Microsoft releases Seeing AI for blind users
Microsoft's free Seeing AI app, released in 2017, narrates text, scenes, currency, and faces for blind and low-vision users.
China's New Generation AI Development Plan (2017)
China's State Council issues a national strategy setting a three-step goal to make the country the world's primary AI innovation center by 2030.
The Asilomar AI Principles
A 2017 set of 23 principles for beneficial AI, drafted at a Future of Life Institute conference and signed by thousands of researchers.
Apple puts a Neural Engine in the iPhone with the A11 Bionic
In 2017 Apple's A11 Bionic shipped a dedicated Neural Engine doing 600 billion operations per second, bringing on-device machine learning to the iPhone.
Apple unlocks the iPhone with a face and an on-device neural engine
Apple introduces Face ID on the iPhone X, mapping a face with 30,000 infrared dots and matching it on a dedicated neural engine chip.
Intel announces Loihi, a self-learning neuromorphic chip
In 2017 Intel revealed Loihi, a neuromorphic research chip with 130,000 spiking neurons that learns on-chip using brain-like timing rules.
AlphaGo Zero learns Go from scratch
AlphaGo Zero reached superhuman Go strength using only self-play, with no human games, and beat the version that had defeated Lee Sedol by 100 games to 0.
Replika launches, an AI friend born from grief
Replika launched publicly in November 2017 as an AI companion app, grown from a chatbot one founder built to remember a friend who had died.
The Slaughterbots video campaign
In November 2017 the Future of Life Institute released Slaughterbots, a dramatized film warning about swarms of autonomous killer drones.
AlphaZero masters chess, shogi and Go
DeepMind's AlphaZero taught itself chess, shogi and Go from scratch by self-play, reaching superhuman strength in each and defeating the best existing programs.
Apple Deploys Local Differential Privacy at Scale
Apple detailed shipping local differential privacy to hundreds of millions of devices, gathering usage data without seeing any user's raw data.
Microsoft and Facebook launch ONNX for model interoperability
ONNX, an open format letting AI models move between frameworks, was introduced by Microsoft and Facebook in 2017 with its V1 released that December.
Microsoft commits $50 million to AI for Earth
In December 2017 Microsoft pledged $50 million over five years to put AI tools in the hands of environmental groups.
A neural network finds an eighth planet around Kepler-90
In 2017 a Google neural network sifted Kepler data and found Kepler-90i, making Kepler-90 the first known eight-planet system besides our own.
A neural network finds an eighth planet around Kepler-90
In 2017 a Google neural network sifting Kepler data found Kepler-90i, giving a distant star eight planets like our own Sun.
Amazon Go opens its first checkout-free store to the public
Amazon opened its first Just Walk Out store to the public in Seattle, using computer vision and sensor fusion to skip checkout.
Viz.ai wins FDA clearance for the first AI stroke triage software (2018)
In February 2018 the FDA granted Viz.ai De Novo clearance for the first AI triage software, which flags large-vessel-occlusion strokes on CT scans.
S&P Global acquires Kensho for about $550 million
In 2018 S&P Global bought AI analytics firm Kensho for roughly $550 million, one of the largest AI acquisitions in finance.
Skydio R1, a fully self-flying camera drone
Skydio shipped the R1, a consumer drone that navigated and filmed entirely on its own using onboard cameras and deep learning, no piloting required.
FDA authorizes IDx-DR, the first autonomous AI diagnostic
The FDA authorized IDx-DR in 2018, an AI that screens for diabetic retinopathy from retinal images without a clinician - the first autonomous AI diagnostic.
Google Duplex demonstrates AI phone calls
Google reveals Duplex, an AI that makes phone calls in a human-sounding voice to book appointments, complete with ums and pauses.
Google's Lookout app narrates the world for blind users
Google's Lookout, announced in 2018, uses a phone camera and computer vision to speak aloud the objects, text and people around a blind user.
OpenAI's AI and Compute analysis
OpenAI's 2018 analysis found compute in the largest AI training runs had doubled every 3.4 months since 2012, the case that computation drives AI capability.
Japan's Article 30-4: a permissive AI-training exception
Japan amended its Copyright Act to broadly allow using copyrighted works for data analysis, making it one of the most AI-training-friendly regimes.
India's National Strategy for AI (#AIforAll)
India's NITI Aayog published its National Strategy for AI in 2018, branded #AIforAll and aimed at inclusive growth.
Databricks open-sources MLflow for the machine-learning lifecycle
On June 5, 2018, Databricks released MLflow, an open-source platform for tracking experiments, packaging code, and managing models across the ML lifecycle.
OpenAI's first GPT
OpenAI's first GPT showed that pre-training a Transformer on large amounts of unlabeled text and then fine-tuning it set a new bar across many language tasks.
The Pentagon establishes the Joint AI Center
In June 2018 the US Defense Department created the Joint Artificial Intelligence Center to coordinate AI across the military.
DeepMind and Moorfields match experts at diagnosing eye disease
A 2018 Nature Medicine study showed a DeepMind-Moorfields AI could refer over 50 eye diseases from OCT scans as accurately as expert clinicians.
DeepMind Hands Data-Centre Cooling to Autonomous AI Control
DeepMind moved from AI recommendations to AI directly controlling Google data-centre cooling, saving about 30 percent of cooling energy.
BERT brings deep bidirectional pre-training to language
Google researchers release BERT, a pre-trained bidirectional Transformer that set new state-of-the-art on eleven NLP benchmarks.
Portrait of Edmond de Belamy sells at Christie's for $432,500
The first AI-generated artwork auctioned by a major house sold at Christie's for $432,500, far above its $7,000-$10,000 estimate.
SpiNNaker reaches a million cores to simulate the brain
In 2018 Manchester's SpiNNaker neuromorphic machine reached a million ARM cores, built to simulate spiking neurons in biological real time.
Google's Night Sight brings computational photography to near-darkness
In 2018 Google's Night Sight let Pixel phones shoot clean handheld photos in light too dim to see, by merging many frames in software.
CIMON, the first AI assistant on the space station
In 2018 the free-flying robot CIMON became the first AI-powered crew assistant to operate aboard the International Space Station.
Rio Tinto runs AutoHaul, the first long-distance autonomous heavy-haul railway
Rio Tinto's AutoHaul became the world's first heavy-haul, long-distance autonomous rail network, moving iron ore across the Pilbara.
The Montreal Declaration for Responsible AI
Launched in 2018 after a year of public deliberation, the Montreal Declaration set out 10 principles for responsible AI development.
Apple Watch ships an FDA-cleared ECG and AFib notification
In December 2018 Apple put an FDA-cleared single-lead ECG and irregular-rhythm notification on the wrist, taking consumer heart screening mainstream.
Google launches Live Transcribe for the deaf
Google's Live Transcribe, launched in 2019, gives deaf and hard-of-hearing users real-time captions of nearby speech on Android phones.
OpenAI's GPT-2 shows language models can learn tasks unsupervised
OpenAI's 1.5-billion-parameter GPT-2, trained only to predict text, performed many language tasks with no task-specific training.
DeepMind Uses Machine Learning to Boost the Value of Wind Energy
DeepMind applied neural networks to 700 MW of wind power, predicting output 36 hours ahead and raising its value by about 20 percent.
MIT Mini Cheetah becomes the first four-legged robot to backflip
MIT's 20-pound Mini Cheetah quadruped did a 360-degree backflip, the first four-legged robot to pull off the maneuver.
Sotheby's auctions its first AI artwork, Klingemann's Memories of Passersby I
Mario Klingemann's GAN installation that generates endless portraits in real time became the first AI artwork sold at Sotheby's.
Computational imaging produces the first black hole picture
In 2019 the Event Horizon Telescope reconstructed the first image of a black hole using computational imaging algorithms.
The first image of a black hole and its imaging algorithms
The 2019 Event Horizon Telescope image of M87's black hole was reconstructed by independent teams using CLEAN and regularized maximum likelihood methods.
OpenAI Five beats the Dota 2 world champions
OpenAI Five became the first AI system to defeat the reigning world champions at an esport, beating Dota 2 champions OG in April 2019.
EU DSM Directive: the text-and-data-mining opt-out
The EU's Digital Single Market Directive let rightsholders opt out of commercial text-and-data mining, shaping how AI can be trained in Europe.
MineRL Minecraft competition at NeurIPS 2019
MineRL ran a NeurIPS competition to obtain a diamond in Minecraft using few samples and human demonstrations; no team succeeded that year.