation, determined in part by the details of the implementation methods and the nature of the task-environment; in this paper, however, the primary focus is on theoretical aspects of the modeling process.
- 6.
It should be noted that causal relations cannot be replaced by probabilities. Pearl [12] (p. 36) states: “...causality deals with how probability functions change in response to influences (e.g., new conditions or interventions) that originate from outside the probability space, while probability theory, even when given a fully specified joint density function on all (temporally-indexed) variables in the space, cannot tell us how that function would change under such external influences. Thus, ‘doing’ is not reducible to ‘seeing’, and there is no point trying to fuse the two together.”.
- 7.
Relevance is determined at “the top” by top-level goals, and at the “bottom” by incoming stimuli through sensors; in between the pattern matching on the models’ LT and RT determines their relevance.
References
Conant, R.C., Ashby, W.R.: Every good regulator of a system must be a model of that system\(\dagger \). Int. J. Syst. Sci. 1(2), 89–97 (1970)
French, R.M.: Catastrophic forgetting in connectionist networks: causes, consequences and solutions. Trends Cogn. Sci. 3(4), 128–135 (1999)
Halpern, J.Y., Pearl, J.: Causes and explanations: A structural-model approach – part 1: Causes. CoRR abs/1301.2275 (2013). http://arxiv.org/abs/1301.2275
Hasselmo, M.E.: Avoiding catastrophic forgetting. Trends Cogn. Sci. 21(6), 407–408 (2017)
Helgason, H.P., Nivel, E., Thórisson, K.R.: On attention mechanisms for agi architectures:a design proposal. In: Bach, J., Goertzel, B., Iklé, M. (eds.) AGI 2012. LNCS (LNAI), vol. 7716, pp. 89–98. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-35506-6_10
Kirkpatrick, J., et al.: Overcoming catastrophic forgetting in neural networks. Proc. Nat. Acad. Sci. 114(13), 3521–3526 (2017)
Lazaric, A.: Transfer in reinforcement learning: a framework and a survey. In: Wiering, M., van Otterlo, M. (eds.) Reinforcement Learning. Adaptation, Learning, and Optimization, vol. 12, pp. 143–173. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-27645-3_5
Lu, J., Behbood, V., Hao, P., Zuo, H., Xue, S., Zhang, G.: Transfer learning using computational intelligence: a survey. Knowl.-Based Syst. 80, 14–23 (2015)
Nivel, E., et al.: Bounded Recursive Self-Improvement. Technical RUTR-SCS13006. Reykjavik University Department of Computer Science, Reykjavik, Iceland (2013)
Nivel, E., Thórisson, K.R., Steunebrink, B., Schmidhuber, J.: Anytime bounded rationality. In: Bieger, J., Goertzel, B., Potapov, A. (eds.) AGI 2015. LNCS (LNAI), vol. 9205, pp. 121–130. Springer, Cham (2015). https://doi.org/10.1007/978-3-319-21365-1_13
Nivel, E., et al.: Bounded seed-AGI. In: Goertzel, B., Orseau, L., Snaider, J. (eds.) AGI 2014. LNCS (LNAI), vol. 8598, pp. 85–96. Springer, Cham (2014). https://doi.org/10.1007/978-3-319-09274-4_9
Pearl, J.: Bayesianism and causality, or, why I am only a half-Bayesian. In: Corfield, D., Williamson, J. (eds.) Foundations of bayesianism. Applied Logic Series, vol. 24, pp. 19–36. Springer, Dordrecht (2001). https://doi.org/10.1007/978-94-017-1586-7_2
Pollock, J.L.: Defeasible reasoning and degrees of justification. Argum. Comput. 1(1), 7–22 (2010)
Steunebrink, B.R., Thórisson, K.R., Schmidhuber, J.: Growing recursive self-improvers. In: Steunebrink, B., Wang, P., Goertzel, B. (eds.) AGI -2016. LNCS (LNAI), vol. 9782, pp. 129–139. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-41649-6_13
Thórisson, K.R., Kremelberg, D., Steunebrink, B.R., Nivel, E.: About understanding. In: Steunebrink, B., Wang, P., Goertzel, B. (eds.) AGI -2016. LNCS (LNAI), vol. 9782, pp. 106–117. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-41649-6_11
Thórisson, K.R., et al.: Autonomous acquisition of situated natural communication. Comput. Sci. Inf. Syst. 9(2), 115–131 (2014). outstanding Paper Award
Wang, P.: Insufficient knowledge and resources-a biological constraint and its functional implications. In: AAAI Fall Symposium: Biologically Inspired Cognitive Architectures. Citeseer (2009)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer Nature Switzerland AG
About this paper
Cite this paper
Thórisson, K.R., Talbot, A. (2018). Cumulative Learning with Causal-Relational Models. In: Iklé, M., Franz, A., Rzepka, R., Goertzel, B. (eds) Artificial General Intelligence. AGI 2018. Lecture Notes in Computer Science(), vol 10999. Springer, Cham. https://doi.org/10.1007/978-3-319-97676-1_22
Download citation
DOI: https://doi.org/10.1007/978-3-319-97676-1_22
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-97675-4
Online ISBN: 978-3-319-97676-1
eBook Packages: Computer ScienceComputer Science (R0)