How may I help you?

By Greg Stevens on August 1st, 2012

Chat bots have been around for a very long time. One of the first was a computer program called Eliza, written by Joseph Weizenbaum in 1966. It was designed to participate in a conversation in a way that parodied a psychiatrist. Because almost all chat bots created since then have worked on the same principles as Eliza, it’s worth taking a moment to look at exactly how Eliza worked.

The knowledge and personality of Eliza were realised in the form of pattern-matching templates, sets of rules that said: “If the input from the user matches a particular pattern, then produce this other thing as a result.”  So, for example, Eliza had rules such as: “If the input sentence contains the words MY MOTHER, then reply by saying: TELL ME MORE ABOUT YOUR MOTHER.”

The rules could be complex. For example, there could be a rule that said: “If the input sentence had the phrase MAKES ME FEEL followed by some words, then reply by saying: WHAT ELSE MAKES YOU FEEL followed by the same words.”  With this rule, a comment like “Sometimes my job makes me feel like a cog in a wheel” would get Eliza to reply with, “What else makes you feel like a cog in a wheel?”

Rules could even refer to other rules. This was useful, because there are a number of ways that a user can say the same thing. For example, Eliza might have one rule that generates a response for any inputs that begin with the words “MY JOB IS…” and another rule that says, “If the input begins with MY OCCUPATION IS, then use the same rule that is defined for inputs that begins with MY JOB IS…”

As you can imagine, even the most simple-minded conversations can require hundreds or thousands of these rules. But, with enough rules, even this simple system can pull off some very convincing conversations. Dr. Richard Wallace is the author of one of the current leaders in the field of entertainment chat bots, Alice (Artificial Linguistic Internet Computer Entity). Alice has won multiple awards and competitions designed to pit Chat Bots against each other to determine which one is the most realistic. It consists of over 41,000 pattern-matching rules.

When you understand how these systems work, it is clear that they don’t “understand” conversations. They don’t have any knowledge about the things they are talking about, and this, according to Mark Tindal, chief technology officer of Ai-Applied, is the key to the problem.

“None of the virtual assistants that we offer uses pattern matching,” says Mark Tindal. A virtual assistant doesn’t need a bunch of chatty rules that allow it to talk about the weather or how your day was going. “You don’t call up a customer service centre, ask to talk to a supervisor, and then ask them how their day was,” says Tindal.

Ai-Applied is offering a customer service tool. In order to be a functional tool, the agent has to actually recognize the words the client is using, and have the right knowledge to understand them. This technology does exist, in the world of academia and artificial intelligence software. There are natural language processing systems that can take a sentence and understand which parts of the sentence are the noun or the verb, and they can find the words in huge databases of knowledge.

These knowledge domains, called ontologies, contain thousands of terms, along with information about those terms. Organisations have painstakingly put together ontologies on all types of topics and industries, ranging from electrical engineering to mid-sized consumer product manufacturing. Placing natural language processing together with large, industry-specific  knowledge bases creates a powerful platform for a customer service user interface.

Yet the connection with the customer service industry is not there. and this is the niche that Ai-Applied hopes to fill.

[midk]

“They are not ‘chat bots’. They are virtual assistants,” says Mark Tindal. This is an important distinction, especially for a one-year-old company that wants to bring artificial intelligence technology to customer service. From a public relations standpoint, the industry of natural language communication with computer programs – whether you call it  “chat bots”, “virtual assistants” or something else – is crippled. Most people think of computer programs you can talk to as gimmickry: programs that can be fun, but which don’t actually function well enough to provide value.

Ai-Applied doesn’t agree. The company wants to revolutionise the customer service experience on websites. Instead of hunting around menus and clicking through layers of links, Ai-Applied wants you to find the information you need on a website by having a conversation with the website. Tindal says that there is no reason that your experience on a website can’t feel personal, as if you are walking into a small boutique shop. You should be able to ask for what you are looking for, and the website should be smart enough to understand you and answer you properly.

Of course, that vision has been a holy grail of human-computer interaction for half a century. In the 1968 movie 2001: A Space Odyssey, HAL became the most famous virtual assistant in the entertainment industry. It has been the paragon of the type ever since: the computer that can solve complex problems, help you to do your work, look up obscure data, and be a conversation partner and companion all at the same time.

But almost 50 years later, we still do not have HAL. Instead, we have Siri: eminently mockable, entertaining in its own way, but extremely functionally limited. If you need to find the nearest Chinese restaurant, it’s as good as having a knowledgeable friend at your side… as long as you live in the United States. But venture away from the regions that it knows about, or outside of a very limited set of questions and commands, and it becomes clear that Siri doesn’t understand what you are saying. It is not a customer service agent, nor a personal assistant. It is a cleverly-programmed, voice-activated data look-up service.

If Ai-Applied is going to be taken seriously, it has to convince people that the services that it offers are dramatically different from toys like Siri. It will also have to convince people that its products can be differentiated from the hundreds of chat bots out there on the internet today, those gimmicky and entertaining toys that most people are familiar with. They look like an instant message chat, but within a very short time it becomes clear that a simple program at the other end. They can be fun to play with, but for most people the entertainment comes primarily from saying things deliberately to “trip up” the program and catch it sounding stupid. If YouTube is anything to go by, this is the primary way in which people interact with Siri, too.

Mark Tindal understands the challenge that Ai-Applied faces. Before partnering up with Jim Wyatt to work on the project, Tindal worked in Voice Interactive Response (VIR) systems: the voice-activated menu systems that some telephone carriers use for their customer service lines: “In many ways, we had this challenge when I started working in voice recognition and voice sales services.” he says. “It was a technology that was introduced in the mid 1980s, with some very crass, very inefficient, and frankly terrible voice-recognition technology.

“So when we came back, to try to re-introduce voice technology after it had improved, we still had the challenge of people who remembered the old technology and said, ‘Well, that doesn’t work.’ They could not get over the fact that they had been bitten once already.”

With Ai-Applied, Tindal is likely to find himself in the same situation again. This is why it’s important to him that people understand that Ai-Applied is not offering chat bots, but fully-fledged virtual assistants.

[midk]

At this point, Ai-Applied is not writing its own software. Instead, the company acts as an agency. Tindal and Wyatt take their expertise in the customer support and self-service helpdesk industries, and their collaboration with suppliers of cutting-edge artificial intelligence software, and are able to recommend the right artificial intelligence software to meet a client’s needs.

At least, that is the vision: the company is still very much in its formative stages. For the past year, it has been examining various industries, looking for arenas in which a website with a virtual assistant might be most needed and successful. The most promising opportunities are in industries where customer service usually involves simple tasks being done by people: the gaming industry, the betting industry, “white and brown goods” repairs, and so on.

Ai-Applied’s pitch is based on customer experience, and the way that customer experience translates into conversion rates. Most people who visit company websites looking for information get frustrated  or bored, and give up long before they have filled out a form through which they might become a potential customer. Ai-Applied wants to identify those people, and have the website pop up a chat box for them, asking, “How can I help you?”

The immediate vision of Ai-Applied to is get customer service on the web advanced enough that every website has its own personality, and customer service on a website is like the customer service you get when you visit a small shop. The key to this kind of customer service is interaction. The interaction is personal; a nice font or color on a website isn’t personal.

It’s a brilliant vision, truly the “goal of the ages” when it comes to human-computer interaction. Ai-Applied believes the technology has finally arrived in the academic world; their success as a start-up will depend on their ability to leverage that technology in the world of customer support, convincing the world that virtual agents have, in fact, changed.