In the financial field, the application of artificial intelligence and machine learning is far less than other fields. It can be said that it is hitting the wall everywhere. When data scientists ask about the basis of the machine learning model, we can only see a bunch of complicated algorithms that cannot be understood. Artificial intelligence can't make decisions that make people understand.
Machine learning and artificial intelligence have made tremendous advances in accuracy over the past few years. However, regulated industries (such as banks) are still hesitant and often prioritize the accuracy and efficiency of regulatory compliance and algorithmic interpretation. Some companies even think that this technology is not credible or dangerous.
During the 2008 financial crisis, the banking industry realized that their machine learning algorithms were based on flawed assumptions. As a result, financial system regulators decided to require additional control measures and introduced regulatory requirements for “model risk†management of banks and insurance companies.
Banks must also prove that they understand the models they use, so it is regrettable but understandable that they deliberately limit the complexity of their technology, using a generalized linear model that is simpler and more interpretable than everything else.
If you want to build trust in machine learning, try treating it like a human and asking it the same problem.
To trust the advice provided by AI and machine learning, companies from all industries need to work hard to understand it better. Data scientists and PhDs shouldn't be the only ones who can clearly explain machine learning models, because, as AI theorist Eliezer Yudkowsky puts it: "So far, the biggest danger of artificial intelligence is that people prematurely think they understand this. technology.
Trust requires an artificial approachWhen data scientists are asked how machine learning models make decisions, they tend to use complex mathematical equations to answer, making the layman stunned and wondering how the model can be trusted. Will machine learning decisions be treated in the same way as human decision making, will it be more effective? As Udacity co-founder SebasTIan Thrun once said: "Artificial intelligence is almost a humanities. This is actually an attempt to understand human intelligence and human cognition."
So, don't use complicated mathematical equations to determine how a credit officer makes a decision, but simply ask: "What information on the loan application form is most important to your decision? Or, "What value indicates the level of risk, and you How do you decide to accept or reject certain loan applications?
The same artificial approach can be used to determine how the algorithm makes similar decisions. For example, by using a machine learning technique called characteristic impact, it is possible to determine the balance utility balance, the applicant's income, and the loan purpose are the top three most important information of the loan officer algorithm.
By using the ability called reason codes, one can see the most important factor in the estimation of the details of each loan applicant, and by using a technique called partial dependence, you can see that the algorithm will apply for higher income loans. The risk rating is rated lower.
Value of objectivity, scalability and predictabilityBy analyzing how machines make decisions like humans, humans can better understand artificial intelligence and machine learning. In addition, humans can gain confidence in artificial intelligence and machine learning by recognizing the unique capabilities of technology, including:
Solving the problem of credibility and data outliers: Traditional statistical models often need to assume how data is created, the process behind it, and the credibility of the data. However, machine learning eliminates these restrictive assumptions by using highly flexible algorithms that do not give more credibility than it deserves.
Support for modern computers and massive data sets: Unlike manual processes, machine learning does not assume that the world is full of straight lines. Instead, it automatically adjusts the equation to find the best mode and tests which algorithms and modes are best for independently validating the data (rather than just testing the trained data).
Predicting the future with missing values: Advanced machine learning does not require hours of data cleansing. Instead, it can build a blueprint, optimize the data for a particular algorithm, automatically detect missing values, determine which algorithms don't apply missing values, and find the ones that replace missing values. Good value and use the existence of missing values ​​to predict different results.
Don't doubt AI or machine learning advice, let us better understand them by asking us about the same reasoning problems of humans. Let us recognize the objective capabilities of technology in reducing the anomaly of data anomalies and the ability to provide scalable flexibility for today's massive amounts of data.
Perhaps most importantly, let us recognize the ability of AI and machine learning to better predict future outcomes by leveraging the missing information. Because while technology is so powerful that it requires vigilance and formal regulation, consumers and businesses will only benefit if they can establish a correct level of understanding and trust.
3.1 Home Theater Speaker,Home Cinema System,Bluetooth Home Theatre,Home Theatre
GUANGZHOU SOWANGNY ELECTRONIC CO.,LTD , https://www.jerry-power.com