Digital Exprt Blog

Narrow AI and Screenless Interfaces

Artificial Intelligence (AI) has gained a renewed lease on life. This renewal happened thanks to recent popularization of messaging platforms. Seems like many (most?) people today prefer to use technology mostly for exchanging short text messages. Automating the processing of text messages requires intelligence, hence the renewed interest in AI.

Initial Excitement

The global market is shifting thanks to the popularity of mobile computing. Billions of people are spending majority of their time texting, and less and less time using apps. It is to be expected that business offering must follow suit — go where the markets are!

There are many reasons why offering text-based messaging services makes sense. From the end-users’ perspective, it minimizes the need to deal with technology. If we can talk to business services the same way we talk to friends and coworkers, technology disappears from our sight. Instead of dealing with technical issues, we’re conducting business in an intuitive fashion. We then realize, much to our delight, that there is nothing to learn. That comes as a relief because we can get things done and not worry about technicalities.

From the business perspective, there is no need to build a responsive Graphical User Inerface (GUI). Also, there is no need to make sure the interface works across various clients. So we see that conversational user interface brings the win-win-win situation. Triple win sounds like an amazing accomplishment. And so the chatbot revolution began. The gold rush ensued. Chatbot builders focused on replacing traditional apps with conversational interfaces. ‘Bots will replace apps’ became the mantra of the year 2016!

Rapid Disillusionment

Things are not that rosy, unfortunately. As it turns out, it takes much more effort to automate a conversations than you’d expect. This issue boils down to still not having so-called General AI available to us. If we want to handle conversation in an automated fashion, we must use General, or Strong AI.

Today, the state of technology only gives us weak, or narrow AI.

Weak AI can automate a very narrow problem area. For example, it can automate a game of chess. Or, it can automate a better way to store data on a hard drive, and so on. Such narrow specialization is useless when it comes to conducting a conversation.

Once we’ve realized the severe limitations narrow AI imposes, it was time to pivot. Chatbot builders started giving up on trying to offer free-form conversational interface. Instead, they scoped the user experience down to few restricted options. In the process, they started introducing GUI into the messaging channels.

Conversing By Clicking/Taping?

This new initiative resulted in hilarious user experiences. Instead of accepting users’ text input, the messaging app offers only buttons. Replacing textual user interface with GUI is a surefire way to kill the conversation.

Another problem is that messaging channels cannot offer rich GUI experiences. Messaging channels are not meant for displaying buttons. Messaging channels enable conversations, and conversations are textual by their very nature.

Screenless Interfaces

Yet another problem arises from the growing popularity of so-called screenless interfaces. Voice-activated technology is rapidly maturing. Devices such as Amazon Echo help popularize voice-driven interaction.

Of course, voice-driven interaction does not depend on a screen. Users issue messages by speaking. Sophisticated technology responds to voice messages and gets things done for the user.

In such scenarios, graphical user interface becomes obsolete. For example, instead of manipulating graphical widgets, I can say “Siri, wake me up at 7:00 in the morning”. I don’t need to look at nor interact with a screen to set my alarm clock. Very convenient.

Open-ended Conversation

With the removal of screens, we lose the ability to restrict the options available to the user. Voice-activated interaction implies losing the ability to channel the conversation. Users can say anything, and the onus is on our software to process it. In the absence of General AI, we realize we are incapable of offering predictable service.

So, once we lose the GUI harness, we must discourage open-ended conversations. Open-ended conversation under such circumstances is a blind alley. Not only is it inefficient, it can also be harmful. Allowing users to submit any free-form input can result in unpredictable behaviour.

Is there a way out of that predicament?

Stick To Narrow AI

The only way to ensure predictable behaviour of our chatbot is to endow it with narrow AI. As we’ve seen, a bot is incapable of fielding open-ended input. It must let users know that only restricted input makes sense.

OK, but how is our bot going to enforce restricted input using text-driven interface? Asking users politely to respect the protocol is not going to work. Any time we try something like that, we notice how users don’t care about our instructions.

So what to do then? The only way to get users to stick to the narrow protocol is by offering significant value in exchange. Most existing bots attempt to empower users by letting them request trivial services. For example, letting users ask about the weather forecast, or ask what time is it, etc. Users don’t perceive such services as being of high value. If we then ask users to formulate their requests more carefully, they won’t find much incentive to do so.

But if we offer a service that is of high value, users will feel more incentivized to pay attention to how are they requesting such services. Such narrow AI service then becomes a valuable commodity. People tend to gladly return and reuse services that they deem being valuable. And with repeat use, people learn how to request services in a way that guarantees successful delivery.

Conclusion

The only way to successfully implement screenless interfaces is to offer narrow AI automation. A narrow AI can only process very limited vocabulary. The onus is on users to strive to learn that vocabulary. Unless the service on offer is of a very high value, people will not bother mastering the required communication protocol.

What’s an example of a valuable narrow AI service? Imagine an automated travel agent. This agent is capable of finding suitable arrangements without expecting a lot of step-by-step guidance from its users. And because the text-based interaction with the agent will be sporadic and simple, people will find strong incentive to master it.