Editor’s note: This commentary is a response to, “Legislation seeks to outsmart ‘smart’ speakers,” May 23, 2019.
Voice recognition systems have streamlined our daily activities, from turning on the lights to important reminders about taking medications. We have only scratched the surface of what is possible with voice recognition systems. But Assembly Bill 1395 by Assemblyman Jordan Cunningham, a Republican from San Luis Obispo, could halt this technology’s advancement.
AB 1395 would create barriers to the type of learning that voice recognition technology depends on to unleash its full potential for consumers, and it would impose duplicative security provisions that would do nothing to enhance consumer privacy or device security.
Voice recognition systems rely on machine learning to adapt to users’ speech patterns and vocabulary, and are informed by the way users talk to the device. Our speech is nuanced and dynamic.
Users’ age, whether they are a native or non-native speaker of the language, and whether they have a speech impairment also impacts the voice recognition system’s learning. These variables make training critical to the success of voice recognition systems.
As consumers use devices, companies retain records so real-world data is available to train and improve the technology for the benefit of the consumer. This training has allowed voice recognition systems to become more useful. For example, real world data for training is critical for device learning in order to minimize false wakes, which is when a home speaker is activated unintentionally.
Assemblyman Cunningham voiced concerns over devices being activated unintentionally. But his legislation would exacerbate this issue by restricting a company’s ability to avoid false wakes by limiting a manufacturer’s capacity to use stored voice recordings and transcripts to improve the product features.
A company cannot adequately fix bugs if recordings and transcripts can only be retained for the brief time a device needs the information to complete a requested command.
Under the California Consumer Privacy Act approved last year, Californians can access and delete any voice recordings associated with their devices. This safeguard gives consumers control over their data. AB 1395 would add layers of unnecessary, confusing, and unworkable mandates that would conflict with the Consumer Privacy Act.
Voice recognition devices are designed with layers of security, privacy protections, and controls to minimize and protect the data collected. California is already a leader on device security.
Legislation to enhance consumers’ confidence in the devices they use is something we can all support. But AB 1395 would do nothing to advance that goal.
Instead, it will make “smart” devices not-so-smart and impact the accessibility of these devices for all consumers, as well as conflict with existing privacy laws. California has always successfully championed consumers and technology. Legislators should reject AB 1395 because it does neither.