And another thing 70
by Jon Honeyball
Jon Honeyball ponders the full implications of AI, and what companies need to do if they are to reassure the end-user.
HardCopy Issue: 70 | Published: November 4, 2016
The news that Microsoft has decided to take AI seriously is not something to be overlooked. It is of course a logical step when you consider the engine you have at your disposal once you’ve built a cloud infrastructure that is farming almost impossibly large quantities of data belonging to you and me.
That the news is focused on Microsoft rearranging MS Research is also a breath of fresh air. MSR is one of those organisations that was set up to do original thinking, with world class engineers based around the world, including at such academic centres as Cambridge. And there is no question that MSR has contributed significantly to the overall technological ecosystem at Microsoft, and within Microsoft products.
It could be argued that the influence has been more undercover than in plain sight, but that might have been for the best. Where the relationship between MS and MSR failed, according to some I have spoken to over the years, is in coming up with innovations in that extraordinarily difficult area which is ‘just around the corner’. In other words, the broader canvas which Microsoft ought to have kept a closer eye on, but didn’t. Some have argued to me that MSR saw that role as belonging to the product groups in Redmond, while some in Redmond have argued that MSR should have been the one with the eyes on the future just beyond the next product release.
Whatever the truth of the matter, and there will be endless discussions of this over almost unlimited quantities of beer, the reality is that Microsoft missed the perfect storm of ARM processor technology, touch screen glass screens and emerging battery technology. This is what led to the iPhone, a product which turned an industry on its head almost overnight. Nothing Apple did was magical, despite its oft-repeated claims to the contrary, but it did have its eye on the ball of what was possible in the near future, and pulled it all together with great success.
To do AI well requires huge computing power and access to vast databases that can be trawled to teach the engines. Anyone who has spent a boring hour of their life trying to train a voice recognition engine will know the almost futile effort required, and the results are often not pretty. However, pre-analyse thousands of voice patterns and the training can be reduced to almost nothing. Indeed, on a modern cloud-powered AI system, vocal training has all but disappeared. Yesterday, I plugged my new Amazon Echo system into my house. The first vocal command I gave was “Alexa, play me the album A Walk Across The Rooftops by The Blue Nile” and it knew exactly what it had to do. This sort of AI is truly remarkable, and can be game changing for the user.
When it comes to mining data, there is an almost unlimited amount of improvement that can be made. It drives me to distraction that I cannot ask the question, “I was talking to Bob, or maybe it was Joe, about the concrete construction of that bridge in New York, some two weeks ago. Find me what we discussed.” The mix of deep AI along with the concept of a time line is something that we have never really had the opportunity of enjoying; and I don’t count local hard disk indexing engines as being anything other than the topmost millimetre of the toe-nail connected to the toe in the water of real search. So much more needs to be done, and deep AI is the engine that will allow this.
When I look at my internet use, a truly frustrating part of it is simply looking up stuff that I know something about already. The phone number of a restaurant I have previously visited; or maybe I will be in York next week and I want to find an interesting restaurant. I know sort-of what I want, but I have to go through mechanical hoops to find it. An AI engine that had looked at my diary, noticed that I was staying in the Travelodge next to the University for the night, and had come up with a bunch of restaurant suggestions which it could slip into my visible world for consideration, would be a game changer. Even better if it could phone the restaurant and make the booking, although I guess going through a table-booking website would be easier.
When it comes to business data, the search for the elusive two percent has always been the Holy Grail. Finding that small but vital competitive edge has kept many people employed doing endless Excel worksheets, often tied to deep “SELECT * FROM …” type SQL queries. Having cloud services that offer deep learning and mining capabilities will be a game changer for many industries, including those who consolidate in the middle tier. For example, finding airline routes that fulfil a somewhat more fuzzy set of criteria than just when and where could be worth looking at.
There are several underlying problems to consider, though. Firstly, AI research into huge data sets can often uncover things we didn’t know. In other words, answering the question we had not yet asked. While this can be fascinating, or even game changing, it could easily lead to the unveiling of more than we wanted. For example, consider the relatively limited information that an insurance company currently has to work with in order to present you with a policy pricing. Imagine what more it could do when it could data mine your LinkedIn profile and your Facebook postings too.
And therein lies a problem: how much about ourselves are we prepared to reveal, and are we even aware that it is happening? Finding that last two percent in insurance premiums might mean noticing where you happen to drive, and weighting things based upon your proximity to past accidents based on geolocation and time. It might seem ludicrous now, until such an AI-driven data search becomes possible, at which point it almost immediately slips into the mundane and run of the mill. Such a step change has both ups and downs. Mining across global-sized data sets requires more than “SELECT * FROM”, and it has to go considerably further than the sort of fuzzy searching done by the likes of Google or Bing.
Secondly, does the user really know what they are signing up to and agreeing to accept? It could be argued that, as an industry, the software industry has a pretty shoddy reputation for hiding things away on the bottom of page 79 of the licence agreement, which you can scroll through only one quarter of a page at a time. When are we going to see much more openness and clarity about user data, and how it might be used and processed in this new AI world? What rights do I get as a user to examine, limit, edit and ultimately delete such data? We cannot move forward on an assumption that data slurp is just fine. The difference between micro and macro is already overstretched, and it is time for vendors of all persuasions, both large and small, to increase clarity, not hide away behind licensing waffle.
Despite all of this, the move to AI in cloud services will be the next big step. From personal computing to internet, from cloud and now to AI cloud, the benefits are both obvious and hard to predict. And therein lies the challenge: using these engines, whether they are from Microsoft or Google or Amazon, or some other company that has yet to arrive, in ways that delivers seamless integration into people’s lives and yet doesn’t jeopardise the underlying trust required to ensure that users, both personal and business, are prepared to continue to contribute.
We know that the user is the product. It is a fine line between trust and abuse, and I am not sure the industry is, as a whole, in a comfortable place on this right now. Just one look at the financial services industry will show you what happens when ubiquity and limited choice combine with money to create a recipe that is often far from the best interests of the customer, and yet the vendor is so big, so powerful and so mighty that they transcend even government level oversight. This is why Macro Versus Micro is going to be a massively important underlying issue here. How can we gain benefit for all (Macro) whilst still respecting the needs, concerns, privacy, and simply human frailty of the individual (Micro). So far, we have done a pretty bad job on this front. It’s time to come clean, stay clean and start to build solid trust.