Back to blog

Virtual Assistants are Search Engines

Exploring the interplay between search engines and personal assistants, and our experiences building them.

Virtual assistantsSearch

I've spent several years tinkering and exploring the world of virtual assistants, and only over the past few weeks have I realized that they have a unique and perhaps symbiotic relationship with search engine technology. Designing a great search engine is, in my opinion, required in order to get virtual assistants right. That is why, as of today, Operand is shifting its focus as a company (temporarily) away from our consumer virtual assistant, towards our experimental search engine systems. All existing subscriptions will remain active, yet we will pause billing and will not be charging customers for as long as our focus is away from the assistant. All new signups will be disabled as well, since we as a team don't have the bandwidth for on-boarding. Our time will be focused on experimentation and building out some new core systems, which will eventually form the basis of Operand v2. This next assistant, which should be ready by fall, will be entirely re-built on top of a new search-engine focused platform.

Interestingly enough, we've been developing search engine technology indirectly over the past few months while we went about developing foundational features of our virtual assistant. It all started with the feature that neither Xander or I wanted to build -- weather. Maybe this is something to do with the dominance of Alexa or Siri, but the majority of users almost immediately asked for the weather when they first were on-boarded to the assistant. We're still not sure if it was just a useful feature, or if it was some way for users to compare our assistant to other assistants, but we decided it was probably in our best interest to build weather, and to built it in such a way which showcases what we're good at. Notably, we wanted to make sure that users could ask arbitrary questions and still get useful (and correct) answers. For example, "how hot is it and should I wear sunscreen" is considered a valid question and should be answered appropriately.

The next major informative moment, or tipping point if you will, was when we built and released our long-term memory feature. This in a way was the first iteration of our search engine technology, but rather than searching the open internet, we searched your history with our virtual assistant. If you told Operand that your favorite color was blue, or that you took your coffee black, it would be able to regurgitate these answers back to you during a conversation. Obviously knowing your favorite color isn't going to make Operand this "next-generation personal assistant" we've been talking about, but knowing your preferences and being able to take action according to them might. At the moment, we don't really take advantage of this to its fullest extent, but eventually, I'm imagining a world where rather than querying a database for configuration, we could query the customers message history with the assistant, or even some other sources of information such as their financials or emails. At first, this will likely be used for boolean toggle type features, such as calendar event reminders, but can be expanded to encompass a wide variety of configuration values.

I think long-term memory is one of many examples of the fascinating interplay between virtual assistants and search-engines. The virtual assistant provides data to the search engine, and the search engine uses the data to better inform the virtual assistant and can control its behavior.

We've spent the last number of weeks, ever since we released our long-term memory systems, working on and experimenting with search engines, and so far, we've had some interesting results. For example, we've developed a system which integrated with some of our conversational systems to enable users to have a full-on conversation with a website, or even a textbook. Aside from some of the business applications of such a system (which yes, we are exploring), there are some fascinating applications in the context of a virtual assistant. A friend of mine suggested indexing the entirety of Wikipedia, and allowing users to have a conversation with it, which essentially would be like having a conversation with someone who has up-to-date intricate knowledge on almost every possible topic. The point here is that there is a unique relationship between virtual assistants and search engines, and we as a company need to take some time to explore further.

In order to focus our time and energy on this experimental technology, we need to take a step back from some of our existing commitments, namely supporting and on-boarding users to our current version virtual assistant. Our existing users are primarily technologists, and use the product mainly to experiment and try out new features and experiences on a weekly basis. It doesn't feel fair for us to continue charging these users when our full attention isn't towards supporting them and delivering on new features and improvements. Therefore, effective today, we will be disabling new signups and pausing billing for all existing users. If you're an existing users who is reading this, feel free to continue using Operand as you are today.

During this time, our primary focus will be experimentation and technical exploration in and around the search engine space, specifically focused on their relationships to virtual assistants. These efforts will be directed towards the second-generation of the Operand assistant, which all existing users will get as part of their monthly subscription. This second iteration will be based on an entirely new platform in almost all areas, specifically our parsers and contextual understanding systems. Over time, we will release more information about this assistant, and existing users will be able to demo and test these new systems as they are released, in the same way that new features are currently released. Expect a full release of these new systems, in addition to the re-opening of signups this upcoming fall.

To conclude, I want to extend a thank you to all of our supporters, users, investors, and to anyone who has been part of our journey so far. I'm incredibly privileged to have the freedom to work on these sorts of problems, and couldn't be happier doing so.

~ Morgan