Will Artificial Intelligence Ever Make Money?

It may be that AI, by the nature of its structural reality, is easy to commoditize.

Whether providers will be able to charge premiums for elemental units or even build it into their consulting cost structure could prove challenging as the following article explains. JL

Thomas Davenport comments in Harvard Business Review:

Software “micro-services” perform small chunks of functionality on data and then return a result. Because these are small, it’s more difficult to get organizations to pay for them. Because they’re small and modular, they lend themselves to creation by multiple software developers, who often contribute them to open source libraries.This is what has happened in the cognitive domain. It’s going to be difficult to make a good living just by selling cognitive software.

I was recently consulting with a publishing company that is exploring various ways to digitize and contextualize its content. Knowing that some of the company’s competitors had signed deals with IBM’s Watson, I asked several executives why they had not done a Watson deal themselves. “We think that the market for AI software is rapidly commoditizing, and we believe we can assemble the needed capabilities ourselves at much lower cost,” was this company’s party line. Some particularly knowledgeable managers mentioned that they expected the company would instead make use of open source cognitive software made available from various providers. These potential open source providers are not small vendors; they include, for example, Google, Facebook, Microsoft, Amazon, and Yahoo.
Upon hearing this company’s strategy, I was initially a bit surprised. Could machines that can think already be so cheap and available? How could the cognitive software market be commoditized when the marketplace is relatively new? Why would developers of exotic “deep learning” and machine learning software give it away for free? How can IBM expect to make $10B in Watson revenues if it’s not clearly better than the free alternatives?
First, some suggestions as to why commoditization is happening within AI (which, for now at least, is perhaps more aptly termed ‘cognitive technologies’). There is a powerful tendency within all software today to move toward “micro-services” that perform small chunks of functionality on data and then return a result. These typically work as “APIs” or application program interfaces. Because these are small chunks of functionality, it’s more difficult to get people or organizations to pay for them than for larger units of software. Because they’re small and modular, they lend themselves to creation by multiple software developers, who often contribute them to open source libraries.
This is exactly what has happened in the cognitive software domain over the past decade or so. There are now many open source libraries with algorithms for common cognitive functions like neural networks, deep learning (neural networks on steroids), speech parsing and recognition, image recognition, and so forth. Some libraries have been open for many years, while those from Google, Microsoft, Facebook, and Amazon only became freely available in the last year or two. They are typically accessed through a vendor’s cloud (in which case the vendor can make at least make some money) or on programming sites like Github. If a lot of companies and programmers use a particular vendor’s open source cognitive tools, there’s a good chance that: a) the software will become a standard; and b) it will be easy to plug into other products from the same vendor.Even IBM Watson is going in this direction, at least to some degree. The software isn’t free (something has to pay for all those expensive ads), but it is now a set of APIs that perform various cognitive functions, including image analysis, sentiment analysis, and the original (Jeopardy!-style) Q&A. As I count them in the catalog, there are roughly 20 APIs now available through the Watson Developer Cloud, although the number is always changing as new ones are added, experimental ones are dropped, and related APIs are combined. Given the rapid pace of commoditization for cognitive tools, I wouldn’t be at all surprised if at least some of Watson’s APIs are open source before long.
Another factor that is driving commoditization is the move to “bots”—what might be called APIs for intelligent human interface. Bots—sometimes called “chatbots”—are small applications that allow conversational interaction with programs, either through text or voice input. In order to succeed they have to convert speech to text, parse the text, and understand a substantial vocabulary. This sounds hard, but many of the same companies that have made their AI software open source are also making available bots to interface with their own programs and just about everything else. Soon they’ll be ubiquitous; there are already even some open source bot libraries. And since bots are just interfaces—an input like typing or clicking, but much easier—no one is likely to pay a lot for them by themselves.
What all this means is that it’s going to be difficult to make a good living just by selling cognitive software. There will, of course, be a need for lots of external services by companies who don’t have a phalanx of data scientists in their employ. Some consulting will also be needed by many firms to figure out where to use these tools in their businesses. I suspect there will also be some highly customized AI “solutions” that are too detailed and specific to be available through open source—e.g., an image analysis system that can detect a fraudulent check.
But in general, this type of software will mostly be abundant and free. If your company knows what it does, how to use it, and how to integrate it into your business, you’re golden. If you’re planning to sell it, not so much.