Artificial Intelligence is indeed our future, but to what extent will our world drastically change in the more immediate term? Here are a few machine learning myths and the reality behind them:
Expectation: Self-driving cars will outnumber traditional cars in just a few years.
Reality: Automated safety features like backup cameras, assisted braking, adaptive cruise control and assisted parallel parking will become more prevalent, but fully automated vehicles may have more hurdles to leap. Legal questions regarding fault in accidents (driver or manufacturer), along with technological issues such as latency, environment complexity and the lack of advancement of machine learning still stand in the way and probably will for at least another decade.
Expectation: Bots will take my job away.
Reality: Bots will handle menial tasks, opening up opportunities for humans to work on more complex and meaningful work. Research has shown that while job displacement could be as high as 4% of the U.S. workforce by 2024, job creation due to AI as a result of new industries that didn’t exist before could offset some or all of the displacement. Bots are good at mundane, routine, rule-based work. Machine learning algorithms can elevate people by doing what computers do well and allow people to do what they do well.
Expectation: My voice assistant will know all my personal information and share it with the large tech companies.
Reality: Voice assistants such as Google Home, Alexa and Apple HomePod are listening all the time to trigger on the keywords. Then, the device transcribes the text to execute the commands. Each of these companies collects the transcribed data to improve their product’s speech algorithm. People listen to small chunks of text to improve it. This is how the companies can improve problems associated with region or accent. For machine learning to improve, it needs high quality, high quantity data. So, it’s true your data is collected and shared. However, it is anonymized and parceled out for data improvement.
This leads to many questions regarding privacy– what do I as a user have control over? Is it more private if a human doesn’t listen but a computer does? If that person can access only small portions of my data and it is anonymized, am I ok with that?
Expectation: A computer will deny me a future loan.
Reality: Mortgage lenders and other financial brokers are using machine learning algorithms to assist with decision making. Given the large number of data points used for recommending loans, this process is ideal for a computer. The model can make recommendations for approval or denial based on historical data and your data. Because of this, people are concerned about the explainability of computer decisions. The European Union General Data Protection Regulation (GDPR) sets the standard with the right to an explanation.
What this means, in reality, is a change in the way loan officers work. Instead of processing tons of approvals and occasional denials, they will only be re-evaluating denials. It reduces drudgery work, speeds up the loan approval process and optimizes human expertise. Computer models become decision support systems instead of actual deciders.
Because the application of machine learning is relatively new and expanding at fast rates, there are many opportunities and many risks. While some of the more extreme fears are more SciFi than reality, there are still legal and privacy risks that should be addressed. Embrace the future, and work to identify and fix potential problems.