Machine learning – McKinsey Insight

machine learningInteresting piece on the McKinsey Insights site on machine learning. It’s aimed at executives to give them an overview of some key issues and trends and worth a read if you want some examples of how companies and researchers are innovating in this sector.

While useful, it does fall into the trap of overplaying the significance of machine learning in the short term:

Now is the time to grapple with these issues, because the competitive significance of business models turbocharged by machine learning is poised to surge. Indeed, management author Ram Charan suggests that “any organization that is not a math house now or is unable to become one soon is already a legacy company.

Steady on. People were saying similar things about the internet and retailing almost 20 years ago and, despite the rise of online retailing, we still have high streets and old-fashioned shops. Of course, the high street is changing and more money is being spent online but in the UK it still only accounts for 15% of retail spending. But I suppose you get less attention by claiming that things are going to change slowly over the next couple of decades rather than “you’re all going to go out of business next week”.

This caught my eye:

Last fall, they tested the ability of three algorithms developed by external vendors and one built internally to forecast, solely by examining scanned résumés, which of more than 10,000 potential recruits the firm would have accepted. The predictions strongly correlated with the real-world results.

I can see that if this software takes off there will be a whole new business in helping people write CVs which score well with these algorithms, a sort of SEO for job applicants. CV Algorithm Optimisation (CVAO) – wonder if that will catch on.

Lessons for IoT platform builders

Platform LeadershipI’m currently reading Platform Leadership1)Gawer, A. and Cusumano, M. 2002. Platform Leadership. Boston: Harvard Business School Press by Gawer and Cusumano which details how Intel, Microsoft and Cisco became platform leaders in their sectors. This is not your usual management book which relies on anecdotes and selective data to support a hypothesis. The authors conducted many interviews with executives at the firms and present a highly readable and coherent account of the rise of these companies. Obviously, the tech world has moved on a bit over the last 15 years but the accounts in this book of how to build successful software and hardware platforms hold many lessons for companies trying to do the same thing in the internet of things sector today.

One passage caught my eye this morning,

IAL2)Intel Architecture Lab – Intel’s division which promoted cooperation between industry players to extend the reach and capabilities of the PC platform maintains significant influence over the design of interfaces that other companies use to interconnect their components to Intel’s chips. Because IAL has successfully obtained this reputation for impartiality and has been able to influence industry standards, Intel can practically guarantee that there will always be a supply of innovative complements around its new microprocessors – unless a completely new platform and set of complementors to that platform were to emerge without Intel’s influence or input. (Page 78)

How quickly that new platform emerged just a few years after that paragraph was written.

This analysis from the Seeking Alpha website a couple of days ago highlights how Intel’s strategy which worked so well in the age of the PC no longer works in the age of the smartphone,

Intel is a vertically integrated semiconductor company that designs, manufactures and markets its own chips. The integrated Intel strategy stands in stark contrast to the more so specialized wings of the highly complex semiconductor industry.

It seems that Intel is perhaps surrendering the mobile space to ARM and others but, as the article points out, is  focusing more on the Internet of Things,

For its latest quarter, Intel classified its businesses according to Client Computing, Data Center, Internet of Things, Software and Services and Other groups. The Client Computing Group then included chip sales to power desktop, two-in-one, tablet and smartphone machines. In effect, Intel has simply folded its old Mobile and Communications Group into the PC Client unit.

It will be interesting to see how Intel’s IoT strategy develops.

 

Footnotes   [ + ]

1. Gawer, A. and Cusumano, M. 2002. Platform Leadership. Boston: Harvard Business School Press
2. Intel Architecture Lab – Intel’s division which promoted cooperation between industry players to extend the reach and capabilities of the PC platform

Accuracy of IoT sensors

thingfulI was just looking through the latest list of startups receiving support from the Open Data Institute (ODI) and saw Thingful on the list. According to their about page, Thingful is:

“a search engine for the Internet of Things, providing a unique geographical index of connected objects around the world, including energy, radiation, weather, and air quality devices as well as seismographs, iBeacons, ships, aircraft and even animal trackers. Thingful’s powerful search capabilities enable people to find devices, datasets and realtime data sources by geolocation across many popular Internet of Things networks, and presents them using a proprietary patent-pending geospatial device data search ranking methodology, ThingRank®.”

This certainly sounds like a useful service which could do to the IoT what Google did for the web. However, a quick look at what “things” were transmitting data near me has made me wonder about how much we should rely on this data. Within a mile or so from my house I found half a dozen private weather stations (the blue circles) transmitting their data through Thingful (the link circles are Raspberry Pi devices which don’t seem to be doing anything). However, the range of temperatures ( in Fahrenheit) they are showing are 33.3, 41.4, 45, 50.9, 51.6 and 69.5. Clearly, they cannot all be correct. It’s impossible to know if this is because some are in direct sun, others in shade etc. or because the sensors themselves are not working correctly. Whatever the reason, it highlights the need for care when relying on crowd-sourced data. Hopefully, Thingful’s ThingRank will be as effective as Google’s PageRank was in presenting the most relevant results from a search query. However, 17 years after its launch, Google is still having to regularly update its algorithm to filter out low-quality URLs – something Thingful may wish to ponder.

Moving beyond subnets of things?

iot walled gardensMachina Research’s concept of Subnets of Things  to describe closed Internet of Things (IoT) systems where the information generated by the “things” is only accessible to one party and possibly its partners is a good description of where we are now with the IoT. This describes many of the Machine-to-Machine (M2M) systems currently operating in industrial processes to help improve efficiencies across a range of functions. While these systems make sense for their owners, they typically do not represent the vision many hold for an open IoT where information is shared (freely or commercially) across the internet.

So is this bigger vision likely to be realised? At the moment I think it is unlikely, at least in the next five years.

Parallels are often drawn with the mass adoption of the internet and World Wide Web (WWW) in the 1990s and, while direct comparisons are inherently flawed, I do think we can learn something by looking back 20 years.

An early key driver of internet adoption amongst businesses and consumers was email. Being able to communicate for free across an open, non-proprietary network with anyone else on that network was an appealing prospect even though we may curse our overflowing in-boxes now. Closed networks such as Compuserve, Prodigy and AOL had allowed email communications for their subscribers but these were limited to being able to email only other subscribers on the same network. Being able to break free of this walled garden started the inevitable decline of those networks.

Similarly, the WWW broke down the barriers to accessing content that was not solely from network-approved suppliers.

Google encouraged further use of the WWW by both content producers and suppliers because it provided an efficient way to find content which more closely met the needs of searchers than previous search engines had been able to.

The launch of Google Adwords in 2000 and then Adsense in 2003 provided a way for content producers to generate revenue from their websites as well as a highly profitable revenue stream for Google to invest further in their search platform.

This combination of an open communications network with a freely accessible content layer and a service which allowed information discovery and revenue opportunities for third parties goes a long way to explaining why the internet and WWW are so central to our lives today.

Currently, the IoT is at the Compuserve/AOL stage of the internet, around 1990. Technically, we have a range of hardware, platforms, standards and protocols which, in theory, could deliver a true IoT but there are some crucial components missing: an equivalent of the WWW to act as an open platform for data sharing and an equivalent of Google to offer discovery and revenue opportunities.

Missing links in Internet of Things platforms

A succinct and useful summary of the functionality and technologies behind a range of IoT platforms is presented in a paper by Mineraud et al. (2015). A couple of comments in the paper caught my eye:

As data is the core of the wealth produced by the IoT,
mechanisms must be available to ensure the sharing and fusion
of data streams from local and external data sources. Today’s
IoT solutions do not support, or support in a limited fashion,
the fusion and sharing of data streams. Yet, it remains possible
to combine multiple streams into a single application if one
knows the URI to the desired sources of information, but
this represents a technical challenge for application developers.

…The principle of data fusion has already been applied to RSS
feeds by the web service Yahoo! pipes, which enables the
aggregation, manipulation, mashup and fusion of RSS feeds
into one. Hence, such mechanisms support the creation of
innovative and enriched web content. We suggest that such
mechanisms should be integrated to IoT middleware systems
to perform similar operations on data streams.

This seems like a reasonable suggestion but my experience of Yahoo! pipes from years ago is that it is not the most robust system and did not really catch on with mashup developers in the way expected. Things may have changed since I played around with the service about 5 years ago but I suspect not.

Additionally,efficient search engines for data streams must be developed
to maximize the quality of services of IoT applications.

Perhaps someone will be able to do what Google did for the WWW all those years ago.

Below is their table of the IoT platforms they surveyed. It is not clear how they measured what the user expectations of these platforms were – this would be very difficult as users are likely to have a range of expectations.

internet of things platforms