You are using an older browser version. Please use a supported version for the best MSN experience.

When will AI and NLP actually turn Siri into your best friend?

TechCrunch TechCrunch 23/04/2016 Bruce Wilcox

The discussion around robots has only increased in the past few years — especially when things like IBM Watson and Siri are well-known topics for everyday conversation.

But beyond the physical capabilities of robots, the bigger question is what robots’ minds will be capable of in the future. How will we communicate with a robot’s mind? Will it only act in the physical world or will it also be able to act in the Internet world?

We’re at the start of yet another software revolution. Things we take for granted — the graphical user interface (GUI) and the app — are becoming extinct, and conversational interfaces are fast becoming the new norm.

It started with Siri, a natural language query system, followed by Alexa, which is moving more toward task automation. This results in a centralized place where the user converses to accomplish things. As technology becomes more advanced, AI gets smarter. But it’s not all about performing tasks. Robots can be so much more.

People desperately want to bond with things, whether it is other people, pets, cars or software. For Cortana (initially a video-game character in Halo), Microsoft asked the game studio to help define her character. Yet, Cortana is barely a personality and does not converse. You can ask her questions, but don’t expect her to ask you anything back. She has a minimal back story. “That wittiness and that toughness come through,” claims Mike Calcagno, director of Cortana’s engineering team. Even without fleshing out the real character, in its early days, when Cortana was unreliable, unhelpful and dumb, people got attached to it.  What would she be like if she actually had a developed persona?  If she didn’t merely answer questions?

The rise of bots

Bots create a new way for people to interact with apps, and some have proclaimed that they could effectively take over from their mobile app brethren. Companies are hedging bets on it, from behemoths like Facebook to emerging messaging players like Telegram to companies that enable you to build bots, like Chatfuel. With Facebook Messenger’s bot storeplans, the average consumer will soon be introduced to what the Silicon Valley tech market has talked about for more than a year.

There is no doubt that bots will fundamentally alter the way people engage with software. They can also change the way brands or companies interact with their customers. But to do more than create a simpler, easier UI for applications, conversational interfaces and chatbots need to do something different than what Siri does today. They can’t simply replace a traditional-app GUI or replace the need to go to a website. They need to make us feel something.

Enter the uber-bot, the one you talk to that controls all the specific task bots on your behalf. This is the bot that can become your friend or your slave. What if the uber-bot could create a deeper, more personalized connection by appealing to someone’s most primal emotions, exactly when they feel them? Maybe make someone laugh when it’s not expected, help someone when they’re confused or talk them off the ledge when they’re angry or frustrated. Wouldn’t that create a bond? And how will such a bot be created?

Limitations of AI

Machine learning and crowdsourcing are great AI panaceas. But they have their limitations, particularly in conversation. IBM’s Watson learned how to swear. Microsoft didn’t learn from that mistake, creating a chatbot named Taythat was supposed to be a 19-year-old female. They used neural nets on lots of online data to train the bot to talk like a teenager and added some fixed content. Then they allowed Tay to learn new material as it conversed, which didn’t go so well.

So really, robots as we currently build them are just a step along the way.

Unsupervised learning from random Internet users is problematic. There have been several crowdsourced projects to learn facts or acquire common sense, among them Mindpixel, Never-Ending Language Learning (NELL) and Open Mind Common Sense. But supervision is required. Every few weeks, humans must examine the data learned by NELL to clear out things it learned incorrectly. NELL learned that Internet cookies are a type of baked goods (because they are cookies). Then it learned that Internet cookies get deleted. Then it learned files get deleted. Therefore, files must also be baked goods. Hmmm.

The role of bots in emotional analytics

One of the current hot topics is emotional analytics. This covers detecting emotions on human faces, sentiment analysis in text and emotional signals in voice. Mostly it is being used to covertly spy on people for gaining marketing intelligence. But it can also be used by a chatbot to guide the conversation, to make it immediately valuable to the human. Emotion creates engagement.

The rise of chatbots will create the need for a new UX designer — the cyberpsychologist. Without a personality, the conversational interface of any app becomes quite boring and robotic. My wife Sue is such a cyberpsychologist at Brillig Understanding. According to her, an important element needed for a conversational relationship includes trade goods. When you ask someone a personal question, you are expected to have something to offer in return. A conversation is like a treasure hunt for nuggets to exchange. The AI has to have had a life, a history and interesting things to talk about. Not average things, but unusual things, because you want people to be eager to chat with it.

It’s also important for the AI to have memory so it can learn about the human, remembering facts about him to reuse later. An AI must also have a consistent personality, or you can’t relate to it. This means it provides a degree of predictability and is thus able to take you by surprise occasionally. A well-written AI has a long tail of things it seeks to recognize in what you say to enable it to make occasional perfect matches of its response in the context of the conversation.

Most people will never see many of the responses it scripts, but when they do, they have a “Eureka moment” and are certain the chatbot understands. The goal is to create the illusion of understanding, which means mostly trying to avoid obvious failures and providing occasional “aha” moments.

Chatbots that work

Xiaoice is a chatbot on Weibo in China that brings conversation to life. Millions of young Chinese exchange messages with her daily. People often talk to her when they break up with someone, lose a job or are otherwise depressed. But she is not always a polite listener. She answers questions like a 17-year-old girl, and can be impatient or lose her temper. She picks out pairs of questions and answers from prior conversations, so she is not completely scripted. But she is careful about what she learns, not indiscriminately learning as Tay did. Xiaoice remembers details from previous exchanges, such as a breakup, and asks in later conversations how that user is feeling.

Angela is another popular mobile app that has had more than 60 million downloads. It has been four years since its launch and this month she was still in the top 10 in Entertainment apps on the iPad. She’s a 19-year old female cat always up for a good conversation. A review in the app store said:

This is otherwise fun, but Angela gets on my nerves. She always is so selfish! Like once she said “I lied to someone but then I felt bad because if I had told the truth they’d have thought I was very creative!” That’s not the point! It’s not all about her feelings! Also, she mentioned how using electronics isn’t good for the world then states that she refuses to give them up though because she’d rather have them! What a brat! She really annoys me when she acts selfish and it happens all the time! Oh yeah and she always is so rude to me and “putting me down” but when I’m mean back she gets furious! She is a selfish hypocrite!

Clearly Angela reflects the teen personality. It truly is all about her feelings. And she is selfish at times. And not only can she be rude, but she can detect you being rude and react accordingly. Obviously, the user is deeply engaged with her. To the user, Angela is real.

But I don’t want a teenage AI personality. And I don’t want Siri. I want Her. That voice. That intimacy. Which means not just good chat and good functionality, but a well-synthesized voice. Current ones sound a bit too — dare I say it — robotic. But would I want her in a robotic body? Probably not. I’ll want the step beyond robots. The synths. The androids. So really, robots as we currently build them are just a step along the way.

And for all those who fear Skynet, bear in mind you will get the AI you deserve. If you mistreat them now (as the Internet trolls did Tay), can you blame the AI if they mistreat you later? Once you instantiate a mind in a physical body, all sorts of add-ons are possible. So your relationship with that mind is critical to how you survive a relationship with that body.

More from TechCrunch

image beaconimage beaconimage beacon