The wrong people are doing the wrong job, too fast and without any forethought, to capitalize on the tech news hype cycle more than anything else.
Conversational UIs are good for two things really - dealing with massively selection, and providing an interface for entirely naive users.
Ultimately, I see them as much an extension of search interfaces as they are of messenger apps, insofar as given a potentially open-ended number of buttons (like restaurants in your area) it's easier to ask for them in english, despite the fact that the interface has anticipate all the issues that brings (you might not get it right, your request may be ambiguous etc)
And, sometimes it's easier to use english to make a selection - particularly if you don't know how to do something that should be relatively simple, and is consequently relatively simple to describe in english ("Hotel room, turn on the lights." "I want to order a pizza."). Unfortunately, this takes both advanced natural language parsing, and equally advanced response generation. The later is much harder, and I feel generally neglected, by most chatbots and platforms, IMHO.
This makes a lot more sense when using voice rather than text, but ultimately, hybrid voice/gui chat displays are what chatbots are going to evolve to. You will ask for something via voice and in addition to receiving audio responses, you will receive a stream of UI widgets in a chat window containing supplemental information and finer grained controls, such as sliders for the color/brightness of your hotel room light, or an options menu for your pizza order.
Amazon Alexa and Google Now are already quietly doing this, for the most part, but it's going to become a much prominent part of the interaction over time.
After a couple of months of training it it could keep up a conversation for a few minutes along most of the tracks that it had learned. To make it look more life-like I added random delays in answering and sometimes splitting an answer across two messages. Spelling errors were retained as well instead of being corrected.
Of course that's not a 'bot', merely a recorder and replay but it was at least as effective as long as the conversation was kept relatively short.
I think I should be able to ask your chatbot how their day is going, but I don't think I should be able to shitpost with it for 3 hours.
The reason I bring up Chatbot is because they seemed to at least have the conversational part figured out.
Also, Siri is really not such a part of this conversation because it is honestly the only chatbot that I can think of that accomplishes a portion of what we're talking about here.
These little things - even though programmed by some other person - can spice things up and make the bot look even more natural.
Go home Siri.
"Make sure you don't overcook that egg" or
"The suspense is killing me".
Failed cases : 1. Lookup Wikipedia during an amazing race 2. Play a specific song 3. Find an event in my calendar
- It won't work in a loud bar
- It will almost always work in a car where you're the driver and there's a backseat conversation or the radio is on.
- It's fine with street/sidewalk noise.
- Not-terribly-loud coffee shops, stores, etc. are also usually just fine.
That's not how we'll be able to achieve real AI soon.
The reason why 90% of chatbot companies aren't living up to the hype is because they're doing it wrong. Either it's an phone tree, or they're doing naive matching on regexps, or perhaps just throwing a bunch of data and trying to deep learn some intents. It's pretty amateurish. If you used more traditional NLP techniques, you'd get much further.
The other problem with chatbots is that people have become accustomed to general chatbots like Siri, and so even if you say, "my chatbot only does X", you're still going to have people trying all sorts if stuff that's not X. ("Hey, banking bot! What's the melting point of lead?" "I only do banking question." "You suck banking bot.")
We are just way, way off technologically.
Originally designed for and lessons from the game company formerly known as Jellyvision (the game company part of that is now spun off into Jackbox Games), the Jack Principles was a briefly free resource during the late part of the IRC chat bot golden age in the 90s.
The consulting firm that is left at Jellyvision still uses it as a part of their consulting, so they don't give copies away for free anymore. I'd love to see a new edition of the Jack Principles in textbook form, especially now that it is very relevant again.
Funny, I see the interface of bots on a messaging platform as their only value. As for the AI/NLP, Meh. To me the possibility of replacing the umpteen apps and Web pages with a single consistent UI is the appealing part.
In many ways this new generation of chatbot is worse than what we had 20 years ago in IRC -- at least then the functionality was clearly presented.
Well put, sir. I hadn't managed to put that into words quite as succinctly but you've pinpointed exactly my feelings and my reasons for skepticism.
Unless you bother to read up on the commands that it accepts you find yourself greeted 50% of the time with "I don't know how to do that". Now we're on the same turf as any good UX problem - power-users (who have the incentive to learn an interface) vs casual users (who don't and shouldn't have to). Chat bots seem to be targeting the latter but they require the time investment of the former.
I've grown very tired of all the differing UIS I must navigate on a daily basis.
I've never met a command line that wasn't obfuscated. Try learning Unix using only man pages as reference. With bots, you can at least expect reasonable responses to some natural-language requests.
Much like the examples in this article, I haven't seen any bots that can present information in a way that is easier than a normal interface, or that even counts as "conversational."
Maybe I'm being an asshole when I say this, but it just doesn't feel like these chatbot companies have put in the sheer man-hours to work out conversations. They're feeding you the information they have, through a small number of prompts that they predetermine. 100% useless.
Where are the creators of Cleverbot? Shouldn't they be using their platform to capitalize on this? That was the most conversational bot I'd seen so far and it came out forever ago.