In January 1966, an MIT professor named Joseph Weizenbaum decided to conduct an experiment in human-computer interaction. Towards that end, he created an app called ELIZA. The app was organized around a set of scripts, each representing a persona. The most famous of these scripts was the Doctor script, which simulated a Rogerian psychotherapy session. This is the kind of session where the psychotherapist reflects your words back at you to get you to talk more.
People who interacted with ELIZA, knowing that they were interacting with software, reported feeling understood and supported by it, as if they had been conversing with a human psychotherapist. Some of them even revealed personal information to ELIZA.
Here is an example of what an ELIZA session might look like:
ELIZA: Is something troubling you?
USER: All men are alike.
ELIZA: What is the connection, do you suppose?
USER: They're always bugging us about something or other.
ELIZA: Can you think of a specific example?
USER: Well, my boyfriend made me come here.
ELIZA: Is it important to you that your boyfriend made you come here?
USER: He says I'm depressed much of the time.
ELIZA: I am sorry to hear that you are depressed.
USER: It is true. I am unhappy.
ELIZA: Can you explain what made you unhappy?
You can see examples of ELIZA latching onto a pattern in the user input and repeating it back with simple word substitution (“you” for “I”, “your” for “my” and the like). User input that does not conform to a recognized pattern is handled by canned responses (e.g., “What is the connection, do you suppose?”). There is no intelligence in the app whatsoever. ELIZA is orders of magnitude beneath large language models (LLMs) when it comes to the kinds of responses it can generate, yet people treated ELIZA as though it were intelligent. No wonder some people think today’s LLMs are intelligent, when in reality they are not.
This phenomenon in human-computer interaction is referred to as the “ELIZA effect.” Today, 60 years after the introduction of ELIZA, we still see the ELIZA effect when people install AI companion apps on their mobile devices, interact with them and even have relationships with them.
We tend to anthropomorphize things like animals and AI, even ascribing intelligence to them. Why do we do this? A study published in Frontiers in Psychology provides some insight. The authors draw a correlation between having emotional feelings for AI (such as liking it and perceiving emotional support from it) and the need to belong. We all have a need to belong. Perhaps that makes us susceptible to the ELIZA effect.
We need to be aware of and intentional about our relationship with AI, understanding that our perception of real intelligence in AI is only a perception, not reality. Some organizations have already attempted to replace humans with AI to improve profits. If the idea that AI is intelligent becomes commonly accepted, more organizations will do this. These efforts will almost certainly fail to generate the expected results, but they will create much personal and economic disruption in the process.
AI can generate results that appear reasonable and sourced from intelligence. Since AI is not intelligent, the output they generate can’t be assumed to be correct. We need to be in the loop for any endeavor that includes AI. When it comes to AI usage, we need to keep it human.
Leave a Reply