In response to Kai-Fu Lee and Chen Qiufan's A2041, I put together my thoughts on AI in 2021.
Natural Language Processing (NLP)
In the 'information age', communication is both a conduit of value and a source of it. Providing automated yet personalised customer service interactions, generating copy for manuals and advertising, generating new stories in the styles of old writers and being able to converse so naturally that we can bond to them — these are some of the tantalisingly realisable promises of machines that can talk.
Communication with machines has been one of the most coveted capabilities of artificial intelligences, as demonstrated by the prominence of the Turing Test. Natural Language Processing is one of the most heavily invested areas for applied A.I, capturing the broader imagination in the 2013 film, Her.
In October 2021, Microsoft and NVIDIA announced that they had collaborated to built the world's most sophisticated NLP model, MT-NLG (530 billion machine learning parameters), surpassing OpenAI's GPT-3, which was released in just June 2020 (175 billion parameters).
Medical — Assisted Diagnosis and Protein Folding
A second crucial sector for Applied A.I. is the medical sector. In particular Google DeepMind's breakthrough in modelling protein folding in November 2020 (AlphaFold), will drastically speed up the rate of drug discovery and potentially have downstream impacts on genomics as the effects that compounds may have in specific genetic environments can be modelled.
While areas like robot process automation will disrupt many professional services, like investment, insurance and law, AlphaFold stands out, because it isn't replicating a human task that is only impressive because it is being done by something non-human, like telling apart cats and dogs. Nor is it merely increasing the accuracy of a task that is carried out by humans, like playing chess, driving a car or recognising malignancy in tumours.
Prediction of protein folding is an area is practically impossible for humans to do to any significant scale even with advanced statistical methods. This is a clear departure from merely mimicking human intelligence in a specific domain.
Although much of the story around applied A.I. is about the replacement of humans, the true value of the technology will come in those use cases where A.I. does not only outperform humans but either augments human capabilities in parallel or is completely orthogonal to them, providing novel and unprecedented avenues for generating value.