A Blog by Jonathan Low

 

Aug 27, 2016

What Bots Need To Win Us Over

Soul, and a sense of humor? Sounds like a real relationship, which, just maybe, is what it has become. JL

Robert Hoffer report in Venture Beat:

Try building bots that use responses that are at least slightly fun. After all, who wants to talk to a “robot”? If we wanted to do that, we could call an actual person working in a call center and let them read us a dull script. (And) what soul means is that it needs to feel like the bot actually cares.
You can’t offend a bot. Most of them don’t have enough of a personality to be offended. Not only that, few of the bots available today are smart enough to debate with anyone. There are plenty of bot development tools, an immense amount of venture capital, and a massive number of bots. But no one wants to speak to a dumbot. They’re frustrating, and their very existence puts the botverse in jeopardy.
Let’s start by defining dumb. Primarily, it means the bot doesn’t understand basic language. Apple Siri and Amazon Alexa are not that smart. If you say, “Hey Siri, I want to eat shawarma in Cannes,” Siri will get confused and might give you navigation directions instead.
We can chalk that up to the fact that Siri has a limited vocabulary. You’d think typing what you want would help. Say you use a chatbot on Facebook Messenger to help you find a book. Should be easy, right? But first you have to authorize the bot for your Facebook account by clicking a Get Started button, which is an extra step compared to friending someone. Note to Facebook: This should be much easier. Good thing I don’t have to go through extra steps when texting my buddies.
Then the real fun begins. I hit the button, and the bot asks me a question. My natural impulse is to answer, but the bot doesn’t understand natural language. When I asked about the latest mysteries, the bot didn’t know what to do, even though it had listed Mystery as a button I could click.
What’s happening here?
When ActiveBuddy launched SmarterChild in 2001 we had two development teams. One team was devoted entirely to developing the tools required to do basic natural language processing and chat, and another team developed the applications we called “domains” that made talking to SmarterChild feel as close as possible to chatting with one of your IM buddies.
This was the goal. And it worked, albeit … not always. Sometimes we failed to understand what people said to SmarterChild. But when we didn’t parse the text correctly, SmarterChild would apologize and ask for clarification. Again, this felt like chatting with a buddy. By wrapping apps into chat we were able to grow SC into a massive audience. By using NLP tempered with personality and humor we created brand focus and loyalty over 15 years ago without the NLP having to be perfect, well before machine learning showed up. Surely today’s technology ought to be able to do at least as well. But the present situation feels like we’ve gone backwards.
Where is all of the brilliant machine learning? Where is all the cutting-edge A.I.? In my last post I complained that bots need a personality — a soul– to become intimate with their friends. I should have demanded more. What follows is my eight-point bot manifesto that lists the attributes all bots need to succeed, arranged from easiest to hardest.
Bots need:
  1. Basic chat skills: “Hi, how are you?” “Not bad, you?” “I’m doing fine.” “Cool.” Since I’m learning Italian, I can sort of have this conversation in that language. It didn’t take me that long to learn to say grazie mille. Basic chat has been a solved problem basically since before we launched SmarterChild. With machine learning and A.I., this should be a piece of cake. Yet, time and again, bots can’t do basic chat. Shame!
  2. IM-ability: Bots should feel like people in your friend list. Platform builders shouldn’t keep making it seem like we’re downloading an app. We don’t need barriers for bots.
  3. Multi-modality: When displaying a button — which could be a useful interface tool — at least let people type their selection instead of, or in addition to, pressing the button. Displaying a button and then not letting me type the button text makes your bot look really stupid.
  4. Memory: I’ve said it before: Dorybots (those with memory loss, like the blank-minded blue tang in Finding Nemo and Finding Dory) are frustrating. Each time I talk to one of my friends, they learn more about me; my bot-friends should learn about me, too.
  5. Domain knowledge: Bots need basic (if not really deep) sector knowledge, so if you’re a White House bot, how about being able to answer the question “Who lives in the White House?” I know the answer; heck, everyone knows the answer — except chatbots, apparently.
  6. Personality: Famed talk show host John McLaughlin passed away recently (R.I.P.). Most eulogized his loud booming style as “bigger than life.” Brands have personalities too, often characterized by slogans and jingles. Make sure to translate your brand personality into a bot personality. Advertisers, please make your bot “The Real Thing!”
  7. Sense of humor: Maybe it’s humor, more than anything else, that makes us human. Bots shouldn’t take themselves too seriously. Even Alexa tells jokes. A friend of mine once wrote, “Even if people think you’re stupid, as long as they also think you’re harmless, you can get endless mileage from a smile.” Try building bots that use responses that are at least slightly fun. After all, who wants to talk to a “robot”? If we wanted to do that, we could call an actual person working in a call center and let them read us a dull script.
  8. Soul: This is the hardest thing of all. In the case of bots, what soul means is that it needs to feel like the bot actually cares, like some work went into the development of the bot, and like the conversations the bot has with its human friends in messaging are important to someone on the dev team. The Oxford Dictionary’s definition of soul fits: “Emotional or intellectual energy or intensity, especially as revealed in a work of art or an artistic performance.”
So there it is, folks. A straight-talking, tough critique of the current state of our botverse.
But let’s end on an optimistic note. Nobody is giving up. More and more people are throwing their resources — both financial and intellectual — at solving the problems with bots.
The reason is simple. Bots today are in their infancy. Computer applications tend to last much longer than we all expected. I’ve been using programs on my computer — like word processors, databases, spreadsheets, IM applications, and even web browsers in various iterations — for the better part of 20 years. So take heart: We’re going to get this right. We’re a lot farther along today than we once were, with broad adoption in the development community, strong interest by brand builders, strong support by platform providers, and the backing of the entire venture industry. Instead of a lone commercial bot company and three messaging platforms, we’ve got the entire world pulling for bots.
In the classic movie The Outlaw Josey Wales, the character Lone Watie — as played by Chief Dan George — says, “Endeavor to persevere.” You wait, then you act. That’s my advice to all of us working on bots: Endeavor to persevere. Having been at this a big chunk of my adult life, I feel a lot like that character. While I’m proud we’re going in the right direction, if you folks don’t start getting this right, just like Lone Watie, I’m ready to declare war!

0 comments:

Post a Comment