Today, Google’s AI voice assistant is the smartest, Amazon’s Alexa is the most impactful, and Apple’s Siri is the most overlooked (which could be due to Apple’s focus on privacy).
Artificially intelligent (AI) assistants are going to be commonplace in next-generation products because of their functionality. Soon, users will no longer have to spend time learning the system because interaction with AI assistants will resemble a normal human conversation.
Sign up for your weekly dose of what's up in emerging technology.
Here are some of the main factors that will (and should) affect the direction of the growth of smart assistants:
Context of Use
Where and how a product will be used is the first thing that needs to be determined before creating a new AI experience. The design of a product should differ on the basis of its purpose: the session time, interaction medium, and attention span should all be customised for individual products.
Google and Amazon’s virtual assistants, for instance, have different goals: Google wants to deliver ads to consumers, while Amazon wants to sell stuff.
As of now, Amazon’s Alexa is the best in the market for smart homes and home automation. Many of Alexa’s 100,000+ skills are to control the many features of homes: lights, garage doors, security systems, coffee makers, home appliances, and more. It can further order groceries, check your bank balance, pay your bills, and connect to your insurance company.
Meanwhile, Google Assistant can book restaurant reservations, track your appointments, and order your ride-sharing services. It can also, like Alexa, control the smart devices in your home. Compared to Alexa’s 100,000 skills, Google Assistant has 1 million capabilities—making it the smartest voice entertainment in the market.
Skills and Scenarios of Interaction
Skills reference the abilities an AI assistant is in possession of, whereas ‘scenarios of interaction’ refer to how people will use those skills to improve their lives. To continually develop smart assistants, it is important that companies determine what skills and scenarios of interaction they aim to deliver to their consumers.
For example, Amazon recently announced that its voice apps have new conversational abilities to greatly improve their functionality. Its most recent innovation is that you can now talk to Alexa without having to specify what skill (or voice app) you want to use. Alexa will now have a short-term memory so that you don’t have to reference previous conversations you have had with it. For example, if you recently ordered a pizza from Pizza Hut, you can just say, “when will the pizza arrive”— and Alexa will remember the context of your conversation.
The things that companies should consider when defining the core skills of their product include:
- Utility: All AI assistants should serve a clear functional purpose—whether they are universal assistants (capable of doing a variety of things) or niche assistants (such as a dating app chatbot). It’s important to identify a target audience and to know what their expectations are—and to be able to spot specific user tasks where improvements are necessary.
- Discoverability: A good AI assistant will be easy to figure out. Mechanisms should be present that make it easy for users to discover the functionality of the product.
- User Feedback: A feedback mechanism in which users can request new skills for a smart assistant is a great way to remain innovative and continue serving the needs of consumers.
Amazon and Google currently form the duopoly of the voice assistant market: and the more private information they gather from their consumers, the stronger their dominance gets. Therefore, it’s easy for them to favour their own products over those of their rivals.
It’s possible that, in the future, open-source AI developers will release more smart assistants that work only for individual users—instead of a larger company with its own self-serving goals. This will be difficult to happen since open-source resources typically do not have the resources that big companies have at their disposal. To train a virtual assistant is often unreasonably expensive: requiring vast data sets of natural language samples painfully annotated by human experts.
Nevertheless, Mycroft is an example of an option like that available today: it is “private by default and completely customisable.” Similarly, a team of researchers at Stanford has also developed an open-sourced smart assistant called Almond that could widen the competition and give more privacy to consumers.
The closer AI assistants get to conversing with users and making them feel like it both knows them and respects their privacy, the more valuable it will be as a product.