Machine learning as a competitive advantage

Marimar Jiménez, a journalist with Spanish financial daily Cinco Días, sent me some questions for an article (SP) she published about the competitive advantages machine learning can give companies.

Machine learning is, above all, a set of tools that allow a machine to iteratively learn from data and develop models that have not been specifically programmed by a person. Machine learning’s capacity for disruption can provide a competitive advantage: the algorithms are adapted to the data and end up generating better predictions and results than those developed by people, so companies that use machine learning obtain greater efficiency, better performance, more agility and other previously impossible functions.

But as a tool, machine learning is not something that can simply be “bought and installed”, because it depends on the quality and accessibility of data, and therefore requires a “data-centricity” that for many companies is still not possible. In reality, a large part of what is called machine learning remains inflated expectations, unfulfilled promises and the unrealistic hopes of businesses that believe it will turn them into the company of the future. Developing the procedures that allow the collection and preparation of data is enormously complex. Machine learning now faces an epidemic of misinformation. Only those companies able to orient themselves to the generation and processing of data will benefit from machine learning and turn them into real competitive advantages.

Below my answers to Marimar Jiménez.

Q. Do you agree that data has become the main asset of companies? Why? Why do General Electric or Siemens now define themselves as data companies?

A. Data gives a competitive advantage in the machine learning environment. Only companies able to obtain and maintain algorithms that are smarter and more powerful than their competitors’ will be here for the next decade, and the key to getting those algorithms is to have data to feed them. It’s something we’ve been saying for a long time: if you do not direct your business to data, maximizing information intensity and working according to your level of permission, a competitor will appear that is able to do so and that provides its users with better products and services than you. Data becomes the fuel that drives your algorithm motor, but data alone is not enough: you have to know how to define objectives, prepare them, transform them, build models, evaluations, predictions… guiding the company to data is just a first step, and those that follow are not as simple or trivial as many suggest.

Q. There is a lot of talk about big data, data analytics, machine learning, all as pieces of the same puzzle. Is the next big thing automatic learning? How will it impact (or is it already impacting) on businesses? What can we expect from algorithms applied to business?

A. Machine learning has been the next big thing for a while now: this can be seen from the evolution of related tags on my website: I have been talking about this for some time, companies are engaged in it are undergoing major acquisitions, and network giants like Google, Amazon, Facebook, Apple or Microsoft are reorienting all their strategies around the issue. We have gone from seeing an algorithm as something with more computing power, more mathematical brute force than a person (Deep Blue beating Garry Kasparov) to seeing it as something capable of understanding human language better than many people (Watson winning at Jeopardy) and able to do things that no human has done thanks to deep learning (AlphaGo winning the world Go championships) or even to make better decisions than a human in situations of imperfect information (Libratus winning at poker).

The point is not that a machine can now do what a person does, but that it can do it much better. Companies that do not know how to take advantage of something like this will disappear.

Q. UC Berkeley auto-learning expert Michael I. Jordan, says more and more data increases the likelihood of making false connections. Is this a handicap for the advancement of data economy and how can it be avoided?

A. Under the right circumstances, everything correlates with everything. But this is the field in which machine learning stands out: many algorithms can be evaluated based on the results obtained and improvement processes applied to improve those results. Algorithms analyze data and extract rules to generate predictions, detect exceptions, isolate patterns … as we feed algorithms with more data, they improve and can even come up with new hypothetical situations that have not happened previously, playing against themselves to improve the results obtained. Data is obtained in scenarios of all kinds and are applied to the whole system: every time an autonomous vehicle drives by a certain place it contributes data that serves the whole fleet; algorithms are also able to learn from playing games like Grand Theft Auto to generate new situations that would not happen under normal conditions. The important thing is to understand the process: these are not rules or menus from programming to use: machine learning allows possibilities that surpass what we thought was a computer.

Q. What are the challenges facing corporations in this new economy, from a technological, cultural or other point of view?

A. At this moment, the challenges are in guiding companies to generate data that can be analyzed. If the only thing you generate when you sell a product is that, a sale, and you do not have more data on who bought it, their characteristics, or an evaluation of the product or its use, then the competition will put you out of business. But in addition to obtaining data (we all think we have data, but we don’t), we must develop the capacities for its exploitation. In Amazon, human intuition is forbidden when making decisions: if you make a decision, show the data that justifies it.

Q. Are there companies born in the heat of big data and machine learning that would not exist were it not for these technologies? Give me an example. What about a traditional company that is using data well?

A. We are seeing acquisitions and movements: all the big players have carried out large acquisitions of machine learning companies, acquisitions that are a cross between acqui-hire (the acquisition to incorporate talent) and the direct application of capabilities to their processes. All large companies are positioning themselves to incorporate these capabilities, these specialists, offering them environments in which they can develop. We are living through the beginnings of the biggest change that technology has caused, and the impact will be bigger than that of the internet itself.

(En espa?ol, aquí)