The State of AI report is produced by AI investors Nathan Benaich and Ian Hogarth. It analyses the most recent and relevant developments in AI and is updated at the end of every year. If you call yourself data scientist or machine learning engineer this report is a must in your reading list.
Full State of AI Report Here
This report covers five thematic areas, namely, Research, Talent, Industry, Politics and Predictions. After going through the last report from 2020 (177 slides/pages), I was impressed and felt like I needed to share a few highlights worth mentioning in each of its thematic areas.
- Only 15% of research papers on AI publish their code
- Facebook’s PyTorch is fast outpacing Google’s TensorFlow in research papers.
- Pytorch is more popular than TensorFlow in paper implementations on Github.
- A new generation of Transformer language models (GPT-3, T5, BART) is unlocking new NLP (Natural Language Processing) use-cases.
- AI can now repair buggy programs/algorithms.
- Papers in Biology are growing FAST. Over 21,000 papers in 2020 alone.
- Deep learning on cellular microscopy accelerates biological discovery with drug screens.
- Reinforcement Learning agent designs molecules, thus ensuring they can be synthesized in the lab.
- Big contributions on analysing COVID-19 datasets in order to find new symptoms, and predicting COVID hotspots.
- Increased in research about Federated Learning. A concept started by Google five years ago.
- Between 2004 to 2018 52 tenured professors from US Universities have been hired by either Google, DeepMind, Amazon or Microsoft. Most of those professors left in 2018 alone.
- The US knows is trying to keep AI talent. Almost 88% of Chinese students who earn an American PhD are retained in the US for work.
- The majority of top AI researchers working in the US did not study in America (the continent, not the country).
- There has been a backlash to Trump’s proclamation to suspend H1-B visas. Eight federal lawsuits and hundreds of Universities object.
- US States continue with the legislation of Autonomous vehicles policies.
- 3 out of 66 companies with AV testing permits in California are allowed to test without safety drivers since 2018 (Waymo, NURO, Autox)
- Tractable’ AI captures and processes imagery of vehicles accident damage to automatically predict its repair costs. This enables faster accident and disaster recovery intervention.
- Computer vision is used to detect subtle evidence of tampered identity documents.
- self-driving milage in California remains small (microscopic)compared to human driving. self-driving mileage is around 3 million whereas human mileage is around 391 billion.
- Ethical risks of implementing Machine Learning in certain sensitive contexts have been published.
- 50% of the world allows the use of facial recognition. Only three countries have partially banned this technology (Belgium, Luxembourg and Morocco).
- On January 2020 a Detroit police arrested Robert Williams after facial recognition algorithm incorrectly matched the photo on his driver’s license with blurry CCTV footage.
- Facebook’s automatic photo-tagging was in violation of Illinois’ 2008 Biometric Privacy Act. Facebook settled a class-action lawsuit for $650M.
- Startups which intersect AI and defense have raised large financing rounds.
- The Chinese government sets up an additional $29B state-backed fund to reduce its dependency on American semiconductor technology.
- A new wave of countries declaring national AI strategies.
In this section, there are eight predictions. My favourite is that Facebook is expected to make a major breakthrough in augmented and virtual reality with 3D computer vision. For further details on the other predictions or to deep-dive in the State of AI report I invite you to check the full report on the following link:
Full State of AI Report Here