The victory of Google’s AlphaGo over the top-ranked Go players in the world last May is another Artificial Intelligence (AI) advancement that everyone is talking about. Besides playing chess, AI is already working on some other areas such as voice assistance (Apple Siri, Amazon Alexa), facial recognition, and customer helpdesk. It is also the driving force for the autonomous car also. AI increasingly permeates society; will it replace most human jobs in future?
It is strongly possible, but there are three fundamental elements in which AI must advance before that can happen:
Have the ability for autonomous learning – The key for AI to tackle a specified task is learning from the previous data. For Example, if you want an AI program to distinguish between male and female, you can feed millions of photos of males and females for the AI system to learn. If you want to create AI customer service, you can input customer support conversations in the past few years for the AI system to learn about. The intelligence of the AI system depends on the dataset you provide. One advantage that AlphaGo has is that “Go” an ancient Chinese strategy game has a very clear result of win or loss. AlphaGo can run reinforcement training by playing millions of games with itself. From these millions of games just within AlphaGo, it learns or creates some steps that come from human experience. But the reinforcement training for customer service AI or facial recognition AI are much more complex, because the results are not dichotomous, in the end, the AI system does not know if it is right or wrong.
Be able to analyze, perceive, and be capable of discovering and defining rules –AI does not require humans to define rules or enter algorithms to calculate the result, it should be capable of discovering and defining the rules through data mining or self-directed learning. If the customer service system is just trying to match a question with the pre-defined 500 Q&A in the database and give out the corresponding answer, it’s not an AI system. Or if people define some rules for the facial recognition system to distinguish male and female such as the distance between eyes and nose etc., that is not an AI system. For a system to be truly AI, it must perform self-directed learning from the big data sets provided to distinguish between male or female. Just like humans, there is no rule of how to distinguish male and female, you learn through experience.
Have strong computing power – Instead of using a CPU for processing, most AI programs use a GPU (Graphics Processing Units). GPUs are highly-parallel, meaning they have the ability to compute or execute many processes simultaneously. They can have multiple processors that allow making high-performance parallel arithmetic calculations. Hence, they can have much more calculation capability than a general CPU at a lower cost. However, there is a current shortage of GPUs as demand is outweighing supply and OEMs are having a hard time keeping stock.
Judging from industry information, AI is still in its early stages and it will be some time before computers can understand and perform tasks to a human level. At least until the foreseeable future, AI will remain a human tool to improve the efficiency of tasks but not fully take over jobs yet. However, when AI reaches a threshold of advancement, it will inevitably replace certain jobs due to its efficiency, accuracy and cheaper cost.