Hypotheses for AI/Singularity Net

AIMS:

In "Computing Machinery Intelligence," Alan Turing frames a now-famous question:  can an artificially intelligent digital computing system win at the imitation game? Seventy one years later, Ben Goertzl and SingularityNet are trying to show that it can be done--not just in principle--but in practice.

One of the challenges Turing considers to his thesis is framed by Ada Lovelace herself. She states, "(Babbages) Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform".

It seems clear that AI systems employing complex, self-organizing, neural and Bayesian nets are capable of learning. In the terms of the philosophy of logic, they are capable of forms of inference akin to analogy and induction. Melanie Mitchell, for instance, has explored what is requisite for computational systems to learn to use words by analogy.

What is not clear, however, is what might be necessary for artificially intelligent systems to make abductive inferences to novel hypotheses. Generating such hypotheses is, as Ada Lovelace points out, a process that requires the origination of something new. A few decades later, in 1885, C.S Peirce argued that deduction and induction only serve to "render the indefinite definite" In other words, "Deduction Explicates; Induction evaluates: that is all." Only abduction originates new explanatory hypotheses.

The aim of this project is simple:  we propose to research and evaluate the best work by computer scientists, mathematicians and logicians on the logic of abduction. The literature is large. As such, we propose to focus on a particular strand that Peirce suggests might be particularly fruitful, which is the study of topological systems of logic that are diagrammatic in character. The system that he develops is called the existential graphs. The fully developed gamma version is a higher order relational logic that is capable of expressing the sorts of epistemic, alethic and deontic modalities that are typically found in the hypotheses that are articulated as conclusions in abductive forms of inference.

The purpose of evaluating this existing body of research is to identify the most promising avenues of inquiry that might be of interest for developers in the SingularityNet community.

TEAM and RELEVANT EXPERIENCE :

JEFFREY DOWNARD, philosophy professor of ethics, law and the logic of inquiry.

DAVID WATTS, business systems oriented on the problems of blockchain scalability, full stack developer.

STEPHEN LENHART, machine learning, developer, philosophical logic, philosophy of science

Board of Advisors: https://www.keplerscs.com/our-team-1

IP: Open conference and workshop sessions on Zoom in both a live and recorded form. The educational resources we develop will be publicly available via a Youtube channel under copy left permission. The ethical standards and policy recommendations will be published as a set of white papers.

TIMELINE AND BUDGET:

Months 1-3: Research the existing literature on abductive inference with a focus on topological systems of logic (3000)

Website development social media and Youtube channel: (2800)

Months 2-4: Creation and editing of an integrated set of video tutorials on the leading ideas on about abductive inference that might be useful for AI developers in SingularityNet. (3500)

Months 2-5: Foster a community of SingularityNet developers with interests in employing forms of abductive inference in AI (4000)

Months 3-5: Development by the network of a set of paradigmatic case studies of AI systems capable of varying degrees of abductive inference (3000)

Months 3-5: Workshops for AI and machine learning developers to study the develop paradigmatic examples of computational systems capable of abductive inference. (3500)

Months 1-5: Project direction (4500)

Months 1-5: Project oversight and evaluation by members of Board of Advisors (2500)

Total: (26800)

Phase 2:

Months 6-9: testing the beta versions of the abductive learning models

Phase 3:

Months 10-12: developing a model to sustain the community engaging in the abduction in AI project.

Links:

www.KeplerSCS.com

REFERENCES:

Aliseda, Atocha. "Abduction as epistemic change: A Peircean model in artificial intelligence." In Abduction and Induction, pp. 45-58. Springer, Dordrecht, 2000.

Hoffman, Robert R., William J. Clancey, and Shane T. Mueller. "Explaining AI as an Exploratory Process: The Peircean Abduction Model." arXiv preprint arXiv:2009.14795 (2020).

Paul, Gabriele. "AI approaches to abduction." In Abductive Reasoning and Learning, pp. 35-98. Springer, Dordrecht, 2000.

Inoue, Katsumi. "Automated abduction." In Computational Logic: Logic Programming and Beyond, pp. 311-341. Springer, Berlin, Heidelberg, 2002.

Costa, Darin McNabb. "Can creativity be formalized? Peircean reflections on the role of abduction in human intelligence." In Toward Artificial Sapience, pp. 3-14. Springer, London, 2008.

Bellucci, Francesco, and Ahti-Veikko Pietarinen. "Icons, Interrogations, and Graphs: On Peirce's Integrated Notion of Abduction." Transactions of the Charles S. Peirce Society 56, no. 1 (2020): 43-61.