Reading Group #9 on AI and Knowledge

AI and Knowledge

Wednesday 27 January 2021, 15:00–17:00 GMT

Overview

Co-facilitated by: Prof. Stephanie Dick (University of Pennsylvania) 

Stephanie Dick

Synopsis

Many early attempts to reproduce 'intelligence' in computers focused on reasoning. Researchers like Allen Newell and Herbert Simon believed that general rules of reasoning, conceived as information processing rules, constituted human intelligence across domains from military decision making to language translation to chess playing. They aspired to discover and automate the application of those rules believing that the resulting AI would be able to solve problems across domains without needing much additional domain-specific information. However, this approach famously failed, by and large, to produce their desired results. According to some other early AI practitioners, these failures resulted because reasoning alone was not responsible for intelligence. These selected readings represent an alternative approach to artificial intelligence research that developed in the 1970s and continues (in variations) to this day – expert systems research. Expert systems research began from the belief that intelligent behavior in humans was as much a product of what humans know as it was a product of any reasoning capacity. As such, if you want to make intelligent machines, you have to find a way to give them knowledge to reason with/about. One of the founders of expert systems research was Edward Feigenbaum, a PhD student of Simon's at Carnegie Mellon University who went on to help shape AI and computer science research at Stanford. He summarized this approach: 'the problem-solving power exhibited by an intelligent agent's performance is primarily the consequence of its knowledge base, and only secondarily a consequence of the inference method employed. Expert systems must be knowledge rich even if they are methods-poor. ... The power resides in the knowledge.' Accordingly, the task of the expert systems developer was to extract and consolidate expert knowledge and then find a way to encode it for the purposes of automation. A related area of research called 'knowledge acquisition' took shape exploring how expert knowledge could be effectively and efficiently 'extracted.' The following readings introduce the theory of knowledge and intelligence at work in expert systems, some approaches to 'knowledge acquisition,' and some applications, both real and imagined. We also include some examples to highlight the fact that expert systems were also among the earliest commercially successful applications of AI, drawing interest and investment from institutions ranging from Schlumberger – a major oil and gas exploration and data processing company – to the National Health Service in the UK. We look forward to unpacking all of this with you!

Assigned texts

  • Edward Feigenbaum, 'Knowledge Engineering: The Applied Side of Artificial Intelligence', Stanford University Department of Computer Science Report No. STAN-CS-80-812 (September 1980), excerpts.
  • James Heatherton, Todd Vikan, An Introduction to Expert Systems and Knowledge Acquisition Techniques, Air Force Institute of Technology, Report AU-AFIT /LS/TR-90-1 (September 1990), excerpts.
  • Edward Feigenbaum, Pamela McCorduck, H. Penny Ni, 'Working Smarter' in The Rise of the Expert Company: How Visionary Businesses are Using Intelligent Computers to Achieve Higher Productivity and Profits (preprint) [Times Books, 1988]
  • Edward Feigenbaum, Pamela McCorduck, H. Penny Ni, 'The National Health Service' in The Rise of the Expert Company: How Visionary Businesses are Using Intelligent Computers to Achieve Higher Productivity and Profits (preprint) [Times Books, 1988]
  • David Barstow, 'Artificial Intelligence at Schlumberger' in AI Magazine, Vol. 5, No. 4 (1984).
  • Randall Davis et al, 'The DIPMETER Advisor: Interpretation of Geologic Signals' in Proceedings of the 7th International Joint Conference on Artificial Intelligence (August 1981)

Event summary

Prof. Dick presented the group with selected readings pertaining to the development of expert systems. Noting the lack of attention received by the shift to developing expert systems in the project of encoding human cognition, she provided an overview of the changing attitudes toward artificial intelligence in academia and industry, covering:

  • Simon and Newell's interpretation of the human brain and the machine as systems of the same genus. They conceived of human intelligence as an information process comprised of sequences (rules and reasoning) which could exist in natural or artificial systems, including digital ones.
  • The failure of the General Problem Solver model to account for heuristics across systems in the 1960s led to Feigenbaum’s production of a new theory of human intelligence correlating with knowledge instead of reason.
  • The initial production of expert systems in the 1960s and 1970s, the significant commercial investments in expert systems (in contrast to early AI as a primarily academic project), and the production of expert systems as a discipline in the 1980s and 1990s.
  • The shared commitment to knowledge which informs the production of expert systems as organisationally differentiated from more top-down structures of encoding intelligence.
  • The extraction of human experts' knowledge as a pedagogical exercise central to the efficacy of expert systems.

Before returning for Q&A, the cohort were then invited to discuss the following questions in breakout rooms:

  1. What would you use expert systems to teach undergraduate students?
  2. What would you aim to convey to the broader public about expert systems and their histories?
  3. What would you aim to convey to your colleagues about expert systems and their histories?

Works cited (in the chat)

Expertise and expert systems

Formalisms

  • Ensmenger, Nathan. 'The Multiple Meanings of a Flowchart'. Information & Culture 51, no. 3 (August 2016): 321–51.
  • Orr, Julian E. Talking about Machines. Cornell University Press, 1996.
  • Ott, Julia C. When Wall Street Met Main Street: The Quest for an Investors' Democracy. Cambridge, Mass: Harvard University Press, 2011.

Anthropology of AI / data science

  • Forsythe, Diana, and David J. Hess. Studying Those Who Study Us: An Anthropologist in the World of Artificial Intelligence. Writing Science. Stanford, Calif: Stanford University Press, 2001.
  • Ribes, David, Andrew S Hoffman, Steven C Slota, and Geoffrey C Bowker. 'The Logic of Domains'. Social Studies of Science 49, no. 3 (June 2019): 281–309.

Ideology

  • Katz, Yarden. Artificial Whiteness: Politics and Ideology in Artificial Intelligence. New York: Columbia University Press, 2020.
  • Giridharadas, Anand. Winners Take All: The Elite Charade of Changing the World. First edition. New York: Alfred A. Knopf, 2018.
  • Waterhouse, Benjamin C. Lobbying America: The Politics of Business from Nixon to NAFTA. Politics and Society in Twentieth-Century America. Princeton, New Jersey: Princeton University Press, 2014.

Other