With the rise of generative AI, such as ChatGPT, the future of human intellect is in doubt. Perhaps it could replace human jobs. Research shows that with generative AI, 30 percent of hours worked today (2023) could be automated by 2030 in the US (Ellingrud et al., 2023). As a Business Information Management student, this is exciting (from a business perspective): what could we do with those 30% extra hours in the future? On the other hand, as a human being: should we be worried that not all people will have the same opportunities to strive for a job?
One might argue that generative AI, such as ChatGPT, should replace some humans since it is very smart. A clinical psychologist even tested ChatGPT’s IQ to see what the results would be. Since it is not a human, it shows no signs of anxiety and distraction. What did ChatGPT score on the IQ test? A massive IQ of 155, meaning it would be the top 0.01 percent smartest human (Roivainen, 2023).
Although these findings are impressive, we should note that ChatGPT is not a human. ChatGPT can only do tests on the Verbal IQ scale (Vocabulary, Similarities, Comprehension, Information and Arithmetic) since it has no other senses (required for the nonverbal IQ test). This means that the IQ score of 155 is only for verbal communication. It can’t even do simple tasks as ‘What time is it now?’. It will apologize (almost as a human) that it has no capability to provide real-time information and gives me some instructions how to find the current time through a clock or watch.
Furthermore, ChatGPT (or any other language AI model) can’t operate without all the humans working behind it to guide the AI model. Since these models predict characters, words and sentences based on training data sets, feedback is very important. This is the reason why ChatGPT has the option to give feedback on how the generated response was (good or bad, and why). It can’t give any feedback to itself (Us, 2023). To teach ChatGPT not to give answers which would be racist, sexist or any other unwanted response, hundreds of workers from Kenya are paid as little as $2 an hour to do this work (Perrigo, 2023).
As described in this blog: ChatGPT is smart, but not smarter than you. It has some real potential but also some awful flaws.
Literature
Ellingrud, K., Sanghvi, S., Dandona, G. S., Madgavkar, A., Chui, M., White, O., & Hasebe, P. (2023). Generative AI and the future of work in America. In McKinsey & Company. https://www.mckinsey.com/mgi/our-research/generative-ai-and-the-future-of-work-in-america#/
Perrigo, B. (2023, January 18). Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic. Time. https://time.com/6247678/openai-chatgpt-kenya-workers/
Roivainen, E. (2023, March 28). I gave ChatGPT an IQ test. Here’s what I discovered. Scientific American. https://www.scientificamerican.com/article/i-gave-chatgpt-an-iq-test-heres-what-i-discovered/
Us, J. P. N. C. (2023, August 24). ChatGPT and Other Language AIs Are Nothing without Humans. Scientific American. https://www.scientificamerican.com/article/chatgpt-and-other-language-ais-are-nothing-without-humans/
This is a very interesting way to think about it. I particularly like that you move from discussing ChatGPT’s intelligence to its ability to learn, which is arguably a massive part of true high intelligence. At first I was surprised at how much work it takes to get ChatGPT to give less unwanted responses, but it makes sense considering how much data it is trained on. One bit of feedback is not going to radically change it. Then I thought back to your intro about AI replacing work, but given its apparent inflexibility it seems like a real pain to get an AI to learn and adjust to the specifics of a particular business, don’t you think? So maybe the 30% is simple stuff, but do you think AI might ever be able to learn new things faster than humans, and fully replace workers whose job involves actual thinking? Or might the massive amount of data it needs before learning be a persistent problem?
Very interesting blogpost Thijs! Learning about the (lack of) intelligence of ChatGPT is very interesting and brings great insights in the program. ChatGPT is very smart and can be used for verbal tasks like almost no human could, but when asked anything non-verbal it just isn’t the way to go. I’m wondering is this will change in the future. Maybe there will become a way that ChatGPT or another GPT model (like Bing uses) will be able to act based on real-time data. Will that change the capabilities and should we be worried about replacing more than the 30% you mention?