Updated: Oct 24
Nathan Benaich and Ian Hogarth’s The State of AI 2021 is finally here.
In its 4th edition now, this is one of the publications I am anxiously waiting for every year as it provides a great overview of the latest advances of AI, covering a broad spectrum of topics with enough depth to make it worth your reading time.
In the 188 slides of this year’s report you will find plenty of insight on the latest AI research, on the challenges around AI talent, examples of novel AI applications to business as well as a snapshot of the latest evolution in AI regulation and its impact on AI geopolitics.
The report then closes with the usual predictions on where AI will lead us in the next 12 months — rigorously scored every year in hindsight.
Here are some observations.
Health is the new AI black
The synchronized worldwide attention to health issues triggered by the COVID-19 pandemic has accelerated the amount of AI research in the health segment. And it’s an awesome and scary thing at the same time. From using auto-encoders to predict the effect of unseen combinations of novel drugs, to use the new Atomic Rotationally Equivalent Scorer to predict 3D structure of RNA molecules, to train language models on viral sequences to predict virus mutations, the application of AI to health research are truly fascinating.
Large Language Models applications provide a glimpse into a new era in Human Computer Interaction (HCI)
A little over a year from the introduction of GPT3 by OpenAI, we are seeing some exciting applications being created. This year’s introduction of Codex is pushing even more the envelope when it comes to using natural language to instruct a computer to create software but there is something really magical when OpenAI’s DALL-E application generates believable images when prompted with “an armchair in the shape of an avocado” (see image).
The AI race is on (and “the west” is not winning it)
If there is a country that emerges as seriously executing the strategy to become a AI world leader from this report, that is China. Whether you look at number of AI research published or number of STEM PhD students or the 1st and 2nd medals of the 2021 AI City Challenge (all Chinese institutions), the Chinese focus and determination is testified by their achievements.
The amount of research on AI Alignment is scarily low
Considering the implications of the adoption of intelligent systems across the board, I was particularly impressed by how few people are fully focused on studying how to ensure increasingly powerful AI systems have goals that are aligned with humanity: less then 100! (see image). Considering we have already military operations being carried out by AI-guided drone swarms (slide 175) to make an example, this is an alarming fact in my humble opinion.
Source: primary research by State of AI team
These are just few observations that come to my mind by reading the report over the weekend. I urge anyone interested in the topic to read the full report here.
CARMELO IARIA The AI Academy, Founder & CEO
We’d love to hear your voice, please comment on it and share your opinion with us!