On December 13th at a very announcement-packed event in San Francisco, Microsoft shared its views of Artificial Intelligence and the progress it has made thus far.
Back in September, Microsoft created a new AI and research group of about 5,000 people under the leadership of Harry Shum and they have certainly been busy. Microsoft announced several different AI initiatives during the event:
- A new chatbot called Zo.ai that is integrated into messenger app Kik
- Cortana Devices SDK so Cortana can be on all kind of devices including what looked to be an Echo-like smart speaker manufactured and branded by Harman/Kardon
- Calendar.help that has Cortana looped into our email to schedule meetings so she can do it for us
- New Calling capabilities for Skype bots and the ability to include rich media
- Microsoft Translator Live that lets you have a real time conversation with people who speak different languages.
Some key milestones were also shared at the event:
- 67,000 active developers using Microsoft Bot Framework
- Chinese Bot Xiaoice and Japanese bot Rinna measure conversation per session that average 23
- Record length on an interaction with Zo was nine hours and 53 minutes
The list of achievements is significant but it’s the picture that develops when looking at them in total that really shows how invested Microsoft is in AI.
Microsoft Has Yet to Capitalize on the App Economy but is Making Strides with Bots
Nobody would argue that Microsoft missed the whole “app economy” craze. While Windows Phone caught up with iOS and Android from a technology standpoint, the low market share left developers with little interest in developing for the platform. Universal apps improved the landscape a little, as they allowed developers to maximize their effort by having their apps run on different devices, but the gap is still there.
Microsoft seems to have learned from its mistakes and is making sure it will not be left out of the next craze that is chatbots. Many have talked about chatbots as the next killer app and, while I do not see them as such, there is something to be said for some of the roles played by apps today being replaced by bots. For example, a travel chatbot helping you book your vacation in the same way you currently do through an app. So, while Windows might not have the best travel apps, it will allow users to get the job done through bots.
Long term, bots will certainly make the app gap a non-issue but the transition will not happen overnight and Microsoft needs to ensure users will be engaged on Windows in the first place. Having 67,000 developers actively engaged with the Microsoft Bot Framework is encouraging to see as was the long list of services already taking advantage of the framework. A crucial point Microsoft made during the presentation is the need for these chatbots to be freed from the messenger apps where they now reside. Clearly an easy starting point, given the need to encourage interaction with bots similar to the ones we have with a real human being. Bots need to be free to show up in email, on the web and other places we visit every day. Ultimately, bots need to follow me wherever I might need them.
Amazon and Microsoft Share Their AI Approach and Goal
Microsoft, like Amazon, talked about democratizing AI by allowing everyone to integrate Cortana and Alexa in their devices. This is because both want to be the underlying platform of preference for the AI revolution. Ultimately, the revenue that will be generated by empowering other hardware and services will be so much greater than what homegrown hardware could ever generate. In a way, this is no different than what Microsoft did with Windows and PCs. Windows became the platform for the computing revolution and now the Microsoft Bot Framework and Cortana will become the platform on which the AI revolution will be built. Let’s not forget that AI needs constant feeding of data, constant learning which will come from different use cases, on different devices, in different circumstances.
Conversational AI’s Will Wow Consumers….Eventually
Microsoft also talked about how AI cannot only be about IQ but how, in order to create a true bond between humans and digital assistants or bots, AI needs a strong EQ.
I could not agree more with that philosophy. I have spoken before about how Amazon has been able to create a strong bond with Alexa by just giving her a name while Google, with Google Home, has created a certain distance between user and machine. Personifying your assistant helps create that bond but having an assistant that understand emotions, nuances in vocabulary, tone, expressions – if I can be seen – changes the interaction at its core.
Microsoft, is of course, not alone in wanting to focus on natural language in order to create a conversation. Google has been talking about this very topic when showing off Google Home. Yet today, in most of our interactions, the experience feels quite transactional. I ask a question; I get an answer. If I do not get the right answer, I might ask again but that is pretty much it. This is not really how conversations happen.
Personality is a big part of EQ and, if Microsoft wants to build a true conversational AI, it has to focus on creating personalities for Cortana and its bots but also teach them how to speed-read people so they can adapt that personality.
Zo.ai is the successor to the short-lived Tay.ai experiment. Tay was trying to replicate a millennial and very quickly we were exposed to what happens to a millennial’s personality when social media tries to push some buttons.
I interacted briefly with Zo and I have to admit I was not impressed. In her defense, though, she must have been thrown off by the fact that someone who she clearly thought could be her father – she told me she was 22 and that older men do everything better – was conversing with her. What is important is that each and every interaction we have makes Zo more aware. I see Zo going through a mix of biosensor stimulation and socialization so she can learn to cope with different situations in the future.
Delivering Value Now!
All of this will take time and Microsoft is smart in delivering value now so we, as users, start to build our trust with simple tasks and conversations. Microsoft Translator Live and Calendar.help are two great examples of Microsoft using AI to take pain points away from our day to day lives. Having a three-way live conversation with people who do not share a common language is something that, if you have ever worked on an international team or married someone of a different nationality, you can easily relate to. The nightmare of checking availability to set up a meeting with multiple people is also something many of us have experienced. These might be seen more as tasks but they make a big difference in our lives. For me, it could be as simple as allowing my daughter to have a rich conversation with my mom; something I would see as extremely valuable and would create an immediate reliance on Cortana or whoever else is empowering that experience.
The difficulty I see for Microsoft and other AI players is to make sure users connect the dots and see that bots, assistants, translators, calendars, maps, and the list goes on, all share a common brain and that to talk about AI is not the same as to talk about a personal assistant.
12 thoughts on “Microsoft’s AI Is Not Just About Being Smart”
I’m still struggling a bit with what I’d want an AI for. Kind of like smart watches: I can imagine a few cases were it’d be nice, but overall, not worth the bother.
What’s the killer app ? I feel either it can act as a full personal assistant, reminding me of stuff, being proactive, or it’s a “5% of the time” thing that can be done without, that’s actually more disruptive (not in *that* sense ^^) than not having it at all.
Then there will be the certain disconnect between what we think AI is, what it traditionally was, and what BS the marketers re-define what they think it is for them, therefore us.
Remember, it’s on these pages, where a mobile device (mostly iOS based) magically transformed into a PC, but a PC became a truck… ;-P
I could be wrong, but I don’t think AI is what you will actually buy. I think it will be the already typical things that have AI underlying its use and interface. Sort of like Nest’s failed attempt at AI thermostats. Also, even with human driver cars I am willing to bet dollars to donuts AI will become an integral part there, such as cars communicating to each other while we drive, etc. The personal assistant idea, is not intrinsically AI, but it would utilize AI, no doubt, to learn about you.
This is not to say it is the best way to do this, but I don’t see it working any other way. How do you market AI? You don’t. You just use AI to ostensibly make your products better.
Other than that, I got nothin’.
I’m drawing a bit of a blank as to which product/appliance I’d want AI in. I wouldn’t count, say, an oven that recognizes I’m doing a roast and cooking accordingly as AI. Do self-driving cars qualify ? That’s awfully single-task, isn’t AI, or just I, supposed to be versatile ?
Yeah, but I don’t think we are going to get to choose. I remember when people said they didn’t want any computer in their car. That ship has sailed.
But I don’t think it will be an AI for this and a separate AI for that. I think it will be the same Ai in a number of applications.
I think that calling the current generation of machine learning algorithms “Artificial Intelligence” is what is causing confusion. What we really have at this point is only a breakthrough in pattern recognition. Computers have not learnt cognition nor reason, and Deep Learning is not targeting this segment of intelligence at all.
As the most, we have artificial intelligence that matches frogs. We do not have artificial intelligence that understands the circumstances, has empathy, or has cognitive abilities. We are still very far from AI that can match humans, even if they can win a game of Chess or Go.
The evolution from frogs to humans took many hundreds of millions of years. Similarly, it is quite a stretch to assume that the current crop of AI will quickly evolve to match human beings. Yes, AI will most likely evolve to be really helpful and serve as a reliable assistant. However the current limitations will relegate AI to very specific functions that can be satisfied by pattern recognition. In our current lives, there aren’t too many of these functions left that are done at a sufficient scale to warrant automation.
Therefore, there is no surprise that the current AI assistants aren’t very useful or that the killer app isn’t there. What I do think though, is that AI could be used in place of our motor functions. That is, they could take the tediousness out of entering a calendar event. Humans, after reading an email, can easily parse the event information. However, entering the event into the calendar is actually what takes many clicks and time. AI could help us by saving the motor functions necessary to click on the right buttons and enter the right dates by hand. AI is a great substitute for computer macros and shortcuts.
“We are still very far from AI that can match humans, even if they can win a game of Chess or Go.”
Wholeheartedly agree. Deep Blue beat Kasparov. But who, other than the entire team of chess players, strategists, and programmers, was Deep Blue?
Kasprarov didn’t even have a machine opponent, he was opposed by a machine enabled team. So yes, my eyes gloss over.
What is profound about Deep Blue, Watson, and the like is though they are “blunt instruments”, they enable non-Kasparov’s with Kasparov-like abilities.
In terms of Chess etc., from what I understand, the idea is to explore the whole space of possible moves, and to do that for as many steps ahead as is computationally feasible. Depending on the stage of the game, if there are just too many possible moves, then you can use pattern recognition to search for situations that match the current one, and then lean on expert human/historical advice on what to do.
Obviously there is much more to this, but I believe this is the essence.
One might say that in fact, humans are not doing much more than simple pattern matching, and if this is improved upon, we will eventually create something that resembles the human mind. Well, since science has not really solved the human mind, it’s hard to debate either way. However, we should note that the animal brain is divided into many discrete regions each with different functions, and animals that predominantly rely on instincts do not have a large cerebral neocortex, for example. This suggests to me that we need a further discrete innovative step until machine learning can match human beings.
“One might say that in fact, humans are not doing much more than simple pattern matching, and if this is improved upon, we will eventually create something that resembles the human mind.”
Don’t underestimate the random element of human actions, and the “butterfly effect” these can potentially cause. Humans don’t always do the obvious, traditional, optimum, or most logical, sometimes with great effects.
“Creative people live in two worlds. One is the ordinary world which they share with others and in which they are not in any special way set apart from their fellow men. The other is private and it is in this world that the creative acts take place.” Mark Kac
In my opinion AI can be a Family Doctor. As a matter of fact I first thought that was what Google was rooting for with Home when they first came up with a tagline for it: “Always on call”. I know it has being tried before without much success. I think now with an abundance of a personal information and a wearable tech that follows us everywhere we are close to the tipping point.
Another function for AI which I am personally less interested in, but see a lot of usefulness in it is in-House salesman.
The question remains how rudimentary these functions can be served by AI, since I have doubts about its intuitiveness. I think they become “good enough” when AI will learn to be a good entertainer, which goes pretty well with “pattern matching” that some people mentioned here. With an entertainer function being a common denominator I think other functions become more practical.
I love in-house sales visits. Dog food is expensive!
I remember the line the housewife tells to her husband, who is working in an advertisement in TV series “Mad men”. She justifies letting in to the house to a traveling salesman: ” Doesn’t a salesman enter our house every time we watch a TV commercial?” Very true. I thought the intent for AI is also in trying to sell “you” outside the house, like PR or finding a job for example.