Training an AI algorithm – a few thoughts.
My 9-year-old son randomly showed keenness around AI.
My son – “what can AI do?”
Me – “AI can do what it is trained to do.”
Son- “Can AI do better than us?”
Me- “Possibly – mainly depends on how it has been trained and how much training is done.”
Son – “How it can do better than us – when it learns from us?”
Me – “ You learned Slither.io from me, and now you are better.”
Son – “ O – yes, coz- you play with a computer – easy level, and I play online with others, and they are really really good, and I play way more than you.”
(Now he changed the context entirely, and wanted to download another game)
Well…..A good point made here, though. Learning of AI solution depends a lot on the quality of data and process involved in training the algorithm. However, a large dataset does not necessarily ensure quality output. The combination of i) rightly selected dataset and ii) a wise ground-truthing process makes a significant difference. It’s not entirely about tireless training over millions of data-items. If the dataset is not apt, might lead to garbage-in-garbage-out. Right planning around training an algorithm plays a key role.
The ethical aspect must be the overarching principle.