Estimated reading time: 5 minutes

Nolan’s Notes: The Turing Test

You can’t escape it even if you try: Mainstream media coverage of artificial intelligence (AI) is everywhere. Reporters and editors are tossing all sorts of requests at ChatGPT, Bing AI, and the like, then reporting on what they get back for results.

You see typical themes like, “ChatGPT will take away your job,” or the media reporting on music awards that explicitly forbid AI-generated music and lyrics. A large chunk of the coverage points out that the results of an AI request are too simplistic or even erroneous. That has us concerned, but remember, we heard the same concerns back in the 1970s, when robotic arms first rolled onto the factory floor. Those robots didn’t take away all the jobs, they just shifted the skill sets. It didn’t take long for the workers to realize that the robots merely did tedious tasks; it still took people to make sure things ran correctly. The robots weren’t really smart, and they couldn’t do what humans could.

Has anything changed? Is generative AI smart? How does this affect us in the EMS industry? That’s what this month’s issue of SMT007 Magazine is about.

“Have you heard of the Turing test? Computer pioneer Alan Turing proposed that a computer program should be considered able to “think” if a human interacting with the program in natural language was unable to tell whether they were communicating with a human or a machine. Turing proposed this test in a 1950 paper published by Computing Machinery and Intelligence. There are a multitude of sources for the development of the Turing test, but Wikipedia is concise:

In 1966,?Joseph Weizenbaum, a professor and computer scientist at MIT, created a program which appeared to pass the Turing test. The program, known as?ELIZA, worked by examining a user's typed comments for keywords. If a keyword is found, a rule that transforms the user's comments is applied, and the resulting sentence is returned. If a keyword is not found, ELIZA responds either with a generic riposte or by repeating one of the earlier comments.?In addition, Weizenbaum developed ELIZA to replicate the behavior of a?Rogerian psychotherapist, allowing ELIZA to be ‘free to assume the pose of knowing almost nothing of the real world.’ With these techniques, Weizenbaum's program was able to fool some people into believing that they were talking to a real person.”

As an undergrad studying computer science in the early 1980s, I recall interacting with a version of ELIZA as a laboratory exercise. At the time, I was unconvinced and lost interest in the conversation pretty quickly. Of course, I knew in advance that ELIZA was a program, not a person. I remember sharing in class that I found “her” rather droll and boring. Intelligent? Perhaps by Turing’s standards in the 1950s, when UNIVAC could perform a whopping 1,095 instructions per second (IPS), but it was insufficiently convincing in 1983 when an Apple II computer averaged 300,000 IPS. Nowadays, ELIZA is nothing more than a quaint exercise.

In fact, in a recent article published on INSIDER, DeepMind co-founder, Mustafa Suleyman, is proposing a new test for artificial intelligence. Rather than the Turing test, Suleyman suggests in his book, The Coming Wave: Technology, Power, and the Twenty-first Century’s Greatest Dilemma, that AI “should be tested on its ability to turn $100,000 into $1 million.”1 Yes, the Turing Test seems quite quaint, indeed.

Like so many other media channels, we couldn’t resist the urge to test the AI engines. I asked some of our contributors to play with generative AI and see what they got back. Happy Holden’s article on business strategies, for example, includes an extended AI passage. It’s somewhat useful and makes for a good starting point. It demonstrates that the program can respond to your natural language request with a much more sophisticated reply than ELIZA, but is it different enough?

I would argue that the current artificial intelligence engines likewise aren’t actually “intelligent.” All that these generative AI engines are capable of is collating or concatenating content and automating the data gathering process; but it’s still based on calculations. Yes, generative AI tools can create program code, art images, prose, or song lyrics upon demand. But at the core of these functions is the fact that the task is simple data collection, recalculation, and transformation of existing data as applied to language. What’s missing is “heuristics,” which is defined as “encouraging a person to learn, discover, understand, or solve problems independently, as by experimenting, evaluating possible answers or solutions, or by trial and error.”

Personally, I think the AI fearmongering craze misses the point. To my way of thinking, adaptive machine learning is much more interesting, much closer to heuristic thinking. The algorithmic research into sophisticated pattern identification has led to a wide number of advances in our body of knowledge. Computers running these cutting-edge analyses on big data have recognized patterns as wide ranging as retinal eye diseases, to a potential message from another planet located closer to the center of our galaxy. This last one may be far-fetched, but I will suggest that if and when we locate signals from space aliens, an AI pattern matching tool will be at the center of that discovery. It’s this sort of work that is truly powerful.

So, how do we tackle this topic in the July issue of SMT007 Magazine? Well, mostly by talking to real human beings who are topic experts. We intentionally scoped this conversation down to the EMS manufacturing process. Where does AI (machine learning) fit today? Where might it be going soon? This will be an ongoing topic, undoubtedly. We’re only just entering a brave new world where compute power and database capacity allows for more than number crunching—or should I say, allows number crunching to apply to linguistic communication as well?

Reference:
“DeepMinds’ co-founder suggested testing an AI chatbot’s ability to turn $100,00 into $1 million to measure human-like intelligence,” by Sawdah Bhaimiya, INSIDER.com, June 20, 2023.

This column originally appeared in the July 2023 issue of SMT007 Magazine.