In a fascinating report on the State of AI put out by venture capitalist, Nathan Benaich, and angel investor, Ian Hogarth, Benaich explains why they partnered to produce this report, “We believe there is a growing need for accessible, yet detailed and accurate information about the State of AI across several vectors (research, industry, talent, politics and China.) The purpose of our report is to drive an informed conversation about AI progress and its implications for the future.”
Reinforcement Learning
This report analyzed AI papers over the past 25 years and found immense growth in the publication of papers focusing on machine learning and reinforcement learning. Over 50% of the papers discussed machine learning topics and 15% were about reinforcement learning.
Reinforcement Learning is where software agents learn goal-oriented behavior by trial and error in an environment that rewards or penalizes the agent’s action to get them to achieve a certain goal. Most of this research right now is going on through gaming with an effort to beat human performance. The goal is to lead to play driven learning for robots and in turn getting robots to learn dexterity using simulations and curiosity driven exploration, and then move on to production environments.
Natural Language Processing
NLP had a big year with pretrained language models substantially improving performance on a variety of NLP tasks like processing large amounts of text, cross referencing, summarizing and deriving signals from large amounts of data.
Machine Learning for Life Sciences
Deep Learning is allowing expert level diagnosis and treatment referral suggestions in the life sciences. Two deep convolutional neural networks (CNN) were shown to work in concert to significantly outperform state-of-the-art, far earlier than expected. Neural networks are decoding thoughts from brain waves, restoring limb control for the disabled and helping machines learn how to synthesize chemical modules.
The U.S. FDA cleared three AI based diagnostic products in 2018. On 4/11/18 IDx software was approved for diabetic retinopathy detections from eye exams, on 5/24/18 Imagen software was approved for detecting wrist fractures in adult patients from 2D xray images, and on 11/7/18 MaxQ software to prioritize the clinical assessment of adult non-contact head computed tomography cases that exhibit indications of intracranial hemorrhage was approved.
Multiple pharma companies are partnering with AI driven drug development companies. Atomwise and Charles River are using convolutional neural networks to predict the binding capacity of small molecule drugs to target proteins of interest. Exscientia and Celgene claim to reduce the time to discover pre-clinical drug candidates by at least 75%. LabGenius and Tillotts Pharma are using AI driven evolution strategies to develop radically improved protein therapeutics targeting inflammatory bowel disease. Insitro and Gilead are creating disease models for nonalcoholic steatohepatitis (NASH), a chronic liver disease that can result in cancer.
Demand Forecasting
Demand forecasting in a variety of industries is being dramatically impacted through the utilization of AI techniques. The retail industry is stocking more effectively, utilizing dynamic pricing and getting closer to the goal of having their supply chain from manufacturer to customer fully automated. The energy industry is doing a better job of understanding demand. Underestimating demand can lead to blackouts and overestimating mean waste, so a critical need in this industry. The travel industry was one of the early adopters of AI through their need to better predict both hotel and airline demand. Even something as simple as the performance for local businesses can be impacted by better weather prediction leading to staffing or utilization issues.
