In 2015 IBM announced that “100 of its Watson partners have now introduced cognitive enabled apps, products and services into the market.” At the time this was billed by IBM as a major milestone and validation that IBM was building an AI Platform that would rival the Mainframe, PC, Cloud and Mobile. I know this because I was there. Of course each of those previous ‘eras’ did produce game-changing Platforms led by the IBM 360, Windows, AWS and IOS. Watson never achieved Platform status, but instead became a brand that encompasses technology, products and services.
Last month OpenAI announced that “over 300 applications are delivering GPT-3-powered search, conversation, text completion, and other advanced AI features through our API.” I couldn’t help but feel deja vu.
To be clear both Watson and GPT-3 are incredible engineering achievements. Watson opened people’s eyes to the power of natural language understanding and inspired years of innovation in the field. IBM has built a real business around AI and I expect them to continue to be a major player moving forward. OpenAI generally and GPT-3 specifically have accelerated the advancement and adoption of AI and increased accessibility for the average developer.
Given the hype surrounding each new breakthrough in AI, it’s important to distinguish Big-P Platforms from small-p platforms. In software development jargon, a small-p platform is any tool on which applications are developed. Small-p platforms can include algorithms, libraries, languages, frameworks and on up the stack. The point is ‘platform’ is used to describe almost everything in tech with a business model.
The defining characteristic of a Big-P Platform is that an ecosystem develops around the Platform such that the value provided by the ecosystem is more important to the decision to develop on the Platform than the Platform technology itself.
Breakthrough algorithms like GPT-3 and Watson are built upon years of algorithm development in the broader AI community. As such they are inevitably replicated and (at least to-date) surpassed. Large ecosystems simply do not develop around proprietary algorithms no matter how powerful. In fact it’s another giant ecosystem (the one around open-source and fundamental research) that provides the bulwark. The result is that if you’re building a real application (as opposed to a demo or proof of concept), you’ll need to rip and replace your chosen algorithm with something better many times over. In other words, there won’t be an ecosystem that creates a reason to stay. It’s more important that you build the tools and processes that make rip and replace easy than that you choose the right algorithm.
I believe AI does represent a new era in computing. I do believe an AI Big-P Platform (or more likely a few) will emerge as the technology and development tools mature. Algorithms from Watson or OpenAI are incredible achievements and may be the best tools to get the job done. Check them out! But be wary of algorithms that claim to be Platforms.