Teresa Escrig

News and oppinion about Cognitive AI & Robotics

Is modern AI intelligent enough?

leave a comment

Reality is suddenly dawning for many who believe that the current AI tools in the market are a silver bullet to solve everything from cancer to your toughest business problems. A recent TechCrunch article citing a study from global investment bank Jefferies, points out that IBM Watson is not providing the promised business value to their clients. How could that be?

It might be because it, like other similar AI tools in the market today, is not so cognitive after all. Simply speaking, Watson technology is comprised of a bag of answers to a bag of questions for a particular subject, and a Machine Learning (ML) mechanism to best correlate answers to questions. Watson neither understands the meaning of the question, or the meaning of the words it uses. And one can only go so far without understanding.

Initially there were bright and promising results, which has brought AI to the second golden age that we are now in. The way ML (Neural Networks – NN and Deep Learning – DL) is used currently is only a brute force solution to problems. Cleaned data is being dumped into a simple mathematical structure (including many additions and comparisons) that reacts to inputs depending on thresholds and weights (parameters) then getting results with a level of accuracy to classify or predict the relationship of inputs with outputs.

Modern AI is being incorrectly equated in the industry to ML, which has caused at least four systemic problems:


  • With small amounts of data there is no model
  • If our data is biased we train biased models
  • Only the companies with access to lots of data can create the best models. It is difficult for startups without access to data to create healthy competition

2. LACK OF TRANSPARENCY. We do not understand why the AI models arrive to decisions and predictions and therefore can’t be trusted


  • When it doesn’t work we don’t know why or how to fix it
  • Difficult to reach high levels of accuracy
  • Uncertain time to train new models

4. THE FOCUS IS ON TECHNOLOGY INSTEAD OF ON SOLVING BUSINESS PROBLEMS. The current AI conversation is more about what can be solved in a company with a chatbot, or Robotic Process Automation, or what can be classified or predicted.  Instead focus should be on the main pain points of the company that will bring about the most transformational value, using AI and not AI technologies, to arrive to the most simple and elegant solution.

In order to resolve systemic problems of a thing, we need to step back and look into the definition of that thing.

Natural versus Artificial Intelligence.

The way we use intelligence is by first perceiving and understanding the world around us (sensor interpretation or perception). We perceive the world around us using different senses (i.e. vision and sound), then automatically integrate that information which further enriches our understanding of the world (we only become aware of it if there is a mismatch in our senses, i.e. if what we see doesn’t correspond to what we hear). Once we perceive the world, we reason with it, inferring new knowledge from our perception (i.e. this street is a shortcut to my destination). When more reasoning is done (what we call “thinking”), we have a better understanding of that aspect of our reality. Once things become repetitive, we progress to the learning phase – we don’t think about them anymore, they become automatic. The decision-making process can come after reasoning or after learning. When we have a view of things over time, wisdom appears. Consciousness is increased when we are more aware of the world around us. If we substitute senses (in people) by sensors (in machines) as input to the Pillar of Intelligence (see figure above), we have a more holistic definition of AI. Improving our definition will allow us to solve systemic problems.

This holistic view of AI then gives us the capability to provide knowledge and wisdom by abstracting data and information (seePyramid of Knowledge below).

Currently, in the ML approach, data is dumped into the learning phase, skipping the prior layers of intelligence (perception and reasoning), and therefore without understanding. It is like doing things because they work (sometimes) but we don’t know why we do them, their cause, or the consequences. These actions wouldn’t increase our levels of consciousness or awareness of the world. But that is exactly what we do with AI.

I believe that the solution to the 4 systemic problems of modern AI resides in the integration of technologies. For sensor interpretation and integration, this means using Knowledge Representation technologies, including Qualitative Modeling, Ontologies, Web Semantics, and Edge Computing. And for reasoning, using inference tools, including Qualitative Reasoning, and Decision Trees.

The only way to get transparency is by going back to the basics of understanding the problem, the words used, and the operations to automate. “Knowledge is power” can be true thanks to AI. The transformation of data/information into knowledge/wisdom that holistic AI provides will allow real transformation in corporations.

Let’s focus on defining the AI journey of our companies. Solving the main current pain points with the best combination of technology will progressively transform companies to become more efficient. And let’s do it in a responsible way: paying attention to the type of solutions that we develop – only to enhance human capabilities; and making sure that we find better professional alternatives for those whose job will be displaced by AI (that finally will be all of us).

Written by Teresa Escrig

May 24th, 2018 at 3:08 pm

Leave a Reply