Tag Archives: ai

Uncanny Valley

We’re increasingly interacting with machines that masquerade as humans, either as an online chatbot, voice activated tools (Siri, Alexa and friends), and just occasionally as robots. As long as the machine is obviously a machine we’re comfortable interacting with it, and we’ll make allowances for its robot brain.

As our ability to use high levels of artificial intelligence delivers more human-like machines it will become harder to determine whether a machine is a machine based on the interaction.  When the machine starts to imitate a human well we’ll interact as long as we see the human.

The “Uncanny Valley” describes the feeling of being disturbed by interacting  and having the realisation that the machine isn’t as human as we’d thought. Or, conversely, knowing that we’re interacting with a robot but finding the speech, look and movements terribly life-like. It’s that moment when you think “oh, creepy”.  The term was created more than forty years ago by Robotics Professor Masahiro Mori.

As the robots become more human like this effect will, in theory, disappear. After all if we can’t distinguish the machine as a machine then we won’t have the discomfort. We still have some way to go but robotics engineers and AI programmers are getting us closer. Take a look at SAYA, the reception robot created by the Koboyashi Laboratory at Tokyo University of Science. Note this is from 2009, so I’m sure there have been advance in design and interaction since then, in the meantime SAYA seems uncanny to me.

Image:  Templum Ex Obscurum  |  Narshe Talbot |  CC BY2.0

Chat; the New Web

2016 July Chat

In those heady early days of the web personal sites were the rage. Who can remember GeoCities? I had a book review site online back then, I don’t anyone read it – not even my mother. As the web became more prevalent and a commercial option personal sites were pushed to a fringe and later into blogs.

Then came social media, with the biggest platforms attracting millions of accounts (that’s Facebook, LinkedIn, YouTube, Twitter et al). Increasingly social media is being exploited by business and absorbed into communications and marketing teams as another channel. Random conversations are harder, and the troll element can make using the platforms a pretty horrible experience.

Along came messenger tools, with WhatsApp listed as the second biggest social network in some analyses.

Rise of Messenger

Which is great for individuals. But much harder for businesses to exploit, they have to automate responses and processes. For example you can order a Domino’s pizza via an emoji sent via FB messenger. What a world. But to do that you need to first set up your “favourite order” and payment info on a Domino’s account and connect it to your Facebook account. So there’s a process designed to get your pizza order out to you.

This is a simple transaction. It’s essentially a yes/no question you’re answering. Other uses of messenger and chat apps are more ambitious. My bank now offers support via chat, which is a brilliant idea, except that it’s all in Dutch… now my Dutch is OK for day to day things, but my spelling is pretty atrocious so I end up flicking between Google translate and the chat bot – the chat bot gets bored. (I am aware I have just ascribed a human emotional reaction to a piece of software).

Chat bots, the tools companies use to make messenger apps scalable, can only answer the questions in its data base; the “known knowns”.  Which means they can serve as a sort of “FAQ” service, which can be helpful to cost-cutting businesses but less so customers with complex questions. Which means that companies are looking to AI to broaden the range of solutions offered and make the responses smarter.

Artificial Intelligence

Artificial intelligence is the ability of machines to learn and apply reasoning and from a geek perspective it’s exciting. Tech companies are competing to see who can pass the Turing Test. But it’s also a way for businesses to scale conversation, that essential element of human interaction could one day be done by machines.

The experiments in artificial intelligence are exciting and sometimes disturbing; Microsoft’s chatbot Tay was racist within a day. For something more fun, Project Murphy uses Skype and image swapping and watches your reaction to judge how well the match worked.

Most recently I found a website that is completely given over to a chat function. There is no other content at all, and you’re forced to engage, here’s the background.

So of course I tried it.

The engagement possibilities are limited, you’re clicking on a button to go forward, sometimes choosing between two options. It’s got some of the socialisation right, the humour in the interactions works and the Bear Bot cleverly waits a couple of interactions to ask my name. It’s a simple trust builder.

But then it goes a bit wrong. When asked a question I didn’t have enough space for a fully thoughtful response, and couldn’t edit it when I discovered the size limit.  Despite entering nonsense I earned “six fish” for my contribution and the thanks of the Bear Bot. This didn’t really increase my trust – pretty sure my contribution is not being “kept in mind” unless by that they mean “saved on a server somewhere”.

Screen Shot 2016-08-17 at 10.34.05I kept going. But I don’t get to talk in this chatbot, I’m once again a passive consumer of the information chosen for me. And then suddenly there’s a suggestion of inviting someone else into the conversation. That someone is Oliver Reichenstein, he’s kinda a big deal in user design, here’s an interview where he talks about some of his ideas.
Screen Shot 2016-08-17 at 10.36.24

I said no and earnt zero fish for that answer.

But seriously was Oliver there? It seems a poor way for him to spend his valuable time.

Overall the experience of the website as a chat was disappointing.  It seems to be pitched as a way for me to discover information – I was given one interesting link early in the conversation but I’d have found the information faster with a simple search. The interaction feels regressive, most of the time I was given one or two choices, like e-Learning in the 90s. I’d enjoy the interaction on an old-fashioned discussion forum more – if you find the right group the interaction and expertise are awesome.

Having said that I do appreciate that the company is trying something new, I’m curious to see how their experiment evolves.

In Conclusion

Chatbots are useful, they may even be able to support us on more open tasks that deal with the “known knowns” using strong databases, good process design and AI.

But until we pass the Turing Test conversation and interaction will not really scale.