Why I’m bearish on bots

Bots are all the rage in Silicon Valley these days. Everyone is talking about them. Facebook has a bot built into Messenger. Google announced a platform for bots. Many startup pitches focus on bots.

But, so far, I’m not convinced.

I’ve worked on bot-related projects. I’ve worked in messaging, speech recognition and natural-language processing. At AOL, we had a local search bot in 2004. (Not my project, but we had one.)

Today’s bots are pretty stupid. They are based on simple keyword recognition. See this example from a recent Messenger conversation I had after I was hit by an Uber:


Ideally, the bot would realize the context and not even show the “Request a Ride” message. At a minimum, it should be frequency capped so it doesn’t appear every message.

Like I said, bots today are stupid.

Will they get better? Undoubtedly.

Will they replace a personal assistant in the near future? Unlikely.

Consider an obvious use case for bots: appointment scheduling. After all this time, appointments are still tough to schedule.

It sounds simple. Check two calendars and find overlapping free spaces. But it’s not.

Availability isn’t fixed. My availability can vary based on who is asking. Mark Zuckerberg wants a meeting? My calendar is wide open.

For me, multiple factors go into availability. What is my relationship to the person? Are either of us working on an important project? Have I met this person recently? Were they referred by someone I trust?

Have the assistant look at my address book, you say? Well, relationships vary over time.

Location matters. Again, this is context sensitive. I one joked with VC Shervin Pishevar that location varies based on importance of the two parites. Jokingly, we scheduled lunch at Madera, which is basically across the street from most venture firms. But there’s a lot of truth to that. If I’m meeting a friend who lives on the peninsula, we try to alternate peninsula and SF. If I’m meeting a VC, it is usually in Menlo Park. Meeting someone who needs my advice? Probably a coffee shop a few blocks from me.

Location also depends based on our local travel patterns. I try to schedule my meetings on the peninsula on one day to minimize my driving. I know some people spend one day in SF. I adjust to take that into account.

The rules vary for out-of-town visitors. Some of my friends visit from the East Coast. Some do it every other week. I can see them pretty much any time. But for a friend who visits once every six months. I’ll work hard to open up space.

Location can be a Skype call if I need to chat with someone who I won’t likely meet up with.

Timing matters. Is the request urgent? I’ll open up availability. If not, it might wait a couple of weeks.

I’m having surgery next week. It was scheduled according to the surgeon’s schedule because she has a specialized skill. I scheduled everything around it. During the recovery period, I’m not taking meetings. But if someone and something important needs to happen, I’ll suck up the pain.

Then there’s the issue of social projection. Maybe I don’t want to project that I have nothing going on and my calendar is wide open. Or I want to project that someone is really important to me.

Something that seems to be so simple can actually be really complicated.

Is my scheduling more complicated than most? Probably. But if it weren’t complicated, the utility of a bot would be less valuable. Could a bot learn all of these rules? Maybe.

It’ll take a lot of time to get there.

Maybe a bot can put it in on my calendar for 2020.