According to a recent assessment by AI observers, text-generating applications like ChatGPT and image-generating applications like MidJourney consume information faster than humans can produce it. This means that by 2026, the “high quality” content that can be used to train these AI systems may become scarce.
The study’s authors also emphasized that “quality” content, such as books, scientific research, and computer code, are expensive to produce and thus will be the first to be affected by this scarcity. However, it’s worth noting that less high-quality content, such as videos on YouTube and other online content, may not be affected until 2050, as the total quantity of this type of content is difficult to measure.
The researchers who conducted this study also noted that the amount of data used to train these AI systems is growing at a rate of 50% per year, while the availability of “quality” content is growing at a rate of only 7% per year. This disparity in growth rates is what is leading to the prediction of the 2026 “wall” in terms of knowledge accumulation.
It’s worth noting that in the context of this study, the AI system in question is primarily an application that learns to converse, so the quality of the “language data” may not be as important as the overall quantity of data used to train it. This means that the exhaustion of texts produced by humans could potentially be compensated by texts produced by AI systems.