Assessing the Potential of ChatGPT: Lessons from the History of Innovation

[Version in French here]

Unless you’ve been living on Mars for the past few weeks, you couldn’t escape news about ChatGPT, the artificial intelligence tool that answers all your questions: summarizing an article, informing you about the economic crisis, writing a poem, etc. As with any new technology, it is presented as revolutionary by some and futile, useless, or even dangerous by others. While it will take time for the dust to settle, we can nevertheless avoid some of the pitfalls, and above all, the clear-cut positions, by relying on the history of innovation, which offers at least seven lessons for a more nuanced approach to the debate.

The history of innovation is complicated. It is littered with technologies that promised a great future but failed to deliver. Others took years to succeed: the first automobile was invented in 1765, but it wasn’t until the late 1880s that one could buy one. Others, finally, were born in indifference, totally underestimated at their beginning. The Wright brothers’ first flight in 1903, a historic event, made three lines in the local newspaper and it took a very long time to see aviation as anything other than a whim of the rich. The greatest minds have been wrong in their estimation of the potential impact of a new technology. However, here are seven historical lessons to better reflect on the potential of ChatGPT.

1) The success of a new technology is rarely due to its technical performance alone. Innovation diffusion is a social process: the social body accepts the technology but by its use generally leads to a modification. It is therefore a complex process. A very good technology can therefore be rejected because of social criteria linked to dominant mental models. For example, GMOs are rejected in France because there is a mental model “GMO = danger” that has successfully opposed the model “GMO = excellent solution”. The economic, social, and political consequences of a new technology are therefore impossible to predict. What happens to a technology is the product of a process that is both technical (its invention and improvement) and social (its adoption, its rejection, its adaptation for sometimes unexpected uses). When we think about the potential of ChatGPT, we cannot limit ourselves to a discussion of its technical performances. We know, for example, that some of its responses will necessarily go against the beliefs of certain groups, which may lead to hostile reactions. ChatGPT will therefore be “moderated” (i.e., censored), which will lead to further backlash.

2) Any technology, like any tool, has its limits; it is not on these limits that it should be judged. None is universal. Indeed, the limits of ChatGPT were very quickly pointed out: lack of references, questionable positions, sometimes bizarre results, lack of creativity, weaknesses in certain tasks, etc. Understanding these limitations is critical to determining where the technology will be relevant and where it will not. Pointing to these limitations and dismissing it outright is a mistake. An image creation software can’t turn you into an artist, but that’s not why it’s not useful. One should not reject a technology because of its limitations, but focus on its potential, to understand what it allows us to do in a new way.

3) There is a tendency to judge a new technology by the yardstick of current technology. However, a new technology introduces new performance criteria, and it is on these that it must be judged. 3D printers do not offer the quality of traditional factory manufacturing, so they are compared unfavorably to it, but that is not what they are trying to do. They provide, in areas where the quality they offer is sufficient, a very useful flexibility and customization. So, they are relevant for some uses, and not for others. The fact that the performance of a new technology is inferior in some respects and superior in others explains why it rarely replaces the existing technology completely. We continue to use traditional ovens in addition to microwave ovens, propeller planes and not just jets, paper diaries and not just electronic ones, etc. Sometimes, the new technology will remain inferior on historical criteria (a cell phone can run out of battery or lose the network, unlike a landline phone), and sometimes it will end up surpassing the current technology on all criteria, as was the case for digital photography from the 2000s onwards, in which case the switchover becomes total, and the old technology disappears.

4) A new technology improves its performance over time. To evaluate it, we must therefore look not at where it is at the beginning, but where it can go, which is obviously difficult, if not impossible. The online translation service Google Translate was launched in 2006. What magic to submit a text that came back translated in a few seconds! Of course, it required a lot of rework. Maybe GT did about 50% of the work, but the time saving was considerable. Then, the quality gradually improved. In 2017, DeepL was released, which marked a significant improvement. A translated text now required only a few minutes of finishing; we reached about 80%. Nowadays, the quality has become remarkable. There is almost no need to rework the text. If machine translation had been judged against human translation in 2006, it would have been definitively rejected.

Adoption and rejection of new technology

5) A new technology tends to be adopted first by non-consumers. The quality of machine translation as of 2006 was much lower than that of a professional translator, except that I didn’t have the time or money to pay for one, and none of them would work to translate three paragraphs here or there. Despite its mediocre performance, machine translation was already of great help. For me, the alternative was GT or nothing, and no matter how mediocre it was, GT was better than nothing, so its performance was good enough for me as a non-consumer. That’s why a new technology is adopted by non-consumers despite its limited performance. It gives them something they couldn’t have before.

6) Users of the existing technology tend to reject the new technology. It is the corollary of the previous point. This is because the performance of the new technology is generally not sufficient for them. I started using Internet telephony in 1998. The quality was terrible, but I could call abroad for the price of a local call. In other words, the performance of Internet telephony was good enough for me, considering the cost. For a business, on the other hand, it was out of the question to use it, as the performance was insufficient for its requirements. For them, it didn’t work, which means: “In terms of my requirements, its performance is insufficient, so I can’t use it”. Logically, businesses initially rejected Internet telephony, which took off through individual use.

7) We tend to put forward the disadvantages of a new technology rather than its advantages. Talk about robots, and you get an answer about unemployment. Talk about biotech and genetics, you get Frankenstein. Talk about AI, you get Skynet and machine domination. This is not new. The advent of photography made people fear the disappearance of artists, and of radio that of musicians. We focus on what will disappear, or risks disappearing, without imagining what can be created: the printed book of Gutenberg made the magnificent, illuminated manuscripts, true works of art, disappear, thus depriving the copyist monks of a source of income, but it opened reading and writing to all. With books, a scholar no longer had to learn texts by heart; he could therefore use the capacity of his brain thus freed for much more creative tasks. It is impossible not to think of a similar effect for ChatGPT, which will free us from some tasks to do other, more creative ones.

Homo ludens

The history of innovation is complex, and it is difficult, if not impossible, to predict the impact of a technology on society. Its success often depends on social factors and societal mental models, and its potential should not be judged solely on its limitations or technical performance. Its success depends on what will be done with it, and thus on a strange dance between its creators and its users. And so, if you want to make up your mind about ChatGPT, start using it to discover its strengths and weaknesses. Take advantage of the former and be forgiving of the latter; watch it evolve, like we did for machine translation. In short, ignore the moral preachers and overconfident experts, and play with it. Have fun!

😉 The title of this article was created by ChatGPT.

📖 If you want to know more about the management of disruptions, you can read my book: A Manager’s Guide to Disruptive Innovation.

📬 If you enjoyed this blog post, why not subscribe?

2 responses to “Assessing the Potential of ChatGPT: Lessons from the History of Innovation

  1. Pingback: Sept leçons de l’histoire de l’innovation pour mieux comprendre le potentiel de ChatGPT, l’outil d’intelligence artificielle controversé | Philippe Silberzahn

  2. Pingback: Sept leçons de l'histoire de l'innovation pour mieux comprendre le potentiel de ChatGPT, l'outil d'intelligence artificielle controversé - Philippe Silberzahn

Leave a Reply