Google AI 2018: Breakthroughs in Ethics, Quantum Computing, and AutoML

Ethical Principles and AI

Google released its AI Principles to guide responsible AI development, emphasizing fairness, accountability, and reducing bias in products such as Google Translate and image datasets.

Social Impact AI

Google applied AI to flood prediction, earthquake aftershock forecasting, and collaborated with external researchers using TensorFlow for scientific challenges, also launching the Google AI for Social Impact Challenge with $25 million in grants.

Assistive Technologies

Research on ML‑driven assistive tools produced Google Duplex, Smart Compose, Sound Search, and MobileNet‑based features that improve user interaction and accessibility.

Quantum Computing

Google advanced quantum computing with the 72‑qubit Bristlecone processor, exploring quantum supremacy and releasing the open‑source Cirq framework for quantum algorithms.

Natural Language Understanding

Improvements to the Transformer led to the Universal Transformer and the development of BERT, a deep bidirectional language model that set new state‑of‑the‑art results on 11 NLP tasks and the GLUE benchmark.

Perception Research

Work on image, audio, and video perception improved Google Photos, Lens, and Cloud ML APIs, introduced MobileNetV2, MorphNet, and multimodal models such as Looking to Listen.

Algorithms and Theory

Google explored optimization, distributed combinatorial optimization, algorithmic fairness, and privacy‑preserving learning, earning awards such as the ICLR 2018 Best Paper.

Software Systems

Key system contributions included TensorFlow 1.0 dynamic control flow, Mesh TensorFlow for model parallelism, JAX for accelerated NumPy, and research on ML‑enhanced compilers, networking, and storage.

TPU

Tensor Processing Units powered large‑scale training of models like BERT, were made available via Cloud TPU Pods and Colab, accelerating research across Google products.

Open‑Source Software and Datasets

TensorFlow celebrated its third birthday with major releases, new tools, and extensive community contributions; Google also released Open Images V4, AVA, YouTube‑8M, HDR+ Burst, and Google Landmarks datasets.

Robotics Research

Advances in robot learning enabled grasping of unseen objects and online deep reinforcement learning on real robots, earning best‑paper awards at CoRL ’18 and ICRA ’18.

AI in Other Domains

Machine learning was applied to physical and biomedical sciences, including high‑precision neuron reconstruction, pathology, and clinical prediction from electronic health records, with results published in Nature journals.

Research Outreach

Google supported external researchers through PhD fellowships, AI Residency, Faculty Research Awards, and organized workshops worldwide, fostering a vibrant research ecosystem.

Looking Ahead to 2019

The blog concludes with excitement for broader impact of Google’s AI research in the coming year.