Artificial Intelligence

Google, Amazon and the upcoming battle over AI Assistants

Artificial Intelligence

The stage is set for the coming battle between the big five tech giants: Google, Apple, Facebook, Microsoft, and Amazon (wait, make that four tech giants and one really tech savvy retailer). All are now heavily investing in AI. All now offer personal AI assistants poised to make your life easier.

On the surface, this looks like head to head competition around the AI personal assistant by five of the top tech companies in the U.S.

One of the most recent announcements is from Google. The company launched Google Assistant, which is an integral part of the new Google Home device and Allo (its new messaging app).

Apple is set to announce its additional AI assistant plans mid-June but has already leaked some details. The Information reported that the company is developing its own device for the home and that Siri will open up to third party  “apps.” Earlier this Spring, Microsoft and Facebook made their own announcements which put AI Assistants (or in their parlance, chatbots) front and center. And each has a clear vision for how third party bots can exist as part of their platforms.

On the surface, this looks like head to head competition around the AI personal assistant by five of the top tech companies in the U.S.

But what’s really at stake—will we eventually be paying obeisance to Siri or Alexa, while their competitors whither? Is the personal assistant the key to “the era of Artificial Intelligence”?

Here are my five immediate takeaways:

 

1. The AI Market Segments

One way to view the current setting is to divide what is happening into three distinct market segments (Platform, Service, and Software), and you should know which segment any given company plays in to better understand its ambitions.

AI Platform
Intelligent Platforms have a set of existing features that you use to build bots (Applications). These include Facebook’s Messenger platform, Microsoft’s Bot Platform (its bots can run on Skype) and all the way up to Amazon’s Echo device, which lets third parties create new skills for Alexa.

Pro: The most dramatic and most important dimension is distribution; these platforms come with easy access to hundreds of millions of users. Almost any programmer can build a bot on the platform, and the time from idea to market is short– think days or weeks.

Con: The most obvious downside is a limited opportunity to innovate outside of what the platform was designed for, which to a large degree consists of the sensors and actuators inside their environments required for the bot to exist, leaving much of the actual intelligence to the developer. And we probably have to accept that the bots are not (yet) true intelligent agents. Platform dependencies put large parts of your destiny in the hands of the platform owner. There’s very limited opportunity to optimize much on output accuracy, and so your product quality (outside of design) depends on the platform. Finally, those users are probably not going to be given to you for free, so expect an App Store like tax.

AI as a Service
These outsourced and on-demand machine learning services allow developers to build models and generate predictions in the cloud, without having to engineer and/or maintain the supporting infrastructure. These include offerings from companies like Amazon Web Services, Microsoft’s Azure and of course Google’s Cloud offerings.

Pro: You can build anything for any platform or channel and move forward pretty rapidly– think weeks or months. You do not have to maintain or build your own machine learning infrastructure.

Con: This comes with some sort of recurring cost which might just hurt your unit costs or costs in general. It is likely that some ceiling to accuracy arises given the fixed constraints of the services.

AI Software
You could build your own AI infrastructure from the ground up. Even if you want to do all of this yourself, you’re likely going to end up using either proprietary software and/or open source tools like TensorFlow from Google or say scikit-learn, Theano or Spark’s MLlib. These will help you with the learning aspect of building an AI, much like the services described above, but there’s typically a whole lot more that you’ll still need to build yourself. For example, you’ll need to collect and clean your data, engineer features, evaluate your learned models, serve predictions, incorporate changes to your system and so on.

Pro: This provides the ultimate freedom and allows for very high accuracy in output. True innovation can happen here. It can provide long term cost benefits through infrastructure optimization as well.

Con: This is very time consuming and thus quite expensive as an upfront investment. Anything you build should be thought of as a months- or years-long project.

This is not necessarily a perfect segmentation, but it does provide a good backdrop.  And, given that each of the five tech giants is playing in multiple segments, it suggests to me that this is indeed the era of AI. It also suggests that none of the five is immediately convinced that the personal assistant is the only way to dominate in this new era.

 

2. Cross Device

AI Personal Assistants will be cross device. The recent announcements also point to a scenario in which these assistants will become ubiquitous. Siri is already in your Apple TV, and at the re/code conference, Jeff Bezos announced that Amazon will license Alexa to third parties.

It’s clear that the only way for these behemoths to compete is to make their AI Assistants available across devices.

It’s clear that the only way for these behemoths to compete is to make their AI Assistants available across devices.

This presents some interesting political challenges right away, as Siri’s and Alexa’s skills converge, Apple would seem pretty uninterested in making Alexa available on your iPhone, for example.

The funny thing is that the player who used to be the most protective about their O/S Platform, Microsoft, has been among the quickest to push its assistant across devices. You can get Cortana on Windows 10, iPhone and Android. I think this willingness could let them leapfrog the others, should they not be as forward-thinking.

 

3. Horizontal vs. Vertical AI

The structure of the AI Assistant space is now crystal clear: Horizontal AI will integrate with and enable vertical AI. All five companies have built what I like to call horizontal AI, which is to say AI Assistants that operate more as enablers of more focused services (like x.ai).

Horizontal AI will integrate with and enable vertical AI.

They’ve also all signaled that they would like third parties to develop these more focused services, just as third parties developed the apps that populate the app store and our smart phones. These are early days, so there are few fully realized vertical AI services in market. (We have built an AI personal assistant who does only one thing: schedule meetings for you.) You should expect to see much more activity on this front. The value of these horizontal AI assistants will hinge to a large degree on the quality of vertical agents that each can enable and the seamlessness of those integrations.

But don’t expect exclusive integration deals with Siri, Cortana or the like. Any vertical AI agent will want to offer its agent up everywhere. And there is precedent for that model:  popular apps go cross device quickly.

 

4. Conversational UI

The conversational UI, whether in voice or text, will be the dominant interface for these horizontal AI assistants, and for many, if not almost all, vertical agents as well. Whether you are texting with M or chatting with Alexa or asking Amy to set up a meeting, you will be holding a conversation rather than navigating a visual interface. Right now, those conversations can feel stilted and awkward. This is one area in which a single player could really separate itself from the pack. I’m not placing any bets, but to achieve a truly natural level of interaction, you need masses of data and the right data and the willingness to massage it. Here Apple seems to be most disadvantaged, for the moment.

 

5. No Winner Take All

This is not a winner take all setting. While one or two players may fail to woo consumers to their AI personal assistant, I do not see any reason for a single or even just two players to dominate the market. After all, Alexa gets only marginally better if I use her and you use her too, and this has to do with data collection (the more and more varied data, the smarter she can get).

The opposite may be true for many vertical AI assistants. For example, if Amy were to run everyone’s calendar, she could find the most convenient location for all meeting participants (not just the host) because she would know where each one has been and where each is going on any given day. That means less travel, which saves everyone time.

One thing is indisputable: the major parties in the AI battle have assembled. Now the war for consumers hearts and minds begins. This should be great for the consumer and fun to watch 😉

[And many thanks to x.ai Data Engineer Jeff Smith for his contribution to this piece.]

 

This post originally appeared on LinkedIn Pulse, here.