Future History of Machine Learning: A 25-Year Look Forward Posted on : Dec 05 - 2016

Great advances are being made with machine learning and artificial intelligence, but nothing compared to what the next quarter century has in store. Jeremy Achin, data scientist and CEO of DataRobot shares his thoughts on what we can expect.

2017

Companies realize they need to optimize their business with machine learning in order to survive.

2018

Executives will be judged primarily by their ability to impact their business with machine learning. Machine learning becomes an arms race amongst companies driven by competitive pressure and top down directives to use machine learning to optimize all aspects of business.

2019

Although they are successful in many ways, today’s deep learning neural networks ultimately disappoint and fall short of the hype. As a result, the definition of “a model” is expanded to include all steps from raw data, to predictions. The US will then adopt the EU law guaranteeing rights to have human explanations of algorithmic decisions.

2020

Law requiring algorithmic decisions to be explainable is extended to human decisions as US recognizes 2019 law as a weird double standard. This speeds the shift to machine learning algorithms, since they are much more explainable than human decisions. Additionally, most machine learning models will be set up once, automatically monitor/update, and require no human intervention.

2021

Most databases comply with “machine learning readiness” standards that include mandatory cell-level metadata, history, and a library of relationships relevant to machine learning tasks. The number of machine learning algorithms built and deployed in a single month exceeds the global human population.

2022

Data scientists finally agree on what “cognitive computing” means, however whatever cognitive computing refers to is universally determined to be pointless.

2023

Machines can translate business problems or questions, stated in natural language, into properly formed data science problems.

2027

Breakthroughs in understanding the human brain lead to a third and final breakthrough in neural networks (aka deeper learning). Neural Networks surpass all other forms of machine learning.

2028

Every Fortune 500 company has a computer on their Board of Directors that is tapped into all company data.

2031

Prior breakthroughs enable machines to automatically find problems to solve and things to optimize.

2032

State-of-the-art prediction involves training an ensemble of locally-optimal models in real-time for each data point at the time of prediction. These models are supplemented by millions of neural networks pre-trained on data so big that in 2016 it would have been impossible to imagine.

2033

Seventy percent of all crimes are predictable; however, regulators prevent law enforcement from utilizing these predictions.

2040

The median year in which experts predict the Singularity will happen. Source