This skill makes humans truly irreplaceable in times of AI. (And it’s not empathy or creativity).
Creativity and empathy are often considered as the last bastions of human advantage over artificial intelligence (AI). After all, AI lacks the ability to feel, though it doesn’t seem to need it. It already surpasses humans in some creative and empathetic tasks today. The real last bastion bears a different name: context.
The big question for the future is what role humans play in a world where intellectual abilities are replaced by AI. To this question, the Industrial Revolution provides a comparable precedent, where physical labor was gradually replaced by intellectual labor. Derived from this history, philosopher predict emotional and creative skills as the successors of intellectual abilities, referring to AIs lack of biological feelings. However, this is not entirely accurate.
Yes, GPT-4 is creative
.. and this is not just a reference to Midjourneys’ astounding ability to generate images.
Creative thinking can be measured with tests like the Torrance Tests of Creative Thinking (TTCT), designed to assess convergent and divergent thinking. Convergent thinking is essentially the ability to solve a creative problem in a targeted and coherent manner. Divergent thinking requires the problem to be solved in multiple, unique ways. Two examples of TTCT questions could be:
Convergent thinking: Imagine you discover an unknown island; write a brief story about your exploration and possible surprising discoveries.
Divergent thinking: What could be the possible uses of an empty beverage carton, apart from storing liquids?
Naturally, it was only a matter of time until GPT-4 was challenged with this test. Erik Guzik from the University of Montana [1] administered a TTCT questionnaire to GPT-4 and roughly 3,000 US college students. An external panel later evaluated the answers without knowing which came from ChatGPT. The result: GPT-4 outperformed 99% of its human competitors.
While we shouldn’t overinterpret this result, it’s clear that AI could easily compete with humans in some creative tasks today. And we’re still in the early stages of this impressive technology.
Yes, GPT-4 can also exhibit empathy
.. or at least more so than many physicians on Reddit’s AskDocs forum.
A study from the University of California [2] compared a sample of questions and responses from verified doctors with those of ChatGPT in terms of quality and empathy. Again, reviewers did not know which responses were from ChatGPT and which were from original experts. In both categories, the AI shone with significantly better answers, as can be seen in the distribution function below. Co-author Jessica Kelley described the comparison as follows “ChatGPT messages responded with nuanced and accurate information that often addressed more aspects of the patient’s questions than physician responses.” [2]
Admittedly, this is not a proof, but rather an indication that GPT-4 can also take on tasks of empathy. There are many other exciting articles on this topic that test ChatGPT’s cognitive-empathetic abilities (ChatGPT is capable of cognitive empathy! (Caution: Paywall)).
In general, it can be assumed that artificial intelligence will soon be able to compete with humans in various disciplines of empathy and creativity. But what ability then ensures that humans are not made redundant?
What’s missing, is context
I use GPT-4 in many areas of my life:
· As a business consultant, to discuss the details of work and concepts,
· as a hobby developer, to automate aspects of my daily life,
· as a politically engaged person, to delve into resolutions and complex issues in my municipality,
· and also privately, to ask for nutritional advice or to further educate myself
Of course, I notice the mistakes and made-up statements of GPT-4 in all different settings. Sometimes the answers are unspecific or useless, even after a long discussion. However, in all settings the proportion of insufficient answers is relatively low and will certainly decrease in future versions that OpenAI releases. Therefore, this is certainly not going to be the showstopper for AI, especially since human responses are also occasionally flawed.
Most often, however, I noticed that the cause of an insufficient answer lies in the fact that GPT does not know the context in which I pose my prompts. For example, to answer the question “what should I eat today?” GPT would need to know what I have already eaten that before, what my culinary preferences are, what I am allergic to, and much more. If I don’t provide this context, it recommends a Greek salad; which I don’t like :(.
The answer becomes more satisfying with context (see figure). So, it seems to be my job as human to formulate my questions precisely and provide the necessary context.
The same applies to developper, consulting or political tasks. When I ask GPT to analyze a resolution for building a road and suggest improvements, the result is often superficial and therefore useless; because it lacks context. It doesn’t know the citizens’ desires, road usage statistics, or the estimated cost-benefit ratio. So far, only humans can provide this knowledge, because only humans have access to it. We are fortuned to have a well-trained system of brain and senses to physically grasp any situation in the real world. We have a lot of prior knowledge, which we’ve gained from conversations, studying and experiences in the past. Also we have the ability to independently obtain information by using internet access or communicating with other humans and machines. As long as artificial intelligence does not have comparable abilities, there will be a need for us to make the world understandable for AI.
And even though there are already many impressive tools that can mimic any parts of our human existence - logic, creativity, empathy, the ability to hear, speak and see, or performing research on the internet — it will probably take some time before a functioning system of robotics and AIs is able to keep up with our innate understanding of context and situations. This skill will keep us distinctly human in our rapidly evolving tech world (at least for a while).
