AI assistants are being used by companies to do everything from organizing medical records and streamlining transportation and logistics to chatting with customers and providing 24/7 customer service. At home, digital assistants like Apple’s Siri and Amazon’s Alexa help us call our friends and change the way we buy things. These AI assistants are expected to become more prevalent but also more human-like through the advents of conversational AI and reinforcement learning.
Currently, our frustration with technology stems from the navigation of unintuitive menus, pop-up windows and automated phone systems that can’t understand us when we speak. The cause of many of these frustrations is the fact that we have to communicate with technology on its own terms. To show you what I mean, imagine you want to stream a song on your phone. You first need to learn the language of your smartphone (i.e., how to use it) and then the language of the app itself, rather than being able to get to the song with conversation and dialect that comes naturally to humans.
AI is becoming more human by transitioning to a new human-first approach to language. This approach is built around natural conversation—we wouldn’t need the intermediates of menus and keyboards if we could talk to our digital technology through a conversational AI assistant.
Early versions of this approach already exist through assistants like Apple’s Siri, Amazon’s Alexa and chatbots. These AI assistants are conversational interfaces that allow you to use technology in a more human way. Using the real-time data analysis necessary to hold a conversation, these new AI assistants are becoming more aware of context, the impact of past conversations, and environmental factors. This will continue to make your interaction with AI much more personal and individualized. Soon, AI will be able to converse like a human, but faster and with immediate access to more information.
Making AI more human through the ability to converse will allow us to use technology in a way that is more intuitive. However, there are still some challenges to overcome. For instance, current AI models are trained through “supervised learning.” This means humans must label and categorize massive amounts of data for the AI to learn from. However, this problem is being overcome by teaching AI a very human skill—the ability to learn from its own mistakes.
OpenAI, a San Francisco-based research company has developed an algorithm allowing AI to review its previous actions and learn from its past attempts to complete a task or goal. Conventionally, when teaching AI to complete a task, current algorithms will either give a virtual reward (if the AI is successful) or withhold the reward if it fails. This new approach mimics the way humans master new skills, wherein mistakes are fundamental to getting better so long as you can learn from them.
This type of learning allows AI to learn without having to explicitly program the assistants for a given task, on a specific, human-generated dataset. Just like humans, they will have a core set of skills that enable them to learn how to complete new tasks and improve based on their experiences.
The promise of AI is vast, and as AI assistants become more human, they will be able to integrate into our lives more easily and solve many human-computer interaction problems.