It all started with Eliza Doolittle. I鈥檓 not speaking about my fondness for musicals, I鈥檓 referring to the birth of 鈥渃hatbots.鈥�
ELIZA was the name of a program considered to be the world鈥檚 first chatbot, created in the 1960s by MIT Professor Joseph Weizenbaum. Weizenbaum named it after Eliza Doolittle, the poor flower-seller with a strong Cockney accent whom Professor Henry Higgins teaches to speak like an upper-class lady in the musical My Fair Lady.
Like the fictional Eliza, the computer program learned to 鈥渟peak properly鈥� based on instructions it was given by computer code. ELIZA was able to simulate conversation, pioneering artificial intelligence that inspired much more sophisticated later versions like Siri, Alexa and ChatGPT.
ELIZA was able to 鈥渃onverse鈥� with a person who asked questions or made comments by typing on a keyboard. Her responses seemed as if they were coming from a human. What the program actually did was search for keywords in the user鈥檚 message and then mirror these back in some way that gave the appearance of intelligent conversation. Weizenbaum wrote a famous program called 鈥淒OCTOR鈥� that mimicked the first interview between a patient and a psychotherapist:
User: Men are all alike.
ELIZA: In what way?
User: They鈥檙e always bugging us about something or other.
ELIZA: Can you think of a specific example?
User: Well, my boyfriend made me come here.
ELIZA Your boyfriend made you come here?
User: He says I鈥檓 depressed much of the time.
ELIZA: I鈥檓 sorry to hear you are depressed.
This made ELIZA seem like a good listener, but it did no more than spin the user鈥檚 words back in a friendly fashion. Many users genuinely felt understood, a phenomenon now known as the ELIZA effect, whereby humans attribute intelligence or empathy to computers that are merely programmed to follow patterns of words. This conversational mirroring can make users feel 鈥渉eard,鈥� though the program has no comprehension of meaning.
Weizenbaum was surprised that some people, including his secretary, attributed human-like feelings to the computer program and was adamant that unlike what many users believed, the computer was not 鈥渢hinking,鈥� but was just looking for words or expressions to which it had been programmed to respond. Today鈥檚 chatbots have gone way beyond those early efforts, thanks to far more sophisticated computer technology that makes possible instant access to the massive amounts of information stored in data banks around the world.
How does a chatbot like ChatGTP work? If you ask the question: 鈥淲ho portrayed Eliza Doolittle in the movie version of My Fair Lady?鈥� the answer will immediately come back as Audrey Hepburn. How? Because the program has learned patterns of language and has seen millions of examples where 鈥淢y Fair Lady movie鈥� is followed by 鈥淎udrey Hepburn.鈥� Not only does it give the correct answer, the chatbot offers a further interesting detail. While Hepburn spoke the role, her singing was dubbed by soprano Marni Nixon.
The dubbing of Hepburn鈥檚 voice created the illusion that Hepburn was singing. Chatbots also create an illusion, that of speaking with a human while the user is communicating with a computer that has no understanding or judgment and is reliant on the programs with which it has been equipped to find an appropriate answer by combing through the staggering amount of data in the cybersphere. Most of the time chatbots do an excellent job, but we still have to be aware of the classic dictum 鈥済arbage in, garbage out.鈥�
And as we know, there is a lot of misinformation floating about that can find its way into a chatbot conversation if that chatbot is not properly programmed. Programming is done by humans who, as history has taught us, have been known to make mistakes.
I decided to try Chat GTP with a few of the questions that I typically get asked. I started with: 鈥淚s there any risk of parabens in cosmetics?鈥� Back came an answer in less than a second detailing how parabens are preservatives that inhibit microbial growth by interfering with microbial enzyme activity, that they are stable and compatible with other ingredients, and that while they do have estrogen-like activity, this is thousands of times weaker than that of the body鈥檚 own estradiol. There is no evidence of any harm at the levels allowed in cosmetics, the chatbot concluded. There were more details, all of which, according to聽, were correct.
Next, I tried a question I thought was more challenging: 鈥淗ow does cholesterol build up in arteries?鈥� That obviously prompted the chatbot to search for all articles in which the word 鈥渃holesterol鈥� is followed or preceded by 鈥渁rteries,鈥� undoubtedly numbering in the millions. The elaborate answer explained how cholesterol travels in the blood inside 鈥渓ipoproteins,鈥� how the variety known as low density lipoprotein (LDL) can slip into the inner layer of the arterial wall especially when this is damaged by high blood pressure, smoking or inflammation.
Once trapped, LDL is oxidized, triggering an immune reaction whereby macrophages, a type of white blood cell, try to 鈥渃lean up鈥� the oxidized LDL. This results in the engulfed lipids building up in the arterial wall, over years forming a 鈥減laque鈥� that contains cholesterol, fats and dead cells narrowing the artery and reducing blood flow. A rupture of this plaque causes platelets to rush to the scene and form a blood clot that can cause a heart attack or stroke by blocking an artery. There was more detail about cholesterol measurement, diets that lower the risk of plaque formation and mention that cholesterol is needed by the body to make hormones and bile acids, meaning that the problem is not cholesterol itself but its misplacement. Once again, all of this was in line with聽.
An addendum: When I asked my question about the portrayal of Eliza Doolittle in the movie, ChatGTP also told me that the original Broadway Eliza had been Julie Andrews in 1956, but her star power wasn鈥檛 judged to be enough for the film version, a decision that famously sparked debate at the time. I never got to see that show, but saw her in Camelot, a performance that triggered my fondness for Broadway shows.
I did ask one more question of the chatbot: 鈥淐an Chat GTP be trusted?鈥� Back came the answer: 鈥淗onestly, ChatGPT can be useful, but it shouldn鈥檛 be blindly trusted.鈥� BINGO!
