By this point, most people assume that AI has already surpassed humans in every single skill imaginable, but how realistic is this? Contextual AI has provided data that can shed some light on the matter at hand, and it turns out that AI has matched humans in a wide array of skills with all things having been considered and taken into account. Image recognition came first in 2015 when the ImageNet database was used to train an AI to teach it to tell images apart.
With all of that having been said and now out of the way, it is important to note that speech recognition came next thanks to the Switchboard database in 2017. Following this, two further important skills were acquired in 2018, namely that of handwriting recognition with the MNIST database, as well as reading comprehension with the SQuAD database.
By 2020, the GLUE database had enabled AI to match human skills in terms of language understanding, and the HellaSwag database facilitated AI meeting human level acumen in Common Sense Completion as well. AI has surpass humans in the fields of image recognition, reading comprehension and language understanding. Other skills such as handwriting recognition, speech recognition and common sense completion are at more or less human levels.
In spite of the fact that this is the case, there are still some areas where AI is lagging behind. Grade school math and code generation are two fields where humans are still able to do a better job than AI, though rapid advancements might make AI surpass us sooner rather than later.
However, one thing that bears mentioning is that datasets could plateau by 2026, which would make it harder for AI to advance than might have been the case otherwise. As high quality datasets begin to run out, AI coders and developers might need to get creative otherwise their tech will not be able to progress beyond a certain point. It will be interesting to see how OpenAI and other companies in this space meet these challenges since they pose an existential risk to them in the long run.
H/T: VisualCapitalist
Read next: New Study Claims 81% Of Americans Are Worried About How Their Personal Data Is Being Used By Companies
With all of that having been said and now out of the way, it is important to note that speech recognition came next thanks to the Switchboard database in 2017. Following this, two further important skills were acquired in 2018, namely that of handwriting recognition with the MNIST database, as well as reading comprehension with the SQuAD database.
By 2020, the GLUE database had enabled AI to match human skills in terms of language understanding, and the HellaSwag database facilitated AI meeting human level acumen in Common Sense Completion as well. AI has surpass humans in the fields of image recognition, reading comprehension and language understanding. Other skills such as handwriting recognition, speech recognition and common sense completion are at more or less human levels.
In spite of the fact that this is the case, there are still some areas where AI is lagging behind. Grade school math and code generation are two fields where humans are still able to do a better job than AI, though rapid advancements might make AI surpass us sooner rather than later.
However, one thing that bears mentioning is that datasets could plateau by 2026, which would make it harder for AI to advance than might have been the case otherwise. As high quality datasets begin to run out, AI coders and developers might need to get creative otherwise their tech will not be able to progress beyond a certain point. It will be interesting to see how OpenAI and other companies in this space meet these challenges since they pose an existential risk to them in the long run.
Read next: New Study Claims 81% Of Americans Are Worried About How Their Personal Data Is Being Used By Companies
