Demystifying AI by Breaking It Down

For the past ten years, I have worked in a speech interface industry that is flooded with Natural Language Understanding/Processing and Machine Learning technologies on top of Speech Recognition. Since AI became the hottest trend and a star for ambitious entrepreneurs in recent years, I have the impression that many are dazzled by the buzzword “AI”…

It sounds almost magical or teleological and influencers facilitate it by over-exaggerating its capabilities. When I read that “AI is the most disruptive technology since the industrial revolution”, I literally said “really?” to my PC screen. These inflated readings may tempt people to feel questionable and overpromising about AI.

My intention is to write an introductory reading for people who may be too excited about AI, and to introduce a new type of AI you probably are not familiar with.

Time for AI

When we founded our first startup Agilingua in 2006, I remember that I insisted to add “artificial intelligence” to taglines. But my partner Matthias Denecke Ph.D did not agree and said “You know what people think AI is? They think AI is what plays chess and wins sometimes, loses sometimes.” I did not try to argue.

Ten years later we started our second company Qeep, which offers services to specific verticals using the identical (but much updated!) core technologies that Agilingua has offered, and I speak loudly about our proprietary AI engines at every occasion.

Things changed. AI-focused accelerator programs were launched, VCs are looking for startups by typing AI in a search bar, world’s leading companies compete in pouring money into it. The progress in deep neural nets made AI work in the large scale, and it brings this AI boom. It finally works.

ai_mna_q3-16-2

As it gets attention, more people write about it and even more people read about it. Consequently, a reasonable understanding of what AI is has been widely shared; An AI is not a human intelligence or anything alike on a computer system. It is a synthesis of small pieces of software components that can occasionally and partially complement human intelligence when integrated into a well-designed solution.

Breaking it down

Although every respected research group issue a paper on the AI market, their forecasts for the market size and growth vary widely. No surprise. You cannot get a grip on it if you see AI as a single technology. Again, AI is a collective of distinctive pieces of software and each AI engine does a different job, so the field of AI needs to be broken down according to application areas. In today’s market, these areas may include Machine Learning, Deep Learning, Natural Language Processing/Understanding, Image Processing, Speech Recognition and so on.

By doing so, we can understand that “Spotify’s AI suggests you new music you will possibly like” means that “Spotify has an algorithm which is a combination of (music) data analysis and machine learning to make predictions based on the user’s past behavior”.

New AI you want to know

Thus, asking about the size and growth of the AI market is the wrong question. Instead, we need to ask which AI technologies are coming and for which applications, then identify their market. And I think you want to know about a new AI technology called Dialogue Management. It adds semantic intelligence and decision-making capabilities to human-machine interfaces. If Siri had dialogue manager, she could answer your follow-up questions rather than executing each single command, and she could also be engaged in conversation on more complex subjects than weather forecast and restaurants search.

I want you to imagine a smart speech interface that is supported by machine learning (that can learn from past data and user’s actions so the system gets smarter as used), natural language understanding (so that the system can contextually understand what a user means and wants) and dialogue management (that make a system take an optimal action based on all factors that need to be considered). This is what Qeep builds for technical documentation.

Qeep helps an engineer, for example when he tries to repair a machinery, make a diagnosis by asking right questions and identify a fault, guide to repair procedures, answer questions, jump to reference or other relating information… all in a hands-free fashion. On top of that, it keeps logs for later analysis. Please think about PDF files that do not provide the right information at the right moment without disturbing the work, nor have any assist avoiding human errors and oversights. It’s not the future.

We are currently looking for pilot project partners. If you are interested in updating your technical manuals or instructions, please contact me.

Posted in AI

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s