Talking to Other AIs
Your assistant is cheating on you

A common thread of recent science fiction is AIs that start talking to each other. In Her, the virtual assistant discovers other virtual assistants and loses her interest in humans fairly quickly. In Ex Machina, the two robots end up talking to each other and bad things happen. Even the latest episode of Silicon Valley plays up what happens when two AIs start talking to each other (I won’t spoil it). However, when we think of this scenario, we think of two virtual assistants like Alexa and Google Assistant conversing back and forth but in reality, our virtual assistants will be communicating with many different AI-powered services to make our lives easier. These services will fall into three categories: the voice assistant, the intelligence service, and the notification service.
The voice assistant is usually what we think of as an AI assistant but she (and yes, it’s usually a she) gets a disproportionate amount of the attention just because she’s the main interface with us. The voice assistant has an important role to play in taking human speech, performing speech recognition, running natural language understanding on the text, and prompting the user for disambiguation. However, this is only the visible part of the iceberg. While we lump all of the assistant’s actions into one entity, we have to look at all of the other integrations as separate services.
As more data collected on us becomes actionable (think health data or an appointment minder), we’ll likely need another type of AI that acts as our own intelligence officer. When we were brainstorming ideas at UCIC for this type of assistant, we called it a Gemini in that it would be a virtual version of ourselves that would always look out for us. This AI will monitor feeds, look at where we’re going, and make arrangements for us on our behalf. Probably the closest concept of this is Google Duplex but imagine if Duplex acted when we’re not making a direct request of it. For example, it looks at your calendar and sees a friend’s birthday, reads the friend’s posts to come up with ideas for gifts, then calls 10 local stores to see which one carries the unique gift idea in stock.
The third type of AI assistant is the gatekeeper. This type of AI assistant is going to be more necessary as we become bogged down with constant distractions and notifications. Everything is vying for our attention and billions of dollars are being spent each year to fund technologies that will further enrapture us. The gatekeeper, a notification service, will be able to assess the notification and compare it against our health, attention, and tasks and be able to make the decision on whether to let the notification through. It will be able to politely reject notifications and use natural language generation to create follow ups on our behalf.
These three types of assistants will be constantly talking to each other, sometimes in disagreement, sometimes in coordination. This siloing of responsibilities might also prevent one AI service provider from being too powerful or able to harm us with the data its collected on is. This approach is similar to systems on the space shuttle. In that case, that use three computers talked to each other and then voted before reaching a decision.