The Limiting Factor of Voice and Dictation Adoption

Thank you Stephen. There is definitely an education involved when learning a new system. I think back to my days using Win 3.1 and the transition to Windows 95, whose introduction made computing accessible to many more people given its more intuitive interface. Voice will be similar.

If you've used Google's product, as compared to say Siri, you can see how much better the former has become over time. This will happen with voice as well. Just as we've developed software for computers, apps for mobile phones, so too will we develop skills for voice that will create shortcuts for what we want to get done.

Neural networks will help these skills only improve, as they learn the common inputs, syntax, and errors gleaned from interpreting this data. This reinforcement learning will only make future use better over time. The beginning stages may be slow, but the future will accelerate quite quickly as developers will be able to apply training data from other use cases to the medical field and hence move the needle faster than what people realize.

Thank you Mark. Your point is well taken. But consider how many people speak on the phone in close proximity every day about a wide variety of sensitive information - whether those are financial brokers in a bullpen, healthcare call centers, telemarketers, or sales people - there are a wide number of roles that are performed in an open setting where sensitive information is expressed freely.

There will certainly be new protocols put in place. For example, assistants today are associated with the Echo, Google Home, the mobile phone and the like. But in the future, integrating the service to a VOIP network or hardware headset is not far off. Moreover, specific individual voice recognition software coupled with a user's unique pass code or ID could help a system understand and adjust to the correct individual speaker and their specific profile. In this way the appropriate reinforcement learning is applied, so that the system is able to adjust to the needs of the particular person.

Moreover, there will be skills, which today are Macros, applets, or sapplets for other software programs that in the future will run on voice (vapplets?). This will allow people to use shortcuts to get to what they need and it will allow the deep learning to take the appropriate course when adjusting to the specific use case. This is already happening at small scale and it is only going to grow.

Thanks Keith! I agree that the system as it stands is immature today. The article by Tomasz cites how he uses four additional services to publish a simple blog post. At some point these services will be a part of a common system, just as other operating systems such as Windows or Android evolved over time. For example, rather than layout the formula step by step as you've done here, you'll launch what today would be considered a Macro in Excel or in computer science a very specific set of functions and then fill in the the blanks. Moreover, the assistant could prompt you line by line; or by changing your users settings, you would just rattle off a list of inputs; or in the case of reinforcement learning, once you (and others like you) have used this similar approach, the machine learning will kick in and guess what you're looking for. Over time, on the consistent or high repetition inputs, automation and prediction will get quite good at iterating on what you need accomplished.