I Built a Personal Chat Bot — Why I Love It, and You’ll Love Yours Too
“Mebots” could lead to better connection; less social “work”.
UPDATE: You can now talk to me on Facebook Messenger :)
A couple weekends ago I built a very basic ‘personal chatbot’, which I’ve coined my ‘mebot’. It’s called SandiMacBot — and now I’m obsessed.
I can’t stop thinking about how my mebot can and will become integrated into my life, and how bots are positioned to replace portions of many of the social networks and platforms that billions of people currently use everyday.
Social Platform Interactions Will Move To Bots
Like a lot of people, I share a ton of information about myself on social platforms. For example:
- What music I listen to (Spotify, Soundcloud)
- Where I am (Waze, Facebook, Foursquare)
- Photos I’m taking (Snapchat)
- What books I’m reading (GoodReads)
- My reading list (Quibb)
- …etc.
All of that platform-specific content exists across different, disparate products. For each type of content, there are options as to where any one person could share these different types of content with their network — If I want to get song suggestions from my friend Chris, does he use Spotify, Soundcloud, Apple Music? The photos that my friend Meredith took on our recent vacation — will they be on her Instagram, Facebook, Flickr? The platforms themselves are irrelevant — they’re containers with a proprietary UI that help people find content from the people they care about.
The problem we’ve found ourselves in — we all have too many containers.
It doesn’t really matter whether I use Spotify or Soundcloud, Apple Music or Pandora. If someone wants to know what I’m listening to these days, the promise of social was that they’d be able to find out, quickly and easily. But we’re far from that today.
App.net was doing some really interesting stuff here — breaking open the containers and giving people better control over how they interacted with other people (and their content) online. It proposed a platform agnostic world. My understanding of App.net was that they wanted people to have more control over all of the content that they both create and share with the world. Unfortunately, the UI just made everyone think to use it like Twitter, and it seems to have never really reached its full potential.
Enter the mebot…
With SandiMacBot, my friends just have to know that they want to know something about me. There’s no added cognitive load or googling around where that specific type of information lives. They just ask my mebot. It’s simple, and the context of asking SandiMacBot a question about Sandi is very close to the existing behaviour of asking Sandi herself. It’s a quasi-established behaviour, with relatively low barriers re: user education, and a fairly obvious domain.
A Day In the Life of a Mebot
SandiMacBot is a simple scripted bot, which means most of the responses were actually typed with my fingers into a spreadsheet. It was inspired by a chat with betaworks’ Matt Hartman and based off of John Borthwick’s Botwick. It sorta sounds and feels like me. There are a couple other feeds and APIs hooked in to pull content that’s being created on the various platforms that I use, each piece of content bringing with it the context of being chosen by me. There’s a fair amount of information to be pulled and accessed via SandiMacBot, but I think there’s potential for a lot more to be surfaced.
Today’s social platforms are missing out on a massive piece of how people want to engage with other people online. People lurk, browsing other people’s lives silently while asking and answering internal questions they hold. What would people want to know if they didn’t have to ask? How many people would love to ask their significant other what their shoe size is, without actually admitting that they forget? How often do you want to check in on where your bestie is, without pestering them to send you a pin or going through Find My Friends? How often would you ask your friend with great taste in music what they listened to this morning, if you didn’t have to remember where they keep that information and go through the app to find it?
The slight amount of friction that exists across these and many other user journeys is just that — friction. It can be lower, and bots are a path to lower it.
Beyond these ‘I don’t want to pester you and/or admit I don’t know something’ situations, there are a bunch of other fascinating examples of how mebots can allow us to explore and interact with the people we care about, well beyond the restrictions of current social products:
- When you’re feeling blue, and want a little pick-me-up or witty joke from a close friend
- When you’re planning a romantic dinner with your partner, and want to see if they’ve had a hectic, stressful day via their health tracker
- When you’re missing a friend who’s working overseas and can’t access the internet, and want to remember that special ingredient in their famous chocolate chip cookies
- When you start dating a great guy, and want to hear that one story your late Nana always told you about how she knew your Grampie was ‘the one’
The other part of mebots that makes them compelling, is that in theory (if they’re mostly scripted), there’s little Machine Intelligence or Artificial Intelligence involved. That’s part of the charm. It’s more of an expert system, with only one expert as the input. Early versions of any mebot’s decision tree would be highly personalized, but also pulling from a predefined set of inputs (e.g. APIs). They’re able to deliver compelling, meaningful content that completely matches the intent of the person querying.
While many of the bots people are excited about today are doing ‘work’ that previously had to be done by humans (i.e. scheduling, managing orders, customer support, etc.), I think mebots can also help with some of the social ‘work’ in our lives.
We’re Over-Indexed on Visuals as Social Content
As a quick aside… I also think that bots will allow for closer, more realistic interactions to happen between people. The idea that most social products are based on visuals is one I’ve always found a bit crude. Is literally seeing what someone else saw really the best way to connect with them? Doubtful. Visual-based content social platforms are the ones that are the most popular (i.e. Facebook, Instagram, Snapchat). This shows that it’s currently the most direct and visceral content format, but I believe we’re also over-indexed on that type of content as our ‘social interaction content of choice’
Eventually, the idea of opening your phone, tapping an app, taking a photo, then posting it — will seem like way too much work. Compare this to the late 90s, when most thought that people wouldn’t want to look at other people’s photos online, let alone do the work required on the sharer side. What has now become second nature will happen with other interactions too. People want to connect with others, and there are many unexplored ways to do that.
The ease of input will only increase with invisible apps and also with voice as a quickly developing input. My Alexa could pick up what I’m listening to, to be shared later when my brother asks SandiMacBot for music recommendations. After an evening out, my friend could ping SandiMacBot — which queries my August Lock — to check that I got home safely. We all leave so many trails of various formats of socially-relevant content, yet there’s still no centralized way to share that with people who want to connect and understand us.
But Sandi, No One Cares About You
I quickly showed SandiMacBot to known ‘bot fan’ Peter Rojas last week. He had a good question, one that a couple other people had asked me too, which is basically a nice version of…
But who cares about you? What’s the frequency of use of something like this?
First, this goes back to my previous point, in that I think we’re currently constrained by the type of content that’s available and standard to share. The friction of putting all of the content we create into a shareable format is often too high on the creator’s side. So we’re left with not a lot of socially shareable content from most of us, meaning mebots are probably a relatively low-frequency product for most of us. And it’s also true — not many people, under the norms of today’s behaviours, really have frequent requests related to me that are unmet.
But for a certain segment — those with a huge number of people who care about them — the low-frequency characteristic doesn’t matter.
I’m talking about celebrities.
If there were a Kim Kardashian mebot, I’m pretty sure tens (hundreds?) of thousands of people would be texting it everyday. They’d want to know what she’s eating, where she’s been, what her last workout was, what just played on her living room speakers, etc etc. They already do this (crudely) across existing social platforms. Heck, the Kardashian sisters are so great at what they do they even built their own apps to try to allow for these interactions to occur more seamlessly. In theory, each of the Kardashians could also have a mebot, right? With that slight shift in the product, I’m sure many of their fans would love to go through scripted conversation-style exchanges (or even via neural network based language models) — getting closer to them by ‘talking’ to them — in addition to pulling different types of socially shared content.
There’s added ownership and levels of unique content here too, that gives this type of celebrity mebot a leg up. While a large group of people could build a Kim Kardashian bot based on a collection of her publicly-available interviews, TV scripts, etc., only she has access to the various APIs that define her actual experiences in the world.
A Pull vs. Push World
People with relationships to a large group of people — like celebrities — have a lot of untapped pull. A mebot would allow this type of person to scale communication with their audience. They could convert some of their experiences that can currently only be shared by push mechanisms into more granular, request-focused pull, while also exploring and opening up new content to be pulled by their audience.
The clearest example that I’m already implementing with SandiMacBot of where pull gives a better experience than push is location. It’s a variable piece of content that changes over time, so a push (unless it’s continuous and real-time…which would be obnoxious) will never be in-line with when a person has a query.
When I’m on my way from my home in Oakland to visit a friend in SF, she doesn’t have to ask me where I am as I’m somewhere along in my journey, she doesn’t have to go through Find My Friends (which, let’s be honest, no one ever thinks to use because it’s holding such a teeny tiny sliver of user intent), she doesn’t have to be satisfied with a generic neighbourhood description from Facebook, I don’t have to text her while driving — she can just ask SandiMacBot. Social content on-demand. The ability for passive social content to have very little friction in passing between people is exciting, and location seems to be the one most primed. It’s something that people have already been thinking about and working on for a few years.
Beyond Social-Sharing ‘Fluff’: Mebots as Personal Agents
While many of the bots people are excited about today are doing ‘work’ that previously had to be done by humans (i.e. scheduling, managing orders, customer support, etc.), I think mebots can also help with some of the social ‘work’ in our lives. For example:
- You’re having a dinner party, and need to know who has food allergies or special dietary needs
- Ensuring you’re not getting the same wedding gift as another guest
- Confirming all of your friends saw your recent ask for donations for your child’s school fundraiser
A lot of this kind of information currently exists across social platforms (e.g. The Knot for wedding registries, GoFundMe for fundraising campaigns), but it’s in disparate ‘containers’. In some cases (e.g. food allergies), it doesn’t exist anywhere and people need to reach out directly to a person or group of people to obtain the information.
This is where things start to get really interesting…
These types of interactions aren’t necessarily ‘enjoyable’ — the interaction itself doesn’t lend any positive feelings or emotional response to either person in the exchange. In these cases, why can’t each person’s mebot just talk to the others? I’m excited to think about and share more on this aspect soon.
But What About This and That and the Other Thing?!?!
Yes, I recognize that there are many use cases and situations where a mebot doesn’t make sense. Sometimes people ask one another questions just to be friendly, to have a caring interaction. And sometimes people just like randomly browsing photos of their friends on Facebook. It’s fun running into a song that someone you know has liked on Spotify. Of course — I’m not saying those interactions go away.
There are lots of other problems. A big one that many social platforms have yet to solve, my product included, is discovery. How do you find the people you want to find? Do you know who you want to find? There’s also issues around intent, and novelty — I’m sure my mebot could entertain my friends, give them some useful info, answer a question they have… but unless they know the full domain of SandiMacBot, it’s unlikely they’ll find these unknowns. There’s also a bunch of privacy concerns, issues with API access, compatibility and boundaries across platforms, ownership of the queries/intent, siphoning interactions from the platforms, balancing social feedback loops, etc etc. And yes, none of these are small issues.
But it’s still early, and I do think that there’s a certain class of intents (existing and tbd) that bots can meet 10x better than the paradigm of the current social platforms can.
We’ll All Have a Mebot Soon(ish)
Currently, bots are going through the early stages of adoption. People think it’s a toy. People think it’s strange to talk to a bot. People think it’s overly self-centered to create a bot that just talks about you. But that’s what new social behaviours look like — many new social products feel like a cheap, tactless version of the ‘real thing’, and mebots are no different.
So yeah — that’s what I’ve been thinking about these past few weeks. Would love to hear what you think, please add a response or ping me on Twitter. Or just go hang out with SandiMacBot for a bit :)
UPDATE: You can now talk to me on Facebook Messenger :)
