Chatbots and Echoborgs: Phil D Hall on the future of conversational AI

Last week, the Simpleweb Studio One team was treated to a talk and ‘fireside’ chat with Phil D Hall, founder of Elzware, about his work with chatbots and AI which he’s devoted his career to the study of. More recently he was technical director for the Echoborg project, part interactive theatre, part AI experiment. And right now, as well as renovating a house and building the odd bit of furniture, he’s working on building photorealistic avatars which, in his words “are really spooky… they’re the other side of the uncanny valley.”

Phil has been engaged in the definition and creation of commercially viable Conversational Artificial Intelligence with Elzware and other companies for the last 16 years. His interest is in building and consulting on semi-automated information gathering and ChatBot delivery, particularly with respect to: infotainment, healthcare and education. He is also a keen supporter of online community and culture and a champion towards acceptable ethics and privacy in Cyberspace.

The talk and conversation spanned from emotion and AI used in gaming to conversational power dynamics between humans and robots and whether Chat Bots really have a future in human society. So to the avoid the taunting ‘you should’ve been there’, we’ve put together some of our favourite discussion points and questions: answered.

Simpleweb: AI can be a tricky subject to get your head around so, for those that don’t know, what’s the difference between machine learning and artificial intelligence?

Phil D Hall: I’d normally describe AI as more an umbrella. At one end of AI is ‘unsupervised’ statistical machine learning. This is the machine intelligence that uses neural networks and other mathematical unsupervised stuff. For example, in the case of Google DeepMind, what it’s doing is taking the huge amounts of data they hold on people and applying rule structures and filters inside a neural network. So you can put ‘data’ in, say a photo of a dog, and the neural network works it out by asking ‘does it have ears’, ‘does it have a tail’ etc, and can then give an output – telling you that it’s a dog.

But of course there are all sorts of problems with this, including issues with data protection and GDPR.

The unsupervised end of machine learning is useful for some areas though. Voice recognition and intelligent search engines for example are useful because they offer something that humans physically cannot do.

In the middle of the umbrella you have supervised machine learning. This is the bit we really need to play in. These are hybrid systems that use both logic and statistics.

And at the far end of the umbrella you have rules and systems – the mechanical/logical end. So this would be things like pattern matching words and if/and statements for example.

SW: Alright, and we’ve heard a lot about EchoBorg and it sounds really cool! But what does it do?

PH: The Echoborg was built by Rik Lander who’s been doing contractive narrative in shows for a long time now. He started off taking an idea that was put in place by two guys at London School of Economics. What they did was an experiment where they took a room and put a human in it who was playing the role of a chatbot, then they invited another human in to take part in the experiment. Their role was to ask questions to the ‘chatbot human’, who had a headset that was being fed the answers from a real chatbot. It was all about power and dynamics between different people. Rick then took this experiment and turned this into a show using Artificial Intelligence Markup Language (AIML).

Conversational AI… or extending intelligence is what I’m trying to get at here. The issues we’re trying to solve involve looking at the structure of AI, partly on the basis of narrative.

You can take a more detailed look into Echoborg by watching a recording of one of the shows below:

 

SW: How clever is the AI behind Echoborg? Did you code for complex philosophical questions or was it entirely spontaneous AI?

PH: It’s a hybrid performance piece using various methods in the background – it has a set of pathways that evolve from the start of the system. So when we first put it together back in Berlin in 2016 it had 6 minutes of conversation, whereas it can now chat for 2 hours. The way that we worked it was through a lot of individual pattern-matching steps, then from the individual steps we have cycles of conversations which are exposing and will either enable or disable additional topics of conversation.

SW: You mention that the development of ChatBots has been slow. Do ChatBots really have a future?

PH: You need developers to build AI, you need engineers to build AI, you need mathematicians to build AI… or at least manipulate data into something that’s statistically intelligent.

But there is nobody training chatbot development in higher education.

The people who are learning how to build chatbots are people like us. It’s the developers who’re having to work it out.

But really, the term ‘ChatBot’ doesn’t do the concept justice. The way I like to position ChatBots instead is as part of a ‘conversational user interface’ which then allows you to talk more openly about it and do something that makes sense.

ChatBots are about managing conversations. You can’t be all things to all people. You can’t have a ChatBot that’s going to understand everything.

SW: What other potential do ChatBots hold?

PH: Well there’s potential to do some good as well. There’s a system that I’ve prototyped for dementia sufferers. We built the system so that it would be able to pick up the user’s cognitive function as it drops away and to be able to engage them in conversation where human beings can’t .

Even on the news just yesterday I heard there was going to be 75,000 more carers for old people required in our region in a few years time which is just not going to work.

The idea of building a ChatBot that can engage with people, helping and allowing them to have a conversation is just brilliant.

My mum has dementia and one day she decided that she’d been to drink tea in the plantations in India even though she hadn’t. When I asked a dementia nurse how she dealt with that she told me nurses were trained not to allow dementia sufferers to have flights of fantasy, so they just ignore them.

But if engaging in a flight of fantasy is what’s going to give what’s left of someone’s brain some level of entertainment, having a conversation about what it’s like to drink tea at a plantation in India, then why shouldn’t it happen? The only way it’s going to happen is by building a ChatBot.

SW: So if you want to change the future for ChatBots and start making one yourself, how would you get started?

PH: If you’re working with children (8-10 year olds) or are a complete beginner, you can have a play with someone else’s ChatBot using Scratch (a free programming language aimed at teaching kids to code), and then take a look at the back-end and have a go at changing it.

When I’m trying to bring anyone else up into building ChatBots from scratch, I tell them to download ChatScript – but don’t read the manuals, there’s 40 of them(!) – and build a pizza delivery bot. Try and make the choices and the loops and structures.

Or if you’re developers, think about how you would use functions to process data and, instead of processing data, process conversation.