-
Who were these people, and how did/does each contribute to the development of artificial intelligence? How did/does each think “intelligence” could be identified? (~50 words each)
Alan Matheson Turing (1912-1954): Biography (Links to an external site.); Work: Computing Machinery and Intelligence (Links to an external site.) (Turing, 1950).
John McCarthy (1927-2011): Homepage, (Links to an external site.) especially What is Artificial Intelligence? (Links to an external site.) (McCarthy, 2007).
Herb Simon (1916-2001): Meet the Nobel Laureates in Economics: Do we understand human behaviour (Links to an external site.) (UBS, n.d.).
Marvin Minsky (1927-2016): AI pioneer Marvin Minsky dies aged 88 (Links to an external site.) (BBC News, 2016).
Timnit Gebru (1982): We read the paper that forced Timnit Gebru out of Google. Here’s what it says (Links to an external site.) (Hao, 2020).
Turing was one of the pioneers of artificial (“machine”) intelligence and defined many foundational concepts. He argued that machines could simulate reasoning and devised the Turing test that determined that a computer could “think” if its responses to conversations were indistinguishable from human responses.
McCarthy coined the term artificial intelligence as the “science and engineering of making intelligent machines” (McCarthy, 2007) and wrote Lisp, a programming language widely used for AI research. He argued that machines can “understand” and “have beliefs”, which sparked debate that continues to present day.
Simon created the Logic Theorist and the General Problem Solver to demonstrate that computer programs could solve problems, advancing the notion that machines could “think”. The Logic Theorist used deductive reasoning and proved mathematical theorems better than humans in some cases. The General Problem Solver broke down problems into components and then solving each component.
Minsky built the first neural network machine, SNARC, which solved problems using primitive machine learning. He studied human psychology using computational ideas and theorized that intelligence could result from interactions of non-intelligent parts. Minsky defined AI as “the science of making machines do things that would require intelligence if done by men” (Minksy, 1968).
Gebru is an advocate of ethics and diversity in AI research. She is an outspoken critic of biases in AI due to the field consisting of predominantly white male researchers, and has succeeded in increasing scrutiny of tech giants such as Google and Amazon.
-
How do “machine (programming) languages” differ from human (natural) ones? (~100 words).
Human languages are learned, evolving, and adaptable based on users. If an error arises from human communication, it can be explained, understood, and resolved.
https://www.youtube.com/watch?v=Q8mD2hsxrhQ&ab_channel=Hugo2608
Human communication combines language with other forms of expression (e.g., tone, emotion, body language). The use of context, culture, and references not “coded” in everyone’s version of a language also adds nuance. Human languages can be translated by users and develop accents, dialects, or even new languages without outside intervention.
Machine languages are complete upon released with little room for change. They are logical, unambiguous, and require no interpretation. Machine languages are often incompatible and require outside translators.
-
How does “machine (artificial) intelligence” differ from the human version? (~100 words).
The definition of human intelligence is subjective and can incorporate characteristics such as self-awareness, adaptability, reasoning, and problem-solving. Humans also possess multiple intelligences which work simultaneously to solve problems or adapt to new situations. This makes intelligence challenging to objectively define and quantitatively measure. Since most machines were designed to complete specific tasks, the degree and/or accuracy of task completion can easily be used to determine intelligence. However, with the increasing sophistication of machines, “skill-acquisition efficiency” and “generalization” are more accurate measures of machine intelligence (Chollet, 2019). The two definitions are analogous to a student who memorizes materials compared to one who applies new knowledge.
-
How does “machine learning” differ from human learning? (~100 words)
Machine learning occurs when an algorithm analyzes datasets by comparing set metrics and outcomes, and then applies the same logic to predict outcomes for other datasets. Machines are prone to bias since they cannot differentiate between good and bad data, nor can they correct themselves with critical thinking or self-reflection. Machines learning is limited by the metrics and assumptions set by programmers because they cannot be changed without intervention. Humans learn by building a knowledge base with a myriad of definitions updatable any time new information we judge to be credible is encountered. Humans also use tools such as context and references to learn and predict ambiguous outcomes.
-
And for your LAST challenge, a version of the Turing Test: how do YOUR answers to these questions differ from what a machine could generate? (~200 words)
My answers were written using my prior knowledge and experiences with the terms and concepts given. I knew nothing about the scientists in Question 1 apart from Turing, so I answered using online information almost exclusively. Because of the word limit, I prioritized conveying information over sentence structure or descriptiveness, risking the final product sounding robotic. However, the process of creating the product involved a combination of other skills of which machines are not necessarily capable yet, such as judging the credibility of information sources and synthesizing a paragraph by selecting information and organizing them into sentences with a logical flow. Above all else, I believe prioritization would be the main factor distinguishing my answers from those of a machine. Not only did I prioritize the style of writing and content, but also the time and effort spent by performing a cost-benefit analysis. For this assignment, I can flexibly compartmentalize different tasks required such as reading, searching, writing, and editing, and allocate time appropriately. Finally, the metacognition needed to answer this question is the best Turing test as machines have no such self-awareness. I can recognize the imperfections in my responses and even their causes whereas a machine does not self-reflect.
References
- Chollet, F. (2019). On the Measure of Intelligence (arXiv:1911.01547). arXiv. http://arxiv.org/abs/1911.01547
- McCarthy, J. (2007). What is Artificial Intelligence? http://www-formal.stanford.edu/jmc/whatisai/whatisai.html
- Minsky, M. (1968). Semantic Information Processing, The MIT Press, Cambridge, Mass.