The Virtual AI Assistant Will Not Look Like Us

It is often said that technology will change us but never in the ways we anticipate. Take for instance the internet. Many thought it would enable us to send mail very quickly (email). But nobody assumed that it would be used to connect us socially (Twitter, Facebook) or hold the world’s information (Google, Wikipedia). People had a limited imagination about what the internet would be versus what it became.

However, that last example isn't entirely true. People did think it would hold the world’s information. But they thought it would be like a library. Google & other search engines show us that the world’s information is rather distilled amongst many people. By accessing Google, we access the world’s combined information. In a way, its a glimpse into our collective unconscious.

This brings me to Artificial Intelligence as an assistant, which promises to be one of its biggest applications. With Facebook looking to bring a bot marketplace to the messenger platform and companies building bots on Slack growing like gang busters, this seems to be the hot space. After all it fulfils my personal criteria of a consequential innovation: it uses new technology to do some task that wasn't previously possible with older technology. Machine Learning and AI can better know us and communicate with us than computers were previously capable of doing. Throw in powerful pocket computers in the form of smart phones and it looks like a wrap.

But I have a problem with the input. Its too archaic and feels too inconvenient. We talked to computers in movies like 2001 and Star Trek. Remember that when the original Macintosh came out, nobody had thought of GUIs before (at least as far as story telling is concerned). Looking back at The Next Generation, they seemed more focused on hot keys which were, prior to the Mac, thought of as the next big innovation after typed out commands. AI & Machine Learning seem like they’re on a similar cusp.

I think AI will exist on our phones, but in a messenger or talking format. Its too much effort to take out our phone and talk to it, especially when better alternatives already exist and are already much much cheaper. I’m thinking of the sensors on our phones and smart watches. Your phone being able to pick up your GPS location as you enter into a fast food joint to tell you its not such a good idea or to tell you of the cleanest path to work based on camera shots from similar snowfalls are the better ways to think of AI. As we leave more and more of our life on the Internet, its recorded and query-able. AI should have a better sense of what I want to do prior to doing or even thinking it. This would put it in a league above what humans can do, which would greatly expand people’s willingness to purchase it. Having virtual assistants is a tough sell outside of more business-like situations (Slack / IRC bots) or gimmicks (Facebook Messenger’s Miss Piggy bot). Consumer products will be more demanding and more relevant to the situation I described.

Which brings about an interesting connection to smart watches. So often we think of smart watches as the next smart phone, yet Pebble, Apple, & Google are instead tethering your watch to your phone. This makes it more of an extension to the phone in my mind. The watch is in a better position for certain activities. [1] It also suits itself to quick information. I love my Pebble smart watch for this reason alone. People always ask me “What’s the point when I can check my phone?” but it really is convenient to check my watch quickly for text messages or twitter updates. I almost never check my phone any more. Most messages come from email or Facebook, so I just quickly type out a response on my computer as I’m usually at one anyway.

As for AI, the smart watch seems really relevant to providing quick information it pulls from the phone. Extend this to AI and it seems fairly clear to me: get the AI serving predictions to me here on my watch. Is it 2pm and I haven’t eaten? Let me know of great places nearby. Have I only been eating red meat recently? Recommend I get some vegetables to balance out my diet. [2] These super-recommendation apps look to be at least the first big AI arena. [3]

This also sounds really similar to Internet of Things. I’m actually really sceptical of IoT, but in certain settings it makes sense. Smart watches are one of those settings.

This also really helps with the business model. One could leverage the phone or smart watch, which the customer already owns and presumably purchases every two years, and sell a subscription to one’s AI service. Monthly subscriptions are a tough sell in the consumer market, but better than relying on ads which wouldn’t really fit in here or one-off purchases which wouldn’t generate revenue over time.

The next big technological step always look unintuitive at first. We expect the world to look and act like Star Trek even when its not practical or useful. Instead, think of simplification, convenience, and first principles. People listening to a computer versus yelling at it seems more sensible. This way of thinking is an extension of that. [4]

If you’re looking for help on your next big Machine Learning or Data Science project, let me know! I’m always looking for work in interesting projects. Shoot me an email at

[1] For example, isn’t the accelerometer a great place to track hand motions like a gold swing? (Btw, my Dad is totally looking for this app and will buy an Apple Watch once its made)

[2] How the phone/AI would know this, I don’t quite know. I do feel it can without much hassle to the user though. Also, I just ate so hence the food examples.

[3] And are actually really useful and anti-time suckers. Given how much we spend on social media and mobile games, one can’t expand more from this market of eyeball-attention.

[4] As a closing footnote, people hate interacting with computers. You can tell me how pretty your site looks all you want, but I see people getting aggravated or defeated in the face of technology more often than not. AI that communicates to somebody who doesn’t have to worry about it would be great and a huge improvement. Trying to talk to a chatbot will result in these frustrations more often than not. If I’m wrong, I’d be curious if there’s a working counter example. If there is, please comment below!

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.