AI deforests the knowledge’s ecosystem

Big-tech’s dash to incorporate ChatGPT-like interfaces into their search engines threatens the ecosystem of human knowledge with extinction. Knowledge development is a social activity. It starts with scientists publishing papers and books that build on earlier ones and with practitioners, journalists, and other writers disseminating these findings and their opinions in more accessible forms. It continues through specialized web sites, blogs, the Wikipedia, as well as discussion and Q&A forums. It further builds upon our interactions with these media through web site visits, upvotes, likes, comments, links, and citations. All these elements combined have yielded a rich global knowledge ecosystem that feeds on our interactions to promote the continuous development of useful and engaging content.

ChatGPT and its siblings work by sucking in huge quantities of existing text and using that to train a neural-like structure that models the text’s language — a so-called large language model. They then use that to provide eloquent answers to arbitrary questions. The answers are often surprisingly useful, saving us from the painstaking work of searching, analyzing, and synthesizing an answer from diverse web sources. The problem is that this process robs the knowledge ecosystem from the interactions we would have had in it. Each AI engine reads the content only once for building its model. From then on, all our interactions happen between us, the AI engine, and the model, rather than with the original content.

Our disappearance from the web sites that hosted the original content removes its creators’ tangible and intangible incentives, such as ad revenue, professional advancement, fame, and even the altruistic pleasure of helping other humans. We stop seeing the articles’ authors, “liking” nice posts, watching related ads, and mentally thanking Wikipedia’s volunteers. Lower incentives will naturally shrink the development of original content at the same time as its creators will be competing with low-cost AI-generated stuff. The result is a knowledge ecosystem dominated by AI engines that take out of it substantially more value than they put in. As the publisher and technologist Tim O’Reilly presciently remarked at the beginning of this century, such ecosystems eventually deplete and die.

The bleak prognosis of the AI’s dominance effects is that the development of original knowledge will drastically diminish. Creative authors are unlikely to put effort into writing something that only a handful of AI engines will ever read. Scientists may continue to write papers with original results, but they are also likely to work less on parts that can be written by ChatGPT. Less content and development of it will result in ever fewer corresponding interactions, diminished liveliness, and thereby in a vicious cycle suppressing large parts of knowledge production. The deforested knowledge ecosystem will be dominated an AI monoculture endlessly regurgitating already known facts and plausible-sounding falsehoods.

Sluggish knowledge growth is a sad but not an extraordinary situation. What is exceptional is the exponential growth of knowledge we have witnessed in the past couple of centuries. Over most of our history human knowledge has been stagnant. Now, however, lethargic development of new knowledge is worrying, because the big challenges we face as a species — global warming, demographic aging, pandemics — can only be addressed with substantial scientific and technologic advancements.

What to do? The AI answer engines currently being developed need not act as one-way streets from accumulated knowledge to answers. They can be enhanced to provide feedback from their interactions with us to new types of collaboration platforms where knowledge creators will be rewarded for their used work, will be able to see existing and emerging knowledge gaps, and will be suitably incentivized to address them. AI’s intermediation can thus result in more focused and purposeful human involvement in knowledge development. The shift will mirror earlier changes in humans’ role from ploughing fields, to making clothes, to consolidating accounts. The new form of knowledge creation will take place not in the rich forest-like ecosystems we have now, but in landscapes resembling cultivated fields or manicured gardens. It remains to be seen where this will lead us, but a new knowledge revolution cannot be ruled out.

Comments   Toot! Tweet Share