In the spirit of openness, the large and dedicated Artificial Intelligence (AI) community makes their insights public and accessible for everyone. The impact of this openness on AI productization will have the effect that AI is not going to be a product differentiation, but an affordance.
AI Research is Open
Currently, the Artificial Intelligence (AI) and Machine Learning (ML) communities are two of the most vibrant research communities. Theorists work along practitioners to push the boundaries of what is possible and together they advanced the state-of-the-art tremendously in the past decade.
Open science and research publication platforms are an important cornerstone to disseminate ML & AI research results. The open science model short-circuits the traditional, peer-reviewed publication process by making research material accessible as soon as it is ready. This speeds up ongoing research efforts tremendously, because researchers can immediately pick up results from other groups and respond to them in their research. In addition, research blogs and tutorials help new entrants to climb up the steep learning curve for AI. Webinars and online courses make it easy to benefit from the teachings of some of the brightest minds in the community.
Commercial offerings for AI platforms aim at tearing down the storage and compute barriers for AI. They provide interfaces and backend systems that facilitate implementing and integrating AI in a system greatly. As a result, everyone with a good understanding of AI can go ahead and leverage AI for their prototypes and products.
AI Success Stories
Useful products solve real-world challenges, thus, create value. Solving a challenge intrinsic to AI has its own merits, but success in research does not directly translate to the realm of products. Product success is driven by customers. It can be measured through its rate of adoption. Naturally, a successful product will soon have imitators who are competing for customer attention and market share. Product differentiation creates a competitive advantage for the product’s seller, as customers view these products as unique or superior.
Adding AI components in a product neither makes it better, nor different per default. Failed startups protocoled that they ‘overestimated machine learning’ in their product. In fact, ML & AI is usually just one piece of the puzzle. Let’s look at some examples of successful applications using ML and their enablers: Video streaming services recommend movies to their customers based on preferences and watching history. Their prerequisite is a system that delivers movies to their users on demand. Web search and social networks categorize images automatically. This requires the capability to store and retrieve relevant images when queried. Credit card companies analyze their client’s spending behavior to identify fraudulent transactions. This can only be done in an active processing system that enables red-flagging of certain transactions.
AI Improving Effectiveness
The common theme in these examples is that there is a system. ML complements this system, for example, by augmenting and controlling user interaction. All by themselves these systems are already creating value. Even if these companies would scrub the ML portion, their systems would still deliver great service. In these examples, ML & AI are an incremental improvement. The key takeaway point from this line of reasoning is that AI is not a shortcut for building a great system. More bluntly, AI does not turn a mediocre system in a top product.
Additionally, the competitive advantage of a product using AI is slim. Even though a market participant might be able to jump ahead because of integrating AI technology, the lead margin will decrease, because competitors will catch up quickly. Reasons are the openness of AI research and the broad availability of AI tools and frameworks. In the light of the rapid advancement, keeping competitors from catching up is hard. Novel applications of AI are quickly discovered by the AI community and replicated them in the research realm. For instance, this is one of OpenAI’s goals.
Ironically, the same thing that enables us to build advanced AI products keeps AI from being a product differentiator. To build a true AI-based product, AI has to go beyond incremental improvements. Think about riding a car without a steering wheel. Clearly impossible without AI algorithms taking control under the hood.
AI as Affordance
In product design, affordances define how the product could possibly be used. More formal, affordances determine the relationship between the properties of an object and the capabilities of the agent using it. For instance, a steering wheel affords steering, a handle affords grabbing, a knob turning, etc.
AI adds a new dimension of interaction: it enables placing handles and knobs that would not exist without AI. An illustrative example are self-driving cars: Removing the AI portion, the car would not be functional. It is impassible for human driver to use such a car. It lacks the basic affordances like steering wheels and gas throttles that are required for human operation. AI removes the affordance for these parts from the self-driving vehicle. These are the AI products that you’ll want to build, those are true differentiators.
By enabling new ways of using and interacting with technology, AI facilitates an entirely new user experience. For instance, the lifestyle enabled through self-driving cars is only the tip of the iceberg. Voice control and robots capable to interpret human behavior would be other instances that show how AI is going to penetrate and turn our lives upside down.
AI can improve the user experience to the level where perceived intelligence and true intelligence blend. This is where the Turing test comes into play: a human user cannot tell the difference based on his/her interactions. The predictive power of AI algorithms can anticipate - in a constrained environment and on a constrained timescale - what actions a user might take, or what results from certain actions. Advanced Driver Assistance systems (ADAS), for instance, are built to filter out bad driving decisions that might lead to crashes.
In the classical definition of affordance, affordances are passive. Affordances manifest the possible ways a product can be used. Using AI, a product can actively reject certain behaviors. For instance, ADAS systems and self-driving cars are built to avoid crashes. In other words, these cars afford not-crashing: it’s just nearly impossible to crash a car by using the defined ways of interaction. Active rejection thus reduces false and hidden affordances and guide the user to intended use. Active rejection is one example for a new dimension of interaction. More are to be discovered.
Takeaway
The true power of AI is in creating new affordances. Augmenting systems with AI can increase efficiency, but adding complete new dimensions of interaction will enable AI technology leadership.