AiThority Interview with Jordi Torras, CEO and Founder, Inbenta

What is your experience about the journey into Artificial Intelligence? What made you found Inbenta?

I majored in Computer Science and specialized in Artificial Intelligence. Back then, we were always talking about things like Expert Systems, Genetic Algorithms, Natural Language Processing, LISP, PROLOG and even Neural Networks. Not fully disclosing here about how long ago was that, but my final project had run in an Intel 8086 computer! I’ve always been fascinated by AI, building software products and creating a viable business out of that. In that sense, Inbenta is the fulfillment of a long-held dream.

Read More: AiThority Interview Series with Jeff Epstein, VP of Product at Comm100

What are your customer-focused AI products? How effective is Artificial Intelligence in providing great customer service?

AI and Machine Learning have proven to be exceptionally helpful at automating some tasks by computing power that would be difficult to program otherwise. For example, building a software program that is able to recognize someone’s face in a picture would be extraordinarily tedious and difficult. Machine Learning is able to make computers build “programs” themselves based on examples, instead of having to put rules and coding in them.

When it comes to excellent customer service, it is crucial for any AI-based product to understand users in their own languages and give relevant answers and solutions to their questions. This falls in the realm of Natural Language Understanding, probably the most difficult area of AI.

Building efficient conversational agents (or chatbots) is a great option for customer support, but making a chatbot that truly processes natural language that uses unique Machine Learning solutions would require a prohibitively immense amount of training data.  Instead, using semantics, lexical functions, and analytical linguistics is necessary to obtain optimal results in a reasonably short time.

How do you differentiate Inbenta from other service providers in the same landscape?

Many of our competitors essentially use some sort of brute-force Machine Learning algorithm to implement NPL that requires large amount of training data. Most vendors call this data as “training utterances”, so that it forces a company to collect all these utterances somewhere. The problem arises when the companies often do not really have those, so they have to start either creating them manually or leave their chatbots undertrained, and therefore experience poor performance.

Inbenta uses a complex semantic-linguistic-statistic engine that combines Machine Learning with a large Lexicon that encompasses deep knowledge of the language being processed, as well as the world we live in, making most or all of the training utterances unnecessary. Thanks to this, deploying a new and intelligent conversational interface with Inbenta that can happen in a matter of days, instead of months.

Read More: AiThority Interview Series with Stuart Brock, Director at Seal Software

How do you see the trend of including ‘AI’ in everything’ impacting businesses?

The development of new training algorithms for neural networks like Deep Learning has enabled many new Machine Learning applications that were unthinkable just five years back. It has also created some unrealistic expectations on the kind of problems that can actually be solved by Machine Learning – many of them have gone on the market, failed and give the impression to many users that we are living an AI hype.

To me, the most relevant example of Machine Learning is chatbots. 70% of the chatbots that started on Facebook have been removed and the overall experience by final users have been less than satisfactory. Ultimately, Natural Language Understanding is the main problem that machines will be able to solve completely and accurately. From Inbenta’s perspective, we provide a technology that reduces unnecessary and redundant call center interactions providing a consistent ROI and a good user experience, but we never overpromise the features.

What are the biggest challenges and opportunities for enterprises such as yours that deliver on complex technologies?

We have been working with AI and NPL for the last 10 years. However, the “AI hype” began about four years ago as a true tsunami. We had to quickly adapt from the “the weird company that still does AI” to an “another company doing AI”.

We had to adapt to a totally new competitive environment, having to move from answering questions like “what do you do?” to “how are you different?”, as well as simultaneously competing with giants like IBM or Microsoft and literally hundreds of small startups with no customers but serious funding.  As a technology startup growing and scaling its business, Inbenta already benefits from excellent opportunities, like a large of customer base and, in our case, having developed completed Lexicons in more than 30 languages – something that our competitors are far from having.

Read More: AiThority Interview Series with Dean Abbott, Co-Founder And Chief Data Scientist at SmarterHQ

What is a simpler definition for our readers about ‘Meaning-Text’ theory?

Think about the following two sentences:

“can I drive your car?”

“can I play your guitar?”

They are totally common and easy to understand. Now consider the following ones:

“can I play your car?”

“can I drive your guitar?”

They are syntactically correct, but they sound weird therefore making them difficult to understand. There is something about “play guitar” and “drive-car” that sounds familiar and to some extent, need to come together.  We could say that the most archetypical usage of a guitar is to “play it” and for a car is to “drive it”. To some extent, we could think of “the most archetypical usage of something” as “operate”. A few examples would be:

“Operate” a “guitar” is “to play it”

“Operate” a “car” is “to drive it”

“Operate” a “motorbike” is “to ride it”

“Operate” a “question” is “to ask it” [also “answer it”]

“Operate” a “favor” is “to do it” [also “ask for it”]

In Meaning-Text theory this an example of “lexical function”. In particular, a lexical function called “Oper” so the previous examples would be denoted as:

Oper(guitar)= play, Oper(car)=drive, Oper(motorbike)=ride, Oper(question)={ask,answer},Oper(favor)={do,ask}

According to MTT, there are around 60 different types of lexical functions, including synonyms and many others. These 60 lexical functions exist in every known human language and the basis of modeling for natural language processing.

How does Inbenta leverage on Semantic Clustering, Gap Analysis & Languages?

Semantic Clustering is putting together user questions that basically mean the same, although might use totally different words. It is a powerful tool to detect what new content should be created and what content should decrease the space, or gap, between the content we have and the content users need – we call this Gap Analysis. The ability to do this in different languages is what allows your company to scale internationally and reach global markets.

Read More: AiThority Interview Series With Scot Marcotte, Chief Technology Officer at Buck

How do you seek information on Artificial Intelligence related topics?

I believe there are two different kinds of sources for AI-related news and information: the hyper-ventilated, industry-driven, sensationalist press on one hand, and the technological and scientific articles on achievements and under-achievements of Machine Learning in different areas.

In reality, I try to keep an eye on both, as eventually all news and publications will have an impact on our business. Often I have to answer questions to customers, partners and investors, so it is important to understand the exact context of these questions.

Where do you see AI/Natural Language Processing and other smart technologies heading beyond 2025?

On one hand, we live in an exponentially accelerated technical revolution, where progress in technology reaches unprecedented speeds. On the other hand, we have seen cataclysmic failures, like predicting when we will see computers really interacting like humans. Just look at HAL9000, the “2001: A Space Odyssey” a movie shot 1968 that fairly represents what were the predictions back in time. If you double check with reality, the most intelligent virtual agent we had in 2001 was Microsoft’s “Clippit” – a software that was described as a “tragic misunderstanding of human nature” and widely reviled among the users as intrusive and annoying. That’s why I tend not to get too adventurous when it comes to predictions!

Read More: AiThority Interview Series with Dr. Massimiliano Versace, CEO and Co-Founder at Neurala

What AI-Focused start-ups and labs are you keenly following?

I believe that TensorFlow initiative is an excellent resource for following the practical progress on Machine Learning, Deep Learning, and applications for Neural Networks. I also follow several NPL forums and I am an avid reader of all topics related to the dichotomy of using statistic versus analytical tools when it comes to human language understanding.

What technologies within AI and computing are you interested in?

I am particularly interested in “Machine Perception”, or “the capability of a computer system to interpret data in a manner that is similar to the way humans use their senses to relate to the world around them”. That encompasses several fields, including machine vision, machine hearing, machine touch. Applications like self-driving cars and robotics need a gigantic push in these areas before they reach any practical applicability – I find this field fascinating.

What is a piece of professional advice that changed your life?

All entrepreneurs have to face challenges and uncertainty. The very decision of starting your own business entails risking your time, money and in certain cases, family and health. The best advice I have received in my life all gravitate around managing emotions. I summarize all those with a cinematographic quote: “danger is real, fear is a choice”.

Read More: AiThority Interview Series with Kobi Marenko, CEO and Co-Founder at Arbe

Tag the one person in the industry whose answers to these questions you would love to read.

I’d love to hear how Igor Mel’čuk answers all these questions!

Thank you, Jordi ! That was fun and hope to see you back on AiThority soon.