Another Week In AI & Machine Learning

Driveless

So the hot news this week is that people keep crashing into driverless cars. Wired magazine discusses the issue that the way in which the current generation of prototypes being driven on America's roads are experiencing a high level of accidents, apparently because they don't behave is the same way that ones driven by humans do. 86% of collisions in California this year have been due to being rearended or side wiped! Autonomous cars appear to be over cautious and annoy human drivers, not only by stopping unexpectedly but also because they are over compliant with the letter of the law in interpreting situations. So why don't they have signs on the car, just like a learner does to warn drivers that they need to keep clear? It's a pretty similar situation and people do tend to steer as clear as practicable from learners who they know to be unpredictable.

AWS Capability Continues to Grow

AWS also held its AWS Innovate Online Conference. In his "state of the nation" talk, Boaz Ziniman outline current capabilities available Off the Cloud (OTC) rather than Off the Shelf (OTS). Basically, AWS still provides impressive capabilities which allow a developer, data scientist or organisation to just start using AI and scale rapidly. Though they have been adding to this capability with impressively powerful capabilities around visual recognition and voice processing as well as bundled environments which are pretty much deploy and go. this is very liberating, because it provides a readily used Pay as You Go (PAYG) capability, which avoids much of the traditional issues of continuously having to research, select, implement, integrate and tune the suite of tools and infrastructure needed, as well as continuously returning to CAPEX approval and purchase processes for scaling to deal with increasing volumes of performance issues. 

There's a caveat though. Some of the capabilities, e.g. recognising a face in a crowd, which may be useful for security solutions, might also be used as the tools for enforcing a police state or merciless pursuit by paparazzi. Enabling such massive AI at scale, will to some extent be another nail in the coffin of personal privacy.

Early Adopters Surge Ahead

Finally, MIT Sloan and Boston Consulting Group released a report "Artificial Intelligence in Business Gets Real" which surveyed the application of AI in business across the globe. Whilst there were the usual scare stories about China being ahead of the West in adoption and that the gap between pioneers and laggards is growing larger there were some interesting points. The Chinese are deploying to meet efficiency and cost needs. Everyone else is deploying, because AI helps them do more and there was a message that AI will not replace jobs, but it will change the nature of work and the skills needed.

The second key message, was to experiment with something simple and demonstrate the benefits, because business leaders who had seen AI in practice, get it and are prepared to invest more, to the point that AI is almost addictive in the way in which it influences investment appetite once a business has tried it.

The final message was really around AI at scale. If you are planning on building your  future business model around AI, then you need to plan, prioritise long term investments and to get effective data governance in place. This does not just mean establishing once source of the truth, with valid, complete and coherent data. It also means having a good handle on version control. Since, if multiple applications of AI depend on the same data, it needs to be synchronised to avoid unintended problems. So Lean Data practices are fundamental to adopting AI at scale.

Finally, it was interesting to hear how paranoid Chinese companies, adopting AI, are about cyber security. Protecting their data from competitors and the continuing availability of data and AI based solutions becomes increasingly important once a company's business model has evolved to adopt AI. So the principle "Cherish your Data" (see The Way of DAU) is key to successful exploitation.

Conclusion

Human issues, ethics and common sense remain central to AI adoption, an Agile mindset encourages adoption and Data Governance is a key enabler.

Artificial Intelligence and Other Tales

Engineers at Google recently published a paper outlining their work developing a custom AI chip which is so efficient that it saved them having to build new data centres to cope with the introduction of AI based services. The paper itself provided a great insight into the scale and nature of Google's operations. However, it was the catalyst for musing on where we are going with AI and what it all means.

In the 1980s there was a lot of hype around artificial intelligence resulting in thousands of university undergraduates learning languages such as LISP and Prolog, and the Japanese Government sinking hundreds of millions of dollars into its Fifth Generation Computing programme in an attempt to progress the technology further. More modestly the UK's DTI published a few pamphlets and books on state of the art Intelligent Knowledge Based Systems (IKBSs) and Neural Networks. A few rather expensive products hit the market. There were a few relatively trivial case stydies and then it all just seemed to fizzle out, gradually becoming forgotten as the Internet Bubble grabbed people's attention.

However, AI did not go away, its development just went into submarine mode. People were there working quietly away in the background on specific applications of AI and bringing them to a level of maturity where they can be adopted wholesale in real life situations. In the process there have been substantive strides, so for example voice recognition systems now require little training and accommodate regional and national accents. Additionally, there has been a little gentle re-branding to adopt the term "Machine Learning". Consequently many people use "personal assistant" agents on their phones or built into household automation devices. Self driving cars are reaching commercial aodption and the next generation of UAV aircraft will be capable of accepting a programmed mission, taking off and completing it without human intervention or interaction, unless some mission parameter is encountered which requires human decision making or to receive required intelligence.

I was literally blown away with incredulity recently, when someone showed me how easy it is to set up a simple machine learning application in Azure and train it to produce a useful output. But I have been reading recent alarmist claims that "AI robots" will replace millions of human professions with a level of scepticism, as this is not substantiated by previous experience.

In the 1980s, when I worked for a bank, there was an interesting article in the Financial Times on the adoption of IT by the banking industry. The gist of the story was that if the banking industry had not adopted IT to mechanism a lot of its work, then the explosion in consumer banking products in the UK which happened at that time would not have been possible without employing the whole of the UK workforce to support it.

A recently touted robotic brick layer is unlikely to eliminate all brick laying jobs because there is a current shortage of bricklayers and robots will not be economic on small sized jobs. Furthermore, the reason that the United States has traditionally enjoyed higher productivity than Europe is not a result of American technical superiority, but the result of the fact that the US has always been resource rich and people poor. A shortage of labour and skills leads to new methods and investments in mechanisation and automation.

Furthermore, the UK Government Digital Strategy is promoting the investment into the development of AI skills, because the Government believes this will grow both the economy and jobs. The real impact will be in the change of the nature of the jobs. People will do less humdrum stuff and explore their curiosity more to invent new things, discover new things and do their jobs more creatively.

Platform Schizophrenia

This year I became aware that there are two definitions to Digital Platforms. Whilst I had been meandering around in IT Space thinking that digital platform meant services like AWS and Azure, our friends in Marketing Space had decided that digital market places were Digital Platforms. So to them ComparetheMarket.com, Deliveroo.com and Uber.com are platforms. 

Anyway moving on from this diversion, it has for some time been a surprise to me that amazon has dominated IT thoughtspace and the market for PaaS based Digital Platforms, whilst belatedly Microsoft pushed into the market with Azure. 

In recent projects I have been involved with both AWS and Azure as well as all sort of fun with the OSS tools which are available on them. To traditionalists coming across the database as a service offerings available is quite amazing. I was also blown away when a Solution Architect who had no experience of Neural Networks was able within 2 weeks to knock up a fully working and trained prototype of a Machine Learning application on Azure.

So it has become increasingly interesting to see that Google, one of the most born in the cloud companies going, has recently started promoting its services. One has to ask why did they wait so long, especially as they have always made much of the fact that their products are all architected around a SOA concept and the ability to expose themselves as services, both internally and externally.

Oracle and IBM have also appeared actively in the market place this year promoting their own special blends. 

The thing about this is that they all have really good stories to tell. You will note that I am not stating any preferences, as to be honest, anything I say about them today will already be wrong tomorrow as this is an ever faster moving situation. Today's facts will be obsolete tomorrow.

So what does it all mean to the average business trying to go Digital?

Firstly, the means are there. You have to be comfortable with the fact that terms and conditions are what they are. You need to examine the pricing and understand how this would play out in some key real world applications. However there is plenty available to "Free Your business from the Tyranny of Infrastructure" and Focus on Value. If you choose reasonable sensibly, you will be able to scale costs with business activity and exploit platforms which support Agile and DevOps so you can move quickly and lightly in the pursuit of opportunities. All the major vendors are investing significantly in security and if you dig deeper, most offer localisation options if data cannot move outside certain jurisdictions. Additionally there are industry certification schemes which many providers are signed up to. So a lot of inhibitors have been addressed.

The key issue is going to be how much do you insulate yourself from the risk that you may need to change platform provider. Business Performance, Legislation, Pricing etc. will change with time. So you may need an exit plan. Therefore, some thought needs to be given to insulating yourself from future supply threats. Where your application is going in for short term gains, e.g. a new financial instrument which will only be around for a a few months or perhaps a couple of years, this is not a problem. But if you are locking yourself into a platforms specific machine learning solution for years, you may need to think how you would deal with problems if the platform vendor ceases training.

In the end, however, we have always faced these problems. Finding a totally vendor agnostic solution has always been too complicated and too costly. So its time to get comfortable with not being in total control. The System of Systems concept of de-optimising components to integrate and optimise the overall performance of the Big System applies. You just need to understand your risk appetitie, your risks, how you want treat them, what you will accept, what you need to insure against and get on with it. The risks of not doing so are far greater.