Artificial Intelligence Machine Learning & Society
Artificial Intelligence Machine Learning & Society Throughout human history, there has always been a dynamic evolution between technology, relations of production and ethics.
When a new technology makes a breakthrough in a lab, the value generated will in turn impact the commodity form, production relations and so on.
After society accepts this new technology, corresponding demands in path, content and even speed of technological development arise in the process of spontaneous adjustment of business models and institutions conceive new ethics and regulations. In this interaction, technological systems and social systems will evolve in resonance, a process called technological revolution.
In the past decade, computational power and data scale have reached a tipping point.
As a result, deep learning has made a breakthrough in feature representation.
Deep learning has made great strides in artificial intelligence’s development, and achieved incredible results in countless application scenarios.
For example: facial recognition accuracy rate has exceeded 97%; the correct answer rate of Google’s intelligent voice assistant reached 92.9% in the 2019 test. In these typical scenarios, AI performance has exceeded that of ordinary people, let alone experts and thus accelerated technology upgrading.
In the past few years, AI has made rapid technological and commercial breakthroughs in areas where business logic is technologically friendly.
Moreover, where ethical regulations are temporarily scarce (e.g., security, real-time scheduling, process optimization, competitive game, and information flow management).
As AI becomes more advanced, humans work on comprehending and retracing how the algorithm came to a result. And since understanding is a prerequisite for trust, explainability is of growing importance for both users and developers.
For example, in the financial risk control scenario, a deep learning model identifies a few users suspected of fraud, but the business department dare not directly use the result, because they can barely understand the results’ origin and judge the accuracy of the results.
Meanwhile, because of the lack of clear basis, they cannot explain to regulators as well.
Another example comes from the medical field.
In this example, a deep learning model determines that a patient has tuberculosis according to the patient’s test data. However, because of the lack of a diagnostic basis, doctors dare not directly adopt it. Instead, they examine the relevant test data based on their own experience and then make their own judgment. From these two examples, we can see that without explainability, the application and promotion of AI models in practical scenarios are less credible. When putting AI models into production, explainable AI helps organizations build trust and ensure the system works as expected.
When the explainable AI is mature, its implementation can truly characterize model accuracy, fairness, transparency and outcomes in AI-powered decision-making.
Explainable AI will eventually make AI more valuable in the business environment and in everyday life.
However, no matter how great machines can do with their computational power and algorithmic rigor, people’s intuition remains unmatched by any AI.
Therefore, I believe that the collaboration utilizing the strengths from both sides will lead to better solutions and innovations and finally enhance life quality.
