A shit UI (basically these are automated phone decision trees) interspersed with quips and emojis is still a shit UI

Timely article! Thanks for sharing. I must admit I'm pretty swept up in the chatbot craze right now. It's exciting, and I'd love to add some context to the discussion.

TL;DR: We have two things happening: Messaging apps becoming hegemonic platforms, and the actual chat paradigm that the user interacts with inside the platform. The former is already happening, and represents a massive opportunity. The latter needs to evolve enough to make chatbots smarter and more useful.

Let me explain:

Messaging apps as platforms The popular messaging apps (FB Messenger, iMessage, WhatsApp, especially WeChat) are becoming aggregators of all the other apps and services we rely on. We're still silo'd in the West, but Asian markets give us a glimpse into the future where 99% of transactions - both social and commerce - are handled through WeChat. So big picture, people are going to start doing a lot more stuff inside messaging platforms. This is a huge opportunity for the chat paradigm gain wider acceptance. For background reading on this, I'd direct your attention to Why Messengers.

Chat paradigm As plenty of people point out here and elsewhere, the 'chat' paradigm has been happening for decades, powered through simple decision-tree logic. What is new is the mainstream acceptance and commercial potential of bots. Not everyone can use a command line interface, but everyone knows how to use FB messenger.

If we evaluate the success or failure of a bot on the criteria of using perfectly crafted responses to natural language, obviously we're far from 'successful' bots. Right now we should evaluate bots on the criteria of removing friction and adding convenience/friendliness to commerce transactions. A great example is The Edit, which texts you a new vinyl suggestion each day. This combination (chat interface with a basic reccomendation engine) shot them to $1mm revenue in <8 months. I don't know about you, but I'd call that bot a fuckin success. For background reading on scripting smarter bots, I'd turn your attention to Random Access Navigation.

Finally, shameless plug: I'm working on an ecommerce chatbot right now. If anyone is itching for a new side project, I'd be stoked to chat about it with you.

Back in my day we had a "chatbot" called Zork. It was a puzzle game, so being inscrutable and willfully obtuse was a feature.

Get ready to scroll!

website.fail. TLDR; Bots fail because AI isn't anywhere near where it needs to be yet to have a human conversation.

Interesting. The biggest issue I have with chatbots is the lack of clarity in possible interactions. Most of the time you have to try many things to see what works and what doesn't.

It'll be fixed eventually once we can start building more intricate systems but right now, I still haven't seen a chatbot that made me go "Woaaah this is so cool".

Bots to me are just drones answering with scripted responses. Back in the day we created bots on IRC to do the exact same thing.. i'm talking the 90's. They are scripts that don't have much depth. Sure you can make them smarter with machine learning but we are way far away from an actual bot to give meaningful responses.

good rundown. current-gen chatbots are limited and proprietary CLIs masquerading as personal assistants.

This maybe will change more quickly than most of the people is expecting, with the evolution of NLP and Reinforcement Learning: https://www.technologyreview.com/s/603501/10-breakthrough-technologies-2017-reinforcement-learning/

I found a typo: "Bots that do one thing well are more helpful that bots that do many things poorly."

Lol, I really had to laugh out loud at the grandma screenshot...

Current chatbots are just another type of wizard, you can't build your product as a giant wizard (or sometimes you can) and hope it's good for everyone.