AI performance is measured against human ability to perform the same tasks, and in many areas AI has already surpassed human ability—for example, in reading comprehension, image recognition, and language understanding. Previously, the tasks AI performed took years and decades to develop and the line was more horizontal, but during the last decade the performance curve has become nearly vertical. The types of tasks AI does were mostly aimed at assisting—for example, handwriting recognition was used by post offices to sort letters, and AI image recognition was used for sorting photos in photo banks.
The real change began in 2019 with predictive reasoning models, which have given AI the ability to not only assist with writing, speech, or image recognition, but to think in a logical way, and in a fraction of the time of a human counterpart. This change is the most pivotal and rapid yet, and the new lines continue to rise nearly vertically.
The ends of the wires on this graphic are loose as a new update is expected this year. We can already expect a few of these capabilities to have crossed the line. But which ones, and by how far?
Data: https://ourworldindata.org/artificial-intelligence
Helen Ilus 