There’s been a lot of hype lately behind Artificial Intelligence (AI) and Cognitive Computing, both terms that generally refer to programs or computing systems that “think” rather than simply perform computations on command. The move to impart cognition into computers has captured public perception, but it’s common to overstate and distract from the way that machine learning and deep learning are actually used.
The concept of cognitive computing is an important distinction for several reasons:
- The primary reason for implementing these kinds of solutions is efficiency, applying the processing power of computers to the kinds of complex problems that are usually left to humans.
- These advanced computational processes require more computing. This takes computing resources and time.
- Intelligent computers must be taught to perform each task.
The promise of artificial intelligence is that one day computers will interface with people much more closely than how people and computers communicate today, even having the computer act on its own. This would bring automation to more complex tasks that require real decision making, not your classic assembly line robots.
Already there are some examples where artificial intelligence shows clear advantages over human users. This is particularly true where volume becomes an issue, like sorting through hundreds of thousands of images or videos. We are seeing today how today’s artificial intelligence is showing real, practical applications in helping to prevent traffic accidents.
Outside of these very structured environments and months-long learning curves, even the most advanced applications will struggle. Cut through the hyperbole of curing cancer and discovering needles in haystacks, and you’ll find that most data successes are still very human efforts. Talented analysts use a combination of technical skill and subject matter expertise to turn raw data into a solution.
In the world of analytics and data science we see these struggles in the form of extensive hardware requirements, costly services engagements, and all too often companies are still looking for a solution to the same problem six months to a year later. They’re struggling to connect the business knowledge with a repeatable, data-guided solutions.
Instead of building smarter computers to bring miraculous solutions, most of these problems are better solved by smarter application of computing. For example, trying to predict the optimal time to trigger a specific stock purchase could be accomplished with a relatively simple model, but imagine a computer trying to determine which stock to purchase. The problem involves a dramatic increase in complexity because the increased number of variables to consider is so much higher. That kind of complexity requires new approaches, and the dynamic nature of the decisions that have to be made doesn’t permit months of research to reach an acceptable level of accuracy.
There’s a long list of complexity problems that people are trying to solve with AI.
What examples would you add to the list?