There is a simple answer to the question that constitutes the title of this blog. They don’t want to. It is not commercially viable for the likes of Google or facebook to create CIs in the same way that it’s not commercially viable for Macdonalds to sell twice-baked cheese souffles. It’s complicated, it’s time consuming and without someone who knows what they’re doing it’s almost certainly going to flop.
Let’s delve deeper into this. An effective, commercial-level conversational AI must be sector-specific, so each one is tailored to the needs of the client. Google and Microsoft are not in the game of creating bespoke AI packages for businesses and there is currently no such thing as an all-knowing or general CI. For example, in order for a banking CI to be able to advise customers on specific products and services, the necessary training data must be in place and the AI equipped with the ability to respond correctly to specific banking related requests.
What the big players offer to the world of CIs
If you want to create a voice or text conversational interface (or chatbot, or virtual assistant or other synonyms) for your business, there’s a mountain of online software that will allow you to do just that. This software is courtesy of the likes of Google, Amazon, IBM, Oracle, and even facebook. A word of caution though. All of these tools, and almost all the others on the market have been designed with one key assumption; that developers with no expertise in computational linguistics and automatic speech recognition can create commercial strength CI services that can automate large numbers of customer queries at once. There are even a bunch of no-code tools which make bold claims such as these.
The assumption that these tools are a simple flatpack solution to a complex problem has created a lot of confusion around the creation of CIs. Yes, it’s possible to create a chatbot from the tools proffered above, but only if that chatbot is supporting a simple use case. Your customers must also be comfortable using rules-based interfaces that make use of multiple-choice buttons or keywords to navigate around. If you’re lucky you may be able to support an initial natural language request via a text or speech channel. Simple chatbots such as these can provide an ROI by tackling basic requests, but they are very limited.
The bottom line is that it’s possible to create conversational interface prototypes for banking and healthcare, but when these are extended for commercial use across a broad spectrum of user queries, the systems fail. This is entirely due to their inability to deal with the complexity of natural language (and speech if it’s a voicebot). The natural response to this kind of failure is to build a solution in-house, but this is simply impracticable. The skills required for this task are so rare that you already need to be an expert to know what you’re looking for in the recruitment process. If you do manage to bag an expert in the field, you’ll find them eye-wateringly expensive, and they will require years of research with a sizable team to build anything close to a commercial model. We know this through our experience of developing internal tools for use with client applications.
Conversational interfaces such as the ones we power are designed for commercial use. They are not rules-based. They interact with customers using natural conversation either by voice or text. They can significantly reduce call centre loads, resolve complex requests, and leave customers happy as a by-product of their experience.
What working with us looks like
action.ai creates your language ontologies, the classifiers, labels the training data, and maintains this going forward for as long as is needed. We can even create the training data if that’s useful – for speech or text. The difference in quality and levels of automation is significant, and the more support offered, the bigger the difference. But in order to create a virtual assistant that automates customer services scalably across a significant range of queries, you’ll need a team of deep learning experts with highly specialist skills in computational linguistics. You will also need senior software engineers with expertise in building virtual assistants and specialist UX designers who understand how to shape and support a conversational model. And if you want to support speech by offering automated services over the telephone, you will need additional and rare skills in automatic speech recognition, speech synthesis, and telecommunications infrastructure.
Clients don’t want such a team in their company so they come to action.ai. With action.ai classifiers, you don’t need to worry about the ambiguity of language, prosody signals on the end of speech, and how people naturally repair conversation within a single turn or across turns. Essentially, we give you a box that does all of that for you.
action.ai is not a consultancy. We do believe in empowering the client but we recognise our clients do not want to build large teams of specialists to crack this problem and then maintain solutions going forward – especially when different applications often have very unique requirements. We’re continually considering the development of tools to empower the client with control but being fully self-sufficient is simply not possible without deep expertise in computational linguistics and automatic speech recognition. So we provide a solution where that expertise is unneeded by providing you with ready-to-go classifiers, either on-premise or hosted on our servers. The client also gets dialogue manager source code that doesn’t require any expertise in any of the areas mentioned. And we allow you to glue in your proprietary systems and call our pipeline of classifiers. Our pipeline of machine learning systems performs over a billion tasks when a user types or speaks to our bots.
In conclusion…
So it’s actually a moot point to ask why the likes of Google don’t offer the types of solutions offered by action.ai. The Google agenda is to empower businesses with tools to create and maintain their own solutions using their tools and servers. Because Google and the like are seeking to make the client sufficient without expertise, the tools they provide are uncomplicated. But straightforward tools result in limited applications that hold little interest for users. That’s why we use our internal expertise and internal tools to shape a solution bespoke for your business and respective customers. If we tried to create sophisticated bots using the likes of Dialogflow, we would need to create supplementary layers of technology. Even then, the end product is likely to be too slow, impossible to maintain, and inaccurate.
We are not in the same game as the Googles, Microsofts, and Amazons. We don’t require banks or healthcare organisations to become experts in machine learning, AI, or language processing. We do the hard work and create bespoke classifiers for a specific company’s needs. Once up and running, we set them up to maintain their own service (if required) and they always have complete control over their own data.
