In this three-part series we discuss three frameworks for viewing the AI market presented by Dennis Mortensen from X.ai at the 2016 AI WithTheBest conference. (Links to Part 1, Part 3)
This time we will discuss the second framework for looking at the AI market: Horizontal vs Vertical AI.
The natural question was then how one would differentiate between a function a Horizontal AI could take up and one that should be delegated to third-party vertical AIs. For example, iOS and Android both come with calendars, music players and calculators but they don’t come with sketching apps because those aren’t common enough. I asked Dennis how he thought Amazon and Apple might draw this distinction.
(My thoughts: There are two elements to consider: Demand and resources needed to produce that feature. If there is sufficient demand, horizontal AI will likely incorporate this feature into their base set. If it is easy to produce, they will do it in-house. If it’s not, they may choose to incorporate features via acquisition or partnership.)
Dennis boldly hoped X.ai would win the vertical before Apple thought it was core. He thought scheduling meetings could be a winner-takes all vertical with network effects.
Three stages in the near term for many settings:
- Existing setting -> All humans. Situations similar from one driver to the next. -> If not have much overlap, not a good vertical to attack.
- If you inject a machine agent into that environment: how to do machine to human interaction? Can’t do 100%, but can do high accuracy prediction. E.g. acr see if accelerate, can take in signals. Can maek apredction, still a pred. Maeks the whole environment enclosed. So move forward on a prediction that you think it’s good. Deosn’t mean it’s 100% ture. (AH.)
- But if car was just a machine car, you can ask it if it’s moving forward and it can say No. -> Andrew talks to Amy. Internal preference negotiation where you maximise happiness for both participants. Negotiate with myself -> product quality goes up.
Q: He thinks many markets won’t do well in a multi-agent setting. Need to adopt almost human-level negotiation tactics e.g. delays.
Q: Agents need different personality depending on context. You need to choose your vertical.
- Hoping for research that figures out what types of personality works for each situation.
- Users want different agents to choose from for different contexts -> Think solution is adapting one agent for say formal vs informal situations. -> Sentiment analysis.
Also think: What counts as a vertical? How specialised are you? it’s an age-old problem really. You might think ‘music’ is a vertical but then you have different genres of music. Within classical music you have violin-playing,
-> It’s still unclear which UX will end up being the right one, and some of us now believe that an assistant isn’t a product in itself, but rather something you add to existing products.
-> One thing in particular we are currently exploring at Snips is how we can add an assistant to existing chat apps instead of building our own chat app (try it here on Android). This would enable us to focus on the assistant, and not worry about the chat part or getting hundreds of millions of users to switch app.
The issue though is that most chat apps don’t have an API, and thus, we can’t access the conversations. Fortunately, we found a way around that on Android by leveraging two specific platform features: the ability to read the content on the screen, and the ability to draw things on top of an existing view.
To test our ideas, we chose to focus on a very specific use case: extracting places mentioned in conversations, showing information about them, and deeplinking into transit apps to go there.
The three-stage process and vertical-horizontal splits both seems intuitive, but I had not thought of either before.
And just for fun: (Cortana is Microsoft/Windows’s smart assistant. Siri is Apple’s.)