Welcome to the era of low-cost predictions. Your products and services will never be the same, and a couple of big companies may wish we’d never arrived.
On Thursday, Amazon Web Services announced that it was selling to the public the same kind of software it uses to figure out what products Amazon puts in front of a shopper, when to stage a sale or who to target with an email offer.
The techniques, called machine learning, are applicable for technology development, finance, bioscience or pretty much anything else that is getting counted and stored online these days. In other words, almost everything.
AWS’s introduction of the product, called Amazon Machine Learning, follows Microsoft’s announcementlast summer of an online machine learning service to complement its Azure public cloud computing business. Last month, Google took a dramatic step in how it holds corporate data by lowering the cost, with an eye to becoming the analysis and prediction hub for the world’s companies.
The services offered by Amazon Web Services, Microsoft and Google draw off each company’s relative strength. Microsoft has Excel, by far the most widely used analysis tool in the world. Most people actively developing businesses on the web are familiar with Google Analytics. AWS appears to offer much of its own internal decision-making process.
With Amazon Web Services’ new features, “it will bring an entirely new decision-making technology into an organization,” said Matt Wood, the head of data science at AWS. “It can provide a company with information on the challenges facing them, or an overlay about what to do next.”
Mr. Wood devoted four years of studying for his Ph.D., which he got 15 years ago, to solving a problem in protein folding, an essential part of genetics research. Using the new AWS product, he said, that would take about a week.
At this point, there is no need to pick a winner in the cloud race among Amazon Web Services, Microsoft and Google. The real point is that if all of these cheap cloud companies are offering low-budget data science, it is almost certain to spread widely. While machine learning is expensive and arcane, if it becomes cheap and easy, it will get stuck onto all sorts of things, not just better online selling, or the way that Netflix uses it to recommend your next movie.
This would be bad news for companies like IBM, which has spent billions on analytics companies and mathematicians, or the SAS Institute, a maker of big predictive software packages. Much of what these companies do will almost certainly become cheaper, based on what Amazon Web Services is charging.
In testing its software, AWS put two engineers on the problem of telling the gender of a person’s name. Using conventional means, the company said, the team gained 92 percent accuracy in 45 days. Using the new Amazon Machine Learning product, one engineer reached the same accuracy in 20 minutes.
“Cities will be able to plan better for the next week, or year,” Mr. Wood said. “Initially we’ll see a lot of uses from customers who already have a lot of data in our system, with things like computer behavior data or customer behavior. Eventually you’ll see it in genomics, the Internet of Things, or large-scale web applications.”