We talked to Dr Cambria ahead of his presentation at the Big Data and AI Leaders Summit (September 12 & 13, Singapore), to discuss AI and the challenges it can both solve and create.
Dr Erik Cambria is an Assistant Professor at Nanyang Technological University and recipient of many awards, including the IEEE Outstanding Early Career award and the AI’s 10 to Watch award by IEEE Intelligent Systems. He is also the founder of SenticNet, a spinoff offering B2B sentiment analysis services.
Tell us about yourself, your background and what you do
I have been always passionate about reasoning and, in particular, common sense reasoning. That brought me to join the Common Sense Computing Initiative at MIT Media Lab back in 2008, when I was still a PhD student. After ten years, I am still interested in the same thing. But I guess it’s cooler to call it AI these days!
When did your organisation start investigating AI technology?
AI can be a fuzzy term. Today most people tend to identify AI with deep learning. But AI, as a research field, was born more than half a century ago. When talking about AI, I think that old school is still cool. In that sense, we have been carrying out AI research for the past ten years.
How important will AI be for your company moving forward? Do you have a clear strategy yet?
We are a B2B company, so we are AI enablers rather than AI users. We are developing a new kind of AI: a hybrid AI that leverages both old-school AI and recent deep learning techniques. Our approach, in fact, is both top-down and bottom-up: top-down for the fact that it leverages symbolic models such as semantic networks and conceptual dependency representations to encode meaning; bottom-up because we use sub-symbolic methods such as deep neural networks and multiple kernel learning to infer syntactic patterns from data.
Coupling symbolic and sub-symbolic AI is key for stepping forward to real understanding. Relying solely on machine learning, in fact, is simply useful to make a ‘good guess’ based on past experience, because sub-symbolic methods only encode correlation and their decision-making process is merely probabilistic.
How do you think AI is going to change the way businesses operate?
It is improving most businesses by automating most processes that needed human intervention before. However, companies must beware of the limitations of deep learning, and especially of its un-interpretability. Some companies for example are using AI for HR. The deep network gets trained by showing it a lot of good and bad examples, that is, good CVs versus bad CVs. Later, given a new CV, the AI will decide whether this is good or bad.
However, we’ll never know how such a decision was taken because the feature selection process is not fully interpretable. The AI may think that being a man, rather than a woman, or belonging to a specific race is an important feature for classification.
A common fear is that AI will take jobs, whilst a common counter to that argument is that AI will create jobs. What are your thoughts?
We are still light-years away from creating a machine that is as intelligent as the dumbest of the humans. The only things we should worry about are ethics (see example given earlier) and goal prioritization. It is unlikely that an AI will suddenly decide to erase humanity, but it could do so by mistake while trying to achieve some other goal, as well as we do not aim to kill ants when we construct a building, but we still do so.
Some people think AI won’t last, whilst some feel this is the biggest shift we have seen in tech. What do you think?
I am not sure it will last under the name of AI, but it will surely last in terms of cognitive- or human-inspired reasoning. As I mentioned earlier, the term AI has recently been associated with deep learning… and that’s a mistake! Deep learning is only one type of AI, namely sub-symbolic AI. I think the future will be a hybrid AI that leverages both symbolic and sub-symbolic learning.
Because sub-symbolic AI only implements one kind of reasoning: bottom-up, that is, learning from experience. And we do not learn everything like that. For example, we do not know that we should not jump out of a window because we have tried that before.
How can AI support a more personal experience for your customers?
We provide B2B sentiment analysis tools that allow companies to really understand their customers. Our tools can break down a product or service review to a set of aspects, such as appearance or usability, and associate a polarity to each of these. This enables better user profiling and, hence, better user intention mining. We also developed an AI for personality detection, which has been featured in the news.
Such an algorithm leverages deep learning for guessing important personality features of the user, which are later used to interpret user feedback. For example, some concepts could be positive for an introvert but negative for an extrovert. Hence, it is important for the sentiment analysis tool to categorize the user first according to his/her personality type.
What do you think will be the key takeaway from your presentation at the upcoming Big Data & AI Leaders Summit?
I will discuss about the limitations of current AI tools and explain how to overcome these by giving specific examples in the context of sentiment analysis.
Join the Big Data & AI Leaders Summit to experience two days of compelling data case studies, new best practices, emerging technologies and practical tips to brush up on your technical skills.
Featured Image via Freepik
This content was originally published here.
About the Author
