A Futuristic Reality: Harnessing The Power Of The Three Layers Of Machine Learning

• Read original article here

Artificial intelligence systems powered by machine learning have been creating headlines with applications as varied as making restaurant reservations by phone, sorting cucumbers, and distinguishing chihuahuas from muffins.

Media buzz aside, many fast-growing startups are taking advantage of machine learning (ML) techniques like neural networks and support vector machines to learn from data, make predictions, improve products, and enhance business decisions. Unfortunately “machine learning theater” – companies pretending to use the technology to make theirs seem more sophisticated for a higher valuation – is also on the rise.

Undeniably, ML is transforming businesses and industries, with some more likely to benefit than others. Like any source of economic power, ML has the potential to fundamentally change what are considered to be core or non-core functions for an enterprise, challenging some prevailing “theory of the firm” assumptions.

While harnessing ML has the potential to yield extraordinary economic power for companies, many machine learning businesses aren’t able to build sustainable “moats.” This article lays out a simplified three-layer framework to explore the kinds of businesses that are using ML to grow at scale:

This layer is the technology upon which ML systems are built – algorithms, frameworks, runtime engines, toolkits and workbenches. Much of the best technology at this stage is open source or delivered as-a-service by large public cloud vendors with track records of continuously lowering prices.  At this layer, users care about the pace of improvement, not just current features, and many inexpensive open source and public cloud ML workbench offerings are improving rapidly and can address various types of ML problems.

Economic moats are hard to build at this layer; sustainably differentiated business can require enormous investments in research & development, marketing and sales. Affordable support is available for popular ML open source frameworks, and public cloud vendors take advantage of their massive scale to regularly lower prices and invest in features and security.

The best businesses at Layer 1 benefit from talent network effects and application ecosystems. As a ML workbench product reaches scale, talent and familiarity in the market naturally builds. Some of the most sustainable businesses recruit developers to build turnkey applications on top of their platform, creating a second network effect and increasing the platform’s appeal to future users and developers.

This layer includes cross-industry turnkey ML capabilities. These are prediction or classification systems that solve problems not “core” to a customer’s business or industry, such as cybersecurity and fraud detection.

At this layer, customers may be generous with their data, as long as it’s treated carefully, since this data is often not seen to be a competitive advantage. For example, security data belonging to a bank or hospital, while sensitive, is not generally considered a source of strategic differentiation. As long as customers are confident their data won’t leak out, many are willing to share their data with a scaling software startup, allowing customers to reap the benefits of the continuous improvements in the startup’s ML system.

In some cases, ML at this layer can generate “data scale” economic power, a result of increased access to data for training models and better feedback on the accuracy of machine predictions. As long as there are not rapidly diminishing returns to marginal data, the market leader can enjoy powerful data scale effects, similar to the advantages Google has with search or Amazon has with product recommendations.

Some human tasks, like judgement, are economic complements to ML systems and will increase in value as ML systems that lower the cost of prediction are widely adopted. Human partnership with complementary ML can allow entirely new kinds of business models to emerge, unlocking additional pools of labor.

This layer is made up of a large number of industry-specific applications that can be improved with prediction or classification systems. Netflix and Amazon exemplify how something as simple as a recommendation engine can reshape entire industries and value chains, and ML-driven applications can be very powerful, even when niche. At this layer, customers may lean toward “build” in a build vs. buy analysis, hesitant to share what they view as key functions or strategic data with a vendor.

As a result, some ML startups with Layer 3 offerings are rethinking traditional boundaries of the firm and going direct to end consumers instead of selling software to traditional industry players. If a startup owns the end-user relationship, it secures a fresh source of training and feedback data to improve its products. Legacy incumbents in the industry may get cut out entirely.

At this layer, marginal increases in prediction accuracy can yield significant increases in utility. Going from 98% accuracy to 99.99% accuracy reduces the error rate by 99.5%. This can dramatically improve the usefulness of an ML system in a domain where error tolerance is very low, such as with self-driving cars. We may soon pass accuracy thresholds beyond which humans will be taken “out of the loop” entirely for certain tasks, which can fundamentally transform entire industries.

In the future, one thing will remain true: building a sustainable business at any layer requires both long-term investors and the ability to attract and retain high-caliber talent. And right now, many incumbent technology vendors have neither. This opens the door for change and may shift economic power from incumbents to growth stage startups powered by ML.

Artificial intelligence systems powered by machine learning have been creating headlines with applications as varied as making restaurant reservations by phone, sorting cucumbers, and distinguishing chihuahuas from muffins.

Media buzz aside, many fast-growing startups are taking advantage of machine learning (ML) techniques like neural networks and support vector machines to learn from data, make predictions, improve products, and enhance business decisions. Unfortunately “machine learning theater” – companies pretending to use the technology to make theirs seem more sophisticated for a higher valuation – is also on the rise.

Undeniably, ML is transforming businesses and industries, with some more likely to benefit than others. Like any source of economic power, ML has the potential to fundamentally change what are considered to be core or non-core functions for an enterprise, challenging some prevailing “theory of the firm” assumptions.

While harnessing ML has the potential to yield extraordinary economic power for companies, many machine learning businesses aren’t able to build sustainable “moats.” This article lays out a simplified three-layer framework to explore the kinds of businesses that are using ML to grow at scale:

This layer is the technology upon which ML systems are built – algorithms, frameworks, runtime engines, toolkits and workbenches. Much of the best technology at this stage is open source or delivered as-a-service by large public cloud vendors with track records of continuously lowering prices.  At this layer, users care about the pace of improvement, not just current features, and many inexpensive open source and public cloud ML workbench offerings are improving rapidly and can address various types of ML problems.

Economic moats are hard to build at this layer; sustainably differentiated business can require enormous investments in research & development, marketing and sales. Affordable support is available for popular ML open source frameworks, and public cloud vendors take advantage of their massive scale to regularly lower prices and invest in features and security.

The best businesses at Layer 1 benefit from talent network effects and application ecosystems. As a ML workbench product reaches scale, talent and familiarity in the market naturally builds. Some of the most sustainable businesses recruit developers to build turnkey applications on top of their platform, creating a second network effect and increasing the platform’s appeal to future users and developers.

This layer includes cross-industry turnkey ML capabilities. These are prediction or classification systems that solve problems not “core” to a customer’s business or industry, such as cybersecurity and fraud detection.

At this layer, customers may be generous with their data, as long as it’s treated carefully, since this data is often not seen to be a competitive advantage. For example, security data belonging to a bank or hospital, while sensitive, is not generally considered a source of strategic differentiation. As long as customers are confident their data won’t leak out, many are willing to share their data with a scaling software startup, allowing customers to reap the benefits of the continuous improvements in the startup’s ML system.

In some cases, ML at this layer can generate “data scale” economic power, a result of increased access to data for training models and better feedback on the accuracy of machine predictions. As long as there are not rapidly diminishing returns to marginal data, the market leader can enjoy powerful data scale effects, similar to the advantages Google has with search or Amazon has with product recommendations.

Some human tasks, like judgement, are economic complements to ML systems and will increase in value as ML systems that lower the cost of prediction are widely adopted. Human partnership with complementary ML can allow entirely new kinds of business models to emerge, unlocking additional pools of labor.

This layer is made up of a large number of industry-specific applications that can be improved with prediction or classification systems. Netflix and Amazon exemplify how something as simple as a recommendation engine can reshape entire industries and value chains, and ML-driven applications can be very powerful, even when niche. At this layer, customers may lean toward “build” in a build vs. buy analysis, hesitant to share what they view as key functions or strategic data with a vendor.

As a result, some ML startups with Layer 3 offerings are rethinking traditional boundaries of the firm and going direct to end consumers instead of selling software to traditional industry players. If a startup owns the end-user relationship, it secures a fresh source of training and feedback data to improve its products. Legacy incumbents in the industry may get cut out entirely.

At this layer, marginal increases in prediction accuracy can yield significant increases in utility. Going from 98% accuracy to 99.99% accuracy reduces the error rate by 99.5%. This can dramatically improve the usefulness of an ML system in a domain where error tolerance is very low, such as with self-driving cars. We may soon pass accuracy thresholds beyond which humans will be taken “out of the loop” entirely for certain tasks, which can fundamentally transform entire industries.

In the future, one thing will remain true: building a sustainable business at any layer requires both long-term investors and the ability to attract and retain high-caliber talent. And right now, many incumbent technology vendors have neither. This opens the door for change and may shift economic power from incumbents to growth stage startups powered by ML.