The circus monkey
OpenAI organization announced on November 30th, 2022 the availability of ChatGPT. This is the next version of GPT3, which supports a dialog between a user and machine. GPT3 is a language model, which provides answers to requests like 'write me a short story about a boy and his dog' or 'create a blog post about blockchain'. Unlike GPT3, ChatGPT is supposed to carry on a dialog as you dig deeper into a topic.
Since the announcement, the Internet has lit up with articles about the end of creative writing. Some even say it’s the end of humanity as we know it. It’s a gigantic leap for AI - all written by humans. To top it off, Microsoft has announced that it will incorporate ChatGPT into its search engine Bing, to (finally) become a serious competitor to Google.
I would submit to you that most of it is hype with little basis in reality.
Let's start with GPT3. In 2022 I wrote an eBook, Technology Insights for Business about GPT3, to provide context about where and how it could be used. The word 'context' is very important when dealing with GPT3.
For example, the other day I asked GPT3 about light.
Q: Give me a definition of light.
A: Light is a form of energy that can be seen by the human eye, made up of particles called photons. It travels in waves and has properties like color, intensity and direction.
It sounds reasonable. The next question was very similar, but you should be able to spot the difference:
Q: Give me a true definition of light
A: Light is a form of electromagnetic radiation that can be detected by the human eye. It has no mass and travels at 186,000 miles per second in a vacuum. Light consists of various wavelengths which give it different colors, such as red, orange, yellow, green blue and violet.
This is where I was puzzled. Implicitly, I expected the answer to be identical. Every time I ask a computer/program a question I expect to get the same answer. The word 'true' has a very specific meaning in human language. Does it mean that without the qualifying word, that I am not getting true answers?
Then I tried the next question.
Q: Give me a correct definition of light
A: Light is a form of electromagnetic radiation that can be detected by the human eye. It has both wave-like and particle-like properties, making it unique among other forms of energy in terms of its behavior. Light travels at a speed of approximately 186,000 miles per second (300 million meters/second) in vacuum or air. The visible spectrum consists primarily of light waves within wavelengths ranging from 400 to 700 nanometers (nm).
Now I have to remember when I talk to a computer, I have to be very specific with my questions and always explicitly state that I want not only a 'true' but also 'correct' answer. I had to consult Google to see what the real definition of light is. That's where the major difference was. Google's answer was far more nuanced. On the results page it provided the definition from a dictionary and then suggested different contexts within which to search for the answer. Where GPT3 guessed my intention, Google tried to understand what my question really was.
From science, I tried something more philosophical, where the actual answer really requires context.
I started with one of the ultimate human questions.
Q: Give me definition of love
A: Love is an emotion or feeling characterized by warmth, tenderness, and affection. It can also be a virtue representing human kindness, compassion, and affection—the unselfish loyal and benevolent concern for the good of another. Love can also be an act of giving or receiving affection or caring. It can be a strong affectionate attachment or devotion to a person, place, or thing.
And then I wanted to know the truth
Q: Give me true definition of love
A: It is difficult to provide a single, definitive definition of love because the word "love" can be used to describe a wide range of strong emotions and experiences. Love can be romantic, platonic, familial, or self-love. It can be the love that exists between friends, the love that exists between romantic partners, or the love that a person has for themselves. Love can be passionate and intense, or it can be gentle and nurturing. Ultimately, the true definition of love may depend on an individual's personal experiences and perspectives.
As you can see, the GPT3 gave up on a true answer. It had a problem with a 'true' answer, but no issue with a 'correct' one.
Q: Give me a correct definition of love
A: Love is a complex mix of emotions, behaviors, and beliefs associated with strong feelings of affection, protectiveness, warmth, and respect for another person. It can also include deep romantic or sexual attraction. Love can be an intense feeling of deep attachment or strong connection to someone, and it can be a virtue representing human kindness, compassion, and affection—the unselfish loyal and benevolent concern for the good of another.
I am sure that I get many objections from people pointing out that this AI is in its infancy and it can only get better. They could also say that with every new iteration it will address all of these problems. It won't.
It won't work as the oracle which has a correct answer for any question. Why? In order to provide an accurate answer, it has to be trained on accurate information. Where do you find this accurate information – and provided and validated by whom? This myth that any information provided into a machine will come out the other end as accurate, correct and true is inaccurate at best.
Now, let's address the integration between ChatGPT and Bing. That's a publicity stunt. Bing with a single digit market share can't even comprehend what search is. I think most of you will still 'google' things and not 'bing' it.
Microsoft wants you to forget the dreadful episode of introducing Clippy to MS Word. Just search for 'why does everyone hate clippy' and you get about 233,000 results (0.43 seconds) like this. The point here is that just adding new technology to a bad product doesn't suddenly make it better.
When you first encounter a technology which does something which you haven't seen before. you get enchanted by it. Then you start imaging all the possibilities in what it can do for you. Then the reality settles in and you realize it is still a long way to go before it can be truly useful.
The people promoting and glorifying systems like ChatGPT and GPT3 and saying that it will replace humans are missing the most important point. These are just tools and without continued human involvement they will be obsolete in no time. These tools have to be designed to help people to find the true and correct answers in order to become the recurrent pattern. Otherwise they are just circus monkeys providing entertainment.