Chatbots are getting better at conversation. Or are they?

Bots are hot. We imagine them to be wickedly intelligent, like Samantha from the movie Her. The reality is duller. Early chatbots suck. They can’t chat. There is no “I” in their AI. They are so hopeless that some people argue that bots would be better without conversation. But without conversation they are not bots. They are more like the Internet on Nokia feature phones (a.k.a. WAP) from 1999. A screen full of dumb buttons. Seriously?

These button-based “bots” are not Siri-like conversational virtual assistants. They are menu-driven programs with a simple graphics UI. They are IVR reincarnate. Remember “Press 1 for arrivals, press 2 for departures, press 3 to check-in”? It was called IVR, for interactive voice response. You loved it, didn’t you?

Conversational AI is not here yet. That said, the buttons-based alternative is not an alternative either. The downsides are many:

1. Boring, un-sexy UI. From vivid apps to blunt Nokia ’99 experience?
2. Incompatible with Google Assistant API, which will be sending unstructured natural language queries to third-party bots. Bots that can’t understand them will lose Google’s referral traffic.
3. No deep linking, no random access. To get one specific answer, you must traverse the whole tree-like menu, from the very top.
4. No Borg Collective. Bots are siloed. They can’t pass information and users to each other. Are we willing to give up on the world wide web?
5. No voice control. Where would you display on-screen buttons if there is no screen? (Think Amazon Echo or Google Home.)

All buttoned-up

Who said the bots must be intelligent? In China, 10 million organizations have “bots” inside WeChat messenger. Are they all like Samantha? No, they aren’t. They are as dumb as a brick. They are not intelligent. They are not conversational. They are screens full of buttons. However, the Chinese use them and seem to be happy.

If it works there, maybe it will work here, too. Shall we forget about Star Trek, Her, and Ex Machina and just build millions of dumb bots?

In fact, that’s what Mark Zuckerberg wants us to do. Facebook’s subliminal message to developers is: “Your AI sucks. If you build a conversational assistant it will be crappy. And we don’t want crap on our platform. So don’t even try.”

Facebook is not alone. WeChat, Slack, Kik, or virtually any other messenger will tell you the same thing: “Don’t let users talk!” Some people are even considering removing the text entry box from the bot framework (or at least hiding it by default). Dan Grover, who worked at WeChat, once wrote a very popular article advancing this point of view.

Back to HTML 1.0?

Messenger is the new browser. Unfortunately, it’s a rather lousy browser. In its feature set it would hardly match Netscape 1.0. For example, Facebook Messenger API doesn’t support even basic formatting. There is no bold, no italic, no font size, no color, not even line break.

To add insult to injury, these new browsers don’t support existing websites. Developers must rewrite their websites to work with these new messenger interfaces. And they have to do it not once, but nearly a dozen times: for Facebook, Slack, Kik, Telegram, Skype, WeChat, WhatsApp, Google Allo, and who knows what else. Needless to say, messenger platforms cannot agree on a single standard. (Thank you, Sir Timothy Berners-Lee, for giving us HTML!)

No deep linking, no random access

Most of the buttoned-up bots don’t even support free-text search. You cannot ask them. They ask you! (“In Soviet Russia, Google asks you!”) It’s like a police interrogation. And you cannot even answer their questions in your own words. They give you a couple of predefined options to choose form, by tapping on a button.

Can you imagine a Wikipedia bot that cannot answer a specific question like “What is the capital of Brazil?” How would you find the answer through all these buttons and interactive menus?  You would have to somehow click through a menu like Geography -> South America -> Brazil -> Key facts -> Capital.

In fact, the buttoned-up bots are even worse than websites without site search. They are like websites without bookmarkable webpages. The are no page URLs. There are no deep links. Unlike a website, you cannot even browse the site by following the links. There are no links. There is only a tree you must traverse from the top.

This is not a Borg Collective

After the bot revolution erupts, every organization that has a website will also have to have a bot. Can you imagine completely isolated websites, not connected in a world wide web?

This is what the buttoned-up bots future would look like. The bots will be siloed. The lack of random access means that the best interconnection between these bots could only be at the top of their menu trees. The web is great. Losing it would be a huge leap backward. Are we willing to give up on it?

Want Google’s referral traffic? Understand natural language!

Unlike Facebook, Kik, Slack, and others, Google Allo and its embedded Assistant are built around natural language interactions. And Assistant will be the master bot, a virtual 411 operator. You won’t talk directly to those millions of bots representing every organization on the planet. You would ask Google Assistant and it will figure out which bot can handle your request and pass it on. That bot must be intelligent enough to understand and act on it.

Amazon’s Alexa and Samsung’s recently acquired Viv also want to be such master bots, forwarding natural language queries to subordinate bots.

Apparently, dumb “buttoned-up” bots are a throwback and a dead-end. Is there an alternative? To be continued…