An artificial intelligence system that can do difficult tasks such as reading, understanding and learning, has for decades been one of artificial intelligence’s biggest challenges.
Reading, understanding and learning have long been the biggest challenges that an artificial intelligence system must overcome. Not surprisingly, when the artificial intelligence model developed by Alibaba surpassed humans in a reading comprehension test last month, several media outlets viewed the event as a radical achievement in the world of artificial intelligence.
However, as AI experts pointed out, the reality is different. The test itself cannot be compared with how a human would read. First, answers that are been given during the test were not generated from understanding the text but from machine learning focusing on the recognition of patterns in short passages. Secondly, the test was performed only on Wikipedia articles that follow a structured and mostly uniform format. Books, news and articles written by people are usually formatted quite differently; be correctly read and understood requires cognitive capabilities. This is different from the capability that today’s artificial intelligence systems employ to discover and identify patterns in text.
AI technology has evolved, but there is still a long way to go (and pure machine learning is not the solution)
According to Gartner’s Top 10 Strategic Technology Trends for 2018, the evolution of intelligent things is one of 10 strategic trends with broad industry impact and significant potential for disruption. In their top trend, the “AI foundation,” they predict that “The ability to use AI to enhance decision making, reinvent business models and ecosystems, and remake the customer experience will drive the payoff for digital initiatives through 2025.”
We can only agree. However, as we pointed out in our post about with AI predictions (Artificial Intelligence in 2018) it is also important to acquire a correct understanding of what Artificial Intelligence is and what it is not. First of all, AI is not machine learning. Even if machine learning can be useful for some tasks, automatic learning is not real.
Let’s look at an example. While perception tasks like image recognition are more applicable for an intelligent system based on machine learning, the situation is different if we consider cognitive tasks like reading and comprehension. Such cognitive tasks require more than a pure machine learning approach because there is greater ambiguity that must be overcome to comprehend language. More complex capabilities are needed to resolve ambiguity for a word that can express different meanings in context.
The idea of an AI world where an artificially intelligent system can magically perform any cognitive task that a human can perform is speculative at best. The road to a real artificially intelligent system based on the core aspects of AI, such as the semantic analysis of text to comprehend the deeper meaning of words and sentences, is long; human intelligence will remain central to such a system.
The Benefits of an Artificial Intelligence System for Business
However, even if human intelligence is and will always be central to a true AI system, there are several examples of artificial intelligence systems that can already resolve many tasks. Here is where organizations should focus, by prioritizing business results enabled by solutions that exploit different AI technologies such as the combination of semantic technology, natural language processing and machine learning.
Want to learn more about the advantages of combining semantic technology and machine learning? Visit our page focusing on our Cogito cognitive technology.
