Issue Number 271 December 15, 2019

This free Information Age Education Newsletter is edited by Dave Moursund and produced by Ken Loge. The newsletter is one component of the Information Age Education (IAE) and Advancement of Globally Appropriate Technology and Education (AGATE) publications.

All back issues of the newsletter and subscription information are available online. Fourteen of the newsletters are available in Spanish. In addition, seven free books based on the newsletters are available.

Dave Moursund’s newly revised and updated book, The Fourth R (Second Edition), is now available in both English and Spanish (Moursund, 2018a, link; Moursund, 2018b, link). The unifying theme of the book is that the 4th R of Reasoning/Computational Thinking is fundamental to empowering today’s students and their teachers throughout the K-12 curriculum. The first edition was published in December, 2016, and the second edition in August, 2018. The Spanish translation of the second edition, La Cuarta R, was published in September, 2018. The three books have now had a combined total of over 76,000 page-views and downloads. More than 16,000 of these are the Spanish edition.

Real and Artificial Intelligence

David Moursund
Professor Emeritus, College of Education
University of Oregon

Artificial Intelligence (AI) is a very long-term interest of mine. I recently read a fascinating new book, Rebooting AI: Building Artificial Intelligence We Can Trust by Gary Marcus and Ernest Davis (Marcus & Davis, 2019). This newsletter includes some of my insights from reading the book. I recommend it highly!

But first, I want to present some ideas about non-artificial intelligence.

Some Definitions of Intelligence

I began my review of AI by consulting several dictionaries, and extracted from them the following short definition:

Intelligence is the ability to acquire and apply knowledge and skills.

This definition probably meets the needs of most people. It contains the two key ideas about human intelligence.

  1. People acquire knowledge and skills. They do this through a combination of formal schooling and participation in informal learning situations. We all are learning 24 hours a day and seven days a week as our brain continuously processes its stored information and combines it with the new information coming from our internal and exterior senses.
  2. People use the knowledge and skills that they have acquired. For example, consider just our somewhat conscious decisions. A study done by Cornell University researchers found that the average adult makes about 35,000 decisions a day (Graff, 2/7/2018, link).

Each person is unique, so that each person’s daily sequence of decisions is also likely unique. Moreover, some decisions are likely bad decisions. So, can we expect that an artificially intelligent robot engaged in humanlike activities will make no bad decisions? This is a fundamental question to be asked as AI becomes more broadly used.

A standard example is driverless cars, with an AI system as the driver. Clearly, humans who are driving cars will occasionally make bad decisions. Suppose that we can produce a driverless car that “only” makes two-percent as many bad decisions as an average human driver. That level of performance would save a very large number of lives and prevent a very large number of injuries. Still, some bad decisions will be made because every driving episode is unique. We cannot expect perfection from driverless cars or other AI systems. Marcus and Davis focus on this idea in Rebooting AI: Building Artificial Intelligence We Can Trust.

Artificial Intelligence and Machine Intelligence

When the term artificial intelligence (AI) was being developed and first widely used, an alternative term was machine intelligence (MI). The two terms are considered to be synonymous. So, what do we mean by AI? Here is one definition (Wikipedia, 2019, link):

Artificial intelligence (AI) is the ability of a computer program or a machine to think and learn. It is also a field of study which tries to make computers "smart". They work on their own without being encoded with commands. John McCarthy came up with the name "artificial intelligence" in 1955.

In general use, the term "artificial intelligence" means a program which mimics human cognition. At least some of the things we associate with other minds, such as learning and problem solving can be done by computers, though not in the same way as we [humans] do.

The first appearance of artificial intelligence is in Greek myths, like Talos of Crete or the bronze robot of Hephaestus. Humanoid robots were built by Yan Shi, Hero of Alexandria, and Al-Jazari. Sentient machines became popular in fiction during the 19th and 20th centuries with the stories of Frankenstein and Rossum's Universal Robots.

Notice the modern emphasis on a particular type of machine that we call a computer. The UNIVAC I, the first commercially produced electronic digital computer in the United States, became available in 1951. When the terms AI and MI were being created, the motivation and emphasis was on electronic digital computers.

Here is a fundamental issue in the current field of AI. When a human acquires knowledge and skills, we expect this to be accompanied by an understanding of what has been acquired. Indeed, this is a fundamental goal in schooling. Rote memory without understanding occurs, but is considered to be a rather inferior type of education. Today’s computers are far, far better than humans at rote memorization and retrieval of what has been memorized. (For example, think of the Web and a search engine such as Google.) Today’s artificially intelligent computers acquire knowledge and skills, but have no understanding of what they are acquiring.

Indeed, that is the essence of the message in the Marcus and Davis book. I routinely use the AI system named Google to retrieve information. Google has no human-like understanding of the results it provides to me. I use my human intelligence and understanding as I read and make use of the information that Google and other resources provided. I use my human intelligence, knowledge, and understanding to decide if the information is valid (not fake news) and whether it is applicable to the task I have at hand.

A Broader Definition of Intelligence

Quoting again from the Wikipedia article cited above:

Intelligence has been defined in many ways: the capacity for logic, understanding, self-awareness, learning, emotional knowledge, reasoning, planning, creativity, critical thinking, and problem solving. More generally, it can be described as the ability to perceive or infer information, and to retain it as knowledge to be applied towards adaptive behaviors within an environment or context.

Intelligence is most often studied in humans but has also been observed in both non-human animals and in plants. Intelligence in machines is called artificial intelligence, which is commonly implemented in computer systems using programs and, sometimes, specialized hardware. [Bold added for emphasis.]

You know that computers have used AI to defeat world champions in games such as chess, Jeopardy, various forms of poker, and Go. You have undoubtedly made use of the Web and a search engine to retrieve information. Likely you have used the spelling and grammar checkers in a word processor. Probably you have used voice input to a computer and, indeed, carried on a conversation with a computer when seeking help from a company’s “help desk.” Spend some time thinking carefully about the bolded sections in the definition above. In your personal opinion, how well are today’s computers doing in meeting this definition of intelligence? For example, do you think current AI systems have self-awareness and understanding?

Rebooting AI: Building Artificial Intelligence We Can Trust

Marcus and Ernest Davis provide an excellent overview of the progress that AI has made over the past 60-some years. In brief summary, this progress has been amazing, but has a very long way to ago in terms of achieving human-like intelligence. Quoting information from their publisher (Penguin Random House, 2019, link):

GARY MARCUS is a scientist, best-selling author, and entrepreneur. He is the founder and CEO of Robust.AI and was founder and CEO of Geometric Intelligence, a machine-learning company acquired by Uber in 2016. He is the author of five books, including Kluge, The Birth of the Mind, and The New York Times best seller Guitar Zero.

ERNEST DAVIS is a professor of computer science at the Courant Institute of Mathematical Science, New York University. One of the world’s leading scientists on commonsense reasoning for artificial intelligence, he is the author of four books, including Representations of Commonsense Knowledge and Verses for the Information Age.

According to Markus and Davis, the problem is that AI systems lack understanding of the world and of the things in it, such as humans and all of the things that humans do. Quoting from Rebooting AI: Building Artificial Intelligence We Can Trust:

The core problem is trust. The narrow AI systems we have now often work—on what they are programmed for—but they can’t be trusted with anything that hasn’t been precisely anticipated by their programmers. That is particularly important when the stakes are high. If a narrow AI system offers you the wrong advertisement on Facebook, nobody is going to die. But if an AI system drives your car into an unusual-looking vehicle that isn’t in its database, or misdiagnoses a cancer patient, it could be serious, even fatal.

What’s missing from AI today––and likely to stay missing, until and unless the field takes a fresh approach––is broad (or “general”) intelligence. AI needs to be able to deal not only with specific situations for which there is an enormous amount of cheaply obtained relevant data, but also problems that are novel, and variations that have not been seen before.

Broad intelligence, where progress has been much slower, is about being able to adapt flexibly to a world that is fundamentally open-ended––which is the one thing humans have, in spades, that machines haven’t yet touched. But that’s where the field needs to go, if we are to take AI to the next level.

As an example, consider the challenge of serving as a general helper in a household, dealing with oral input and oral responses in the context of the various situations that occur, doing cleaning, cooking, serving, taking care of children, and so on. These tasks are quite open-ended and routinely contain never-before-experienced situations. A human parent can do relatively well in these situations, but the very best of today’s AI-based robots are terrible at such a challenge.

Language translation is another example discussed by Markus and Davis. We now have computer systems that can take written or oral input in one language, and produce written or oral translation in any one of a large number of other languages. I am impressed by how well these systems work. However, they do their translation work with no understanding whatsoever of the meaning of the input they are receiving and the output they are producing. The quality of such translations is reasonably good, but is not sufficient for complex tasks such as translations of legal documents, poetry, or in other instances where small errors make a huge difference.

I personally tested one such translation system by taking a page from one of my books written in English, translating it into a second language, translating that result into a third language, and translating that result back to English. The results were usable, but far from the quality we would expect from good human translators.

Web searches are another place where we use AI and have learned to tolerate its weaknesses. From time to time, I ask the Google browser to find a definition for me. While writing this current newsletter, I asked Google to “define Artificial Intelligence.” In less that a second Google was able to identify about 429 million documents in its memory storage units that contain the phrase “artificial intelligence”. The results were rank-ordered using a proprietary system that Google has created and continues to improve.

I am really impressed by this information storage and retrieval system. But—429 million responses? That certainly sounds like overkill to me. Hmm. Which one of the definitions is the best, or most accurate, or most used by leading researchers in the field?

Which of these definition(s) will be most useful to me in the document I am writing? The Google system does not know what I am writing, or my intended audience. Suppose that, instead of searching Google, I was talking to a close professional acquaintance who also is a human expert in AI. This person knows me, my background, and my interests. This person might ask me about how I was going to use the definition, and for other information that would help the expert to more effectively communicate with me in response to my question. Humans do this, and we can hope that a truly expert AI information retrieval system will be able to do this in the future. We are currently a very long way from achieving this goal.

This example illustrates a major issue in AI. What do I want computers to know about me, and about the people I communicate with? Or, take the example of a student in school. The student routinely makes use of computer-assisted learning systems that are driven by AI, and that keep track of many details about a student’s performance and progress. The computer uses this information to make decisions about new content to teach or previous content to review, the type of instructional methods that may work best for a particular student at a particular time, what is likely to work well in assessment for this particular student, the content the student is studying, the time of day, and so on. Hmm. Who else should have access to such detailed information about a student, and how accurate will the computer system be in making such decisions? What if the school’s computer system gets hacked, and a student’s personal information becomes widely available? These are important and challenging questions to be considered as AI continues to evolve and we use its capabilities more widely in our educational systems.

Final Remarks

AI is now a routine part of our world. Our children are growing up in a world that is being changed by AI. So, what do you want our students to be learning about the AI aspects of computers?

This is a very difficult question, and I certainly cannot provide you with a simple, definitive answer. However, here is an analogy that you may find useful. Schools have been teaching the tools we call reading, writing, and arithmetic for more than 5,000 years. Over these years, we have made changes in the content, instructional processes, and assessments used in these aspects of schooling. But, to a large extent, these goals are clear and do not change much from decade to decade or even from century to century. We know that knowledge and skill in using these basic tools will empower their users. Nowadays, we expect all adults to have to have a working knowledge of these three basic tools.

Now, think about computers and AI as new empowering tools. My personal belief is that today’s children should all be gaining a working knowledge of the capabilities and limitations of computers and AI. They should receive an education that empowers the students to make effective use of these new tools. In the same way that instruction in reading, writing, and arithmetic begins when children are at a preschool level and cuts across all disciplines of study, I believe we should be providing children with similar levels of instruction in the use of computers and AI.

I like to pose and think about the following questions:

  1. If a computer can greatly help in solving a problem or accomplishing a task that students are studying in school, how should this affect the content being taught, the teaching and learning processes, and the assessment processes? In addition, what new topics could be added to the curriculum because of the capabilities of computers and AI?
  2. If a computer can greatly help students with their learning, and with learning how to learn topics taught in school, how should this affect the teaching, learning, and assessment processes?

All teachers are faced by these questions as they apply to the courses they are teaching and to their students in these courses. In my opinion, every teacher should have sufficient knowledge about the computers and AI systems that are relevant to the content, pedagogy, and assessment in the courses they teach that will enable them to make effective use of these tools in their teaching.

In terms of specific goals for students, I believe they should be learning some of the capabilities, limitations, and implications of AI in terms of what they already know, what they are learning, and the realities of the world they live in. Here is a simple example. Four-year-old children know, and in some sense understand, that on their next birthday they will add 1 to their current 4 years in order to be five years old. They happily anticipate this happening, and to the accompanying birthday celebration. An inexpensive handheld calculator or a sophisticated AI system can help a child to determine that 4 + 1 = 5. But neither of these has any understanding of a child’s excited anticipation of growing a year older, nor of the coming joys of a birthday party.

References and Resources

Graff, F. (2/7/2018). How many daily decisions do we make? University of North Carolina TV Science. Retrieved 11/26/2019 from http://science.unctv.org/content/reportersblog/choices.

Marcus, G., & Davis, E. (2019). Rebooting AI: Building artificial intelligence we can trust. New York: Penguin Random House.

Moursund, D. (6/25/2019). David Moursund honored at ISTE 2019 conference. IAE Blog. Retrieved 12/1/2019 from https://i-a-e.org/iae-blog/entry/david-moursund-honored-at-iste-2019-conference.html.

Moursund, D. (6/9/2019). Forecasting possible futures of education. IAE Blog. Retrieved 12/1/2019 from https://i-a-e.org/iae-blog/entry/forecasting-possible-futures-of-education.html.

Moursund, D. (8/22/2018). Big brother’s growing capability to listen to and censor you. IAE Blog. Retrieved 12/1/2019 from https://i-a-e.org/iae-blog/entry/big-brother-s-growing-capability-to-listen-to-and-censor-you.html.

Moursund, D. (6/30/2018). Cobots in automotive factories and in education. IAE Newsletter. Retrieved 12/1/2019 from https://i-a-e.org/newsletters/IAE-Newsletter-2018-236.html.

Moursund, D. (2018a). The fourth R (Second edition). Eugene, OR: Information Age Education. Retrieved 12/1/2019 from http://iae-pedia.org/The_Fourth_R_(Second_Edition). Download the Microsoft Word file from http://i-a-e.org/downloads/free-ebooks-by-dave-moursund/307-the-fourth-r-second-edition.html. Download the PDF file from http://i-a-e.org/downloads/free-ebooks-by-dave-moursund/308-the-fourth-r-second-edition-1.html. See the Spanish edition, La cuarta R, below.

Moursund, D. (2018b). La cuarta R. Eugene, OR: Information Age Education. Retrieved 11/21/2019 from http://iae-pedia.org/La_Cuarta_R_(Segunda_Edici%C3%B3n).

Moursund, D. (2016). Artificial intelligence. IAE-pedia. Retrieved 12/1/2019 from http://iae-pedia.org/Artificial_Intelligence.

Moursund, D. (2015a). Brain science for educators and parents. IAE-pedia. Retrieved 12/2/2019 from http://iae-pedia.org/Brain_Science.

Moursund, D. (2015b). Two brains are better than one. IAE-pedia. Retrieved 12/1/2019 from http://iae-pedia.org/Two_Brains_Are_Better_Than_One.

Pavlus, J. (10/17/2019). Machines beat humans on a reading test. But, do they understand? Quanta Magazine. Retrieved 12/2/2019 from http://www.nerdstalker.com/2019/10/machines-beat-humans-on-reading-test.html.

Penguin Random House (2019). Gary Marcus and Ernest Davis. Retrieved 11/20/2019 from https://www.penguinrandomhouse.com/books/603982/rebooting-ai-by-gary-marcus-and-ernest-davis/.

Wikipedia (2019). Intelligence. Retrieved 11/26/2019 from https://en.m.wikipedia.org/wiki/Intelligence.

Author

David Moursund is an Emeritus Professor of Education at the University of Oregon, and editor of the IAE Newsletter. His professional career includes founding the International Society for Technology in Education (ISTE) in 1979, serving as ISTE’s executive officer for 19 years, and establishing ISTE’s flagship publication, Learning and Leading with Technology (now published by ISTE as Empowered Learner). He was the major professor or co-major professor for 82 doctoral students. He has presented hundreds of professional talks and workshops. He has authored or coauthored more than 60 academic books and hundreds of articles. Many of these books are available free online. See http://iaepedia.org/David_Moursund_Books .

In 2007, Moursund founded Information Age Education (IAE). IAE provides free online educational materials via its IAE-pedia, IAE Newsletter, IAE Blog, and IAE books. See http://iaepedia.org/Main_Page#IAE_in_a_Nutshell . Information Age Education is now fully integrated into the 501(c)(3) non-profit corporation, Advancement of Globally Appropriate Technology and Education (AGATE) that was established in 2016. David Moursund is the Chief Executive Officer of AGATE.

Email: moursund@uoregon.edu

Reader Comments

We are using the Disqus commenting system to facilitate comments and discussions pertaining to this newsletter. To use Disqus, please click the Login link below and sign in. If you have questions about how to use Disqus, please refer to this help page.

Readers may also send comments via email directly to moursund@uoregon.edu.

About Information Age Education, Inc.

Information Age Education is a non-profit organization dedicated to improving education for learners of all ages throughout the world. Current IAE activities and free materials include the IAE-pedia at http://iae-pedia.org, a Website containing free books and articles at http://i-a-e.org/, a Blog at http://i-a-e.org/iae-blog.html, and the free newsletter you are now reading. See all back issues of the Blog at http://iae-pedia.org/IAE_Blog and all back issues of the Newsletter at http://i-a-e.org/iae-newsletter.html.