By now we know that Artificial intelligence is the future. Many businesses are investing heavily into AI technology and scientists are making a lot of progress. However, how do we measure that progress? Scientists from the University of Waterloo conducted a study, which revealed that there is no exact method for deciding whether a given problem may be successfully solved by machine learning tools.
There are people who are scared of Artificial intelligence. It is a tool that can do bad, but can also do a lot of good. And scientists are talking about new advancements in this technology all the time. However, effectiveness and capabilities of modern AI are extremely difficult to measure. In fact, when we are assigning tasks to various algorithms, we pretty much cannot predict if they are too difficult. This is quite a worrying concept and scientists say that we should proceed with caution.
If you need to move a heavy cargo, you have to select an appropriate truck. You look at their power, engine specifications, size of the cargo compartment and you decide accurately whether the machine is fit for the job.
There is no such measuring system for AI algorithms.
Furthermore, scientists say that there cannot be one either – if the task is more complex and more general than just a simple “yes or no” situation we can’t distinguish learnable from un-learnable tasks.
Scientists used a learning model called estimating the maximum to capture many common machine learning tasks. They presented AI systems with various tasks, such as identifying the best place to locate a set of distribution facilities to optimize their accessibility for future expected consumers. Analysis showed that it is impossible to predict whether an AI-based tool could handle that task just by using mathematical methods.
This is quite interesting, because scientists assume that they can measure and research virtually everything. This analysis revealed that it may not be possible 100 % of the time. Shai Ben-David, lead author of the study, said: “This finding comes as a surprise to the research community since it has long been believed that once a precise description of a task is provided, it can then be determined whether machine learning algorithms will be able to learn and carry out that task”.
AI systems are still going to be used extensively in the future. In fact, AI technology crossed the border of scientific activity and is already a commercially available tool. It will be interesting to see how scientists will solve the problem of not being able to predict whether a task is too difficult for the algorithm.
Source: University of Waterloo