Boris Mann’s Personal Blog

Dec 2015

Bots will speak English before they speak JSON

Bots will speak English before they speak JSON

AKA How do you know if two replicants are chatting?

I responded to Dennis Mortensen’s tweet (he’s the founder of X.AI, a “personal assistant that schedules meetings for you”) on bots using English:

I was pointing out that it depends on the transport mechanism — that is, how the two agents are “communicating”. A machine readable meta-element in the communications could indicate that there is a bot-to-bot format to use, e.g. a JSON endpoint.

Dennis is right with this last comment, especially in the context of trying to bet or guess whether one bot is talking to another.

But what if you weren’t guessing? I suggested that an X-header in email as a pseudo standard might be an interesting start for early leaders like X.AI.

(I couldn’t find an authoritative link for learning more about X-headers in email. It’s an extension mechanism to put extra information in email messages, and Dyn has an article that talks about’s usage of some custom ones.)

Having two bots “speak” to each other over email could lead to some strange things.

For one, their internal systems would get trained on bot-to-bot interactions, when they should be optimized for bot-to-human. On the other hand, the efficiency of passing around chunks of machine readable code is, I think, desirable.

Or at least, the two bots could expose their interfaces to each other — their API as it were. And then we’re pretty close to the holy grail of two systems passing back and forth data that they know what it is.

Bot Literacy

AI and chatbots is a topic we kick around a lot at the office.

Job title of the future: Strong AI Groomer
Ryan Betts

We ended up having a discussion about the wave of messaging-based interfaces — businesses talking to customers using Facebook Messenger, or LINE, or WeChat.

Just like rudimentary canned responses in email, could bots be used to handle communications at scale? (yes)

Imagine someone talking to a bot through a web chat interface getting suggestions on what to buy on an ecommerce site. Is the general public going to know the difference between a calm and helpful chat bot and a well-trained customer service agent?

Will the bot suggest discounts if you say phrases indicating that you’re going to leave the site and shop elsewhere?

And finally, will people pass around conversation logs that result in discounts? Can you “glitch” the bot into unlocking its inner workings, and game the system?

Bot literacy is something for bots, but it’s something for people, too.

Will some of us want the “command line interface” for the bot, because it’s faster and more efficient than having to walk through a conversational tree?

Economics 2.0

With bot communications and Ethereum-powered smart contracts, it’s time to brush up on our science-fiction reading to see what the future might have in store for us:

“The crew upload their virtual states into new bodies, and find that they are all now bankrupt, unable to compete with the new Economics 2.0 model practised by the posthuman intelligences of the inner system”

Charles Stross, Accelerando (Wikipedia entry)

And, of course, Rushkoff’s Program or Be Programmed very much comes to mind as well.

Just like getting good at framing Google search queries lets you find information, having a certain literacy of interacting with bots (or understand that it is a bot you are communicating with) will become important.

I totally want an AI bot that does nothing but reads all my past tweets, bookmarks, and blog posts and tells me what I mean!