Until now, authorities describe machine learning algorithms based on the learning style which are supervised or unsupervised. Rob Walker brings a new perspective about the subject. He categorized supervised learning algorithms as opaque and transparent. Actually, there is nothing new under the sun! These terms are based on explaining how to reach the decision. However, I have to hand it to Rob, new definitions would be very creative.
Even though Opaque AI algorithms such as neural networks or genetic algorithms are so powerful, explaining how to reach the decision is almost impossible. They are almost black boxes!
In contrast, decisions produced by Transparent AI algorithms such as decision tree or random forest can be explained easily by human.
The question is that do we really need to explain how? If the subject is requiring legal regulations, then you might need to explain how the decision is taken. However, result is mostly considered more important than the means to an end. All is fair in love and war, isn’t it?
For instance, the following antenna is developed by NASA. Its shape is found by a Opaque AI algorithm which is genetic algorithm. Even though nobody can describe why, it creates the best radiation pattern. The result is that this antenna works better than any antenna and no matter how it is designed.
On the other hand, trusting an opeque ai algorithm requires blind confidence. You might remember the Microsoft’s conversation bot, Tay. It is based on deep learning system. But, it has lived only 16 hours and killed by Microsoft because the bot becomes a racist and tweets genocide supporting sentences. Actually, Microsoft is not guilty about Tay. People Tay talked to cause her to be racist. Because the more you talk, the smarter Tay gets but people prefer to abuse it.
So, I would like to translate a famous Turkish idiom. You should plan to overwinter, if you live summer it would be your luck. Defansive behaviour proposes not to trust black boxes. Similarly, Rob Walker launchs t-switch in pegasystems for defensive ones in same keynote. T stands for transparent meanwhile t stands for trust. All opaque ai algorithms is disabled when t-switch is activated. On the other hand, opaque ai algorithms tend to be more powerful than transparent ai algorithms. To sum up, it’s all your choice! Would you select trust or power?