INFLUENCER: "Conversational artificial intelligence is alive and well," says Paul Slattery, Technical Director at AKQA Berlin
One of the most natural things for humans to do – engage in a simple conversation – has been one of the hardest things for machines to master. Yet a new era has started where we might chat with machines more often than we do with each other – as conversation becomes the most effective interface for getting things done. Conversational AI (Artificial Intelligence) is alive and well in applications like Telegram, which passed 100 million active monthly users earlier this year. Facebook has re-launched Messenger as a platform with built in AI; x.ai has spent the last two years teaching ‘Amy’, its calendar AI, how to handle meeting schedules via email; and a 19-year-old has created DoNotPay, a free chatbot AI lawyer that has appealed $3 million in parking tickets. Let’s have a look at how these intelligent systems work, the advantages they present to us and where we might be heading.
Essentially, advances in conversational systems have three drivers:
- Emerging UX models
- Messaging systems
- Artificial Intelligence (AI)
Operating systems are focussing on speech as an integrated feature. This is often the first place where people encounter conversational interfaces. We are becoming more accustomed to the idea of asking Siri or Alexa for directions or a recommendation.
Even text however has lots of advantages:
- It is natural. There is no need to learn any interaction above typing.
- It has a low cognitive load. It does not cost a lot of brain power to read and reply.
- It is asynchronous. I can ask a question while doing something else, picking up the answer in my own time.
We can text bots as well as people. A bot is simply a piece of software that runs inside a messaging platform and can perform basic tasks. So you can send a message to a bot, and it will send you a reply. But it can also perform tasks for you. The bot normally has a finite set of functions. New forms of button, control or keyboard can enrich the conversation. There have been huge improvements in this space, a good example being the Telegram 2.0 bot platform extensions. These use an inline keyboard on a specific message that allow customised interaction without sending any messages.
Messaging is still the killer app for mobile phones. Messaging platforms are poised to replace much of what we use apps for today with a conversational interface. Integrations built into messaging platforms pipe messages around other systems and allow for seamless integration. WeChat thrives on its integrations. It basically combines functions of WhatsApp, Facebook, LinkedIn, Uber, and even Apple Pay into one platform. Fully integrated payment is driving the adoption of new services on WeChat.
Narrow artificial intelligence and cognitive systems have made tremendous advances in the last 10 years. All major technology companies are currently developing their own AI supercomputer systems such as Baidu Minwa, Google DeepMind, Microsoft Project Oxford and IBM Watson. The latter is probably the most famous example of a system that operates on large unconstrained domains to extract meaning. It was developed by the DeepQA team at IBM and uses a combination of algorithms and approaches that are arranged in a kind of pipeline for the task of competing against a human at Jeopardy.
A natural place for these advances is processing the vast amount of text sent over messaging systems and then using these to understand intent. Overlaying other data layers will allow for expanded topics within the conversation. An example of this is the pluggable knowledge system in Viv. Artificial Intelligence can be combined with human intelligence for a multiplier effect. AI can be used as a doorstep to initially process a request. By leveraging the advantages of integrated AI with a human backend this allows for scale and depth. It is a kind of large Turing Test.
Conversational interfaces point us in an exciting direction, liberating us from the many limitations of the mobile interface such as small screen sizes and unwieldy ‘shopping cart’ forms. Platforms such as WeChat show you no longer need to download and launch separate apps for each on-demand service. Over time, the use of natural language processing will automate these concierge services, while retaining a human in-the-loop to ensure accuracy and handle the long tail of requests that cannot be automated.
Paul Slattery is Technical Director at AKQA Berlin