Crystal Balls for everyone: How can we make Artifical Intelligence more accessible to the layman?

Crystal Balls for everyone: How can we make Artifical Intelligence more accessible to the layman?

 The most dangerous, misused and thought-annihilating piece of technology invented in the past 15 years has to be the electronic spreadsheet. Every day, millions of managers boot up their Lotus 1-2-3s and Microsoft Excels, twiddle a few numbers and diligently sucker themselves into thinking that they’re forecasting the future.
Michael Schrage

We have no idea what the future holds for us. Not even our society's best minds do. Here are some of their prognostications:

"We believe the effect of the troubles in the subprime sector on the broader housing market will be limited"
Ben Bernanke in May 2007
We don’t like their sound, and guitar music is on the way out.”
Decca Records Executive on the Beatles
"No! No! No! Bear Stearns is fine. Do not take your money out”
Jim Cramer four days before Bear Sterns was on the brink of insolvency

What motivated me to quote the aforementioned experts ? Surely you have read some of their embarassingly inaccurate predictions if not all. Well….I wanted to make a case for how Machine Learning (ML) can help them and our organizations to improve their predictive abilities.

Ofcourse the AI technique is not a silver bullet. For example not every forecasting problem lends itself to ML. Then there is the question of negative data. It isn't always available. If you build an algorithm to predict the way an individual will cast his ballot by only or mostly polling conservatives, it would not be particularly useful. Thirdly, all ML algorithms recommendations must be backstopped with human intuition which sort of undercuts its utility.

Finally, ML techniques are inscrutable to most myself included. Democratizing the technology therefore will prove to be arduous. I am posting to articulate one of many approaches that might help make AI in general and ML in particular accessible to the layman, something I have christened Machine Learning as a Shared Service (MLaSS). Two quick disclosures:

  1. I might not be the first to have used the term.
  2. Secondly my proposed approach is already in effect at companies like Uber and is the reason why tools like Google’s tensor flow are open source:
To democratize ML

Notwithstanding the derivative nature of my idea, I felt compelled to give my most value added two cents. But first lets talk about electrification and automotives, the dissemination of which has parallels with the inevitable emergence of ML as a pre-eminent forecasting tool.

Electrification: In 1881 there was only one city that was distributing electricity to its residents:Godalming, Surrey, U.K. By the 1930’s 70% of households were electrified in the U.S.

Auto-motives: Henry Ford was a subject of great ridicule at most investor Soiree’s when he first pitched the Model A to their community. By 1927 his company had mass produced 15 million Model T’s and by the end of the second world war Tank Connoisseurs’ were arguing overnight over the engineering superiority of the German Panzers over the American Shermans.

ML is diffusing similarly across industrial and corporate America like electricity and the combustion engine were a century ago. ML too will disrupt many industries, in-fact the world entire, like its predecessor technologies.

It will help corporations keep inventory at optimal levels to avoid excess, predict frequency and severity of natural disasters with increasingly improving accuracy, help economists identify leading indicators for recessions that currently are outside the bounds of human intuition, help politician wage more effective campaigns, allow airlines to maintain more robust airline fleets and physicians to provide better medical care

There is however a powerful constraining force acting against the development of all these critical applications: The limited number of machine learning experts. Google executives expressed the same concerns at the search giant’s developer conference in 2016. How can companies build these aspirational intelligent tools in light of this scarcity. I think the solution is hidden in Organizational Principles: I recommend adding ML capabilities to every firm's suite of shared services, just like HR or Risk Management. Just like interviews for all Business Units (BU) within a firm, regardless of their focus, are co-ordinated by its Human Resource so can all of their intelligent algorithms can be built under the supervision of central MLaS teams.

Of-course its not as easy as snapping my fingers so let me flesh out my proposition.

Let us examine the steps to recursively build and improve Machine Learning algorithms

I think any big organization pending some training can/should delegate the responsibility for Data Gathering, Analyzing and Cleansing to the various BU's for which they would not need any ML expertise. These preliminary steps also do not constitute the best use of an ML Engineer’s time. The most productive use of their time is in-fact helping BU’s with Model Selection, Training, Monitoring and Calibration.

But even for these latter steps, ML experts can obviate requests for hands-on assistance from BU’s by building ML Platforms with high levels of abstraction. Abstraction is just another pretentious noun used in reference to software whose developers go to great lengths to minimize its complexity and render its many interfaces as intuitive to the non-technical user as possible. Consider for example Microsoft’s ML Platform (The Azure ML Studio)

The way it easily allows you to import and split data, select an appropriate model , train and subsequently score it , inspires a lot of confidence in novices like me across an organization to run Machine Learning experiments and build useful prediction engines.

Think of all its applications for drug discovery and development:

  • Molecular Biologists can use it to quickly identify targets for therapeutics
  • Chemists can use it to anticipate pharmacokinetics and pharmacodynamics of their lead compounds
  • Physicians can use it to predict adverse events in response to various treatments administered to their patients
  • Cinical Trial Managers can use the platform to stratify the patients based on genomic profiles to various arms of their studies

None of these technocrats need to be ML experts to realize the possibles articulated above.

As a final thought experiment, think of the year 1990 when the president of the United States only had as much computing power and information access as a poor child carries in his pocket today if he/she owns a smart phone. Think of the world that existed then and the world as we experience it today. It has radically transformed, thanks to the democratization of the internet, the personal computer and the exponential improvements in the price and performance thereof.

Society can usher in a similar transformation were it to empower its people with the tools of AI like it empowered them with the technologies of bygone eras.

要查看或添加评论,请登录

Amit Minhas, FRM的更多文章

  • War Rooms

    War Rooms

    Chief John Anderton put his haptic gloves on and mentally prepares himself for his investigation ahead like a Conductor…

  • Industries of the Future

    Industries of the Future

    Werner Jacobi built the Integrated Circuit (IC) in 1949 Steve Wozniak took several and organized them into a neat…

  • Doctor in your Pocket

    Doctor in your Pocket

    Dozens of Healthcare Information Startups today are scrambling to design the Doctor In your Pocket (DIP) application…

  • Bioinformatics Bazaar

    Bioinformatics Bazaar

    I love Apple's app store (AS). You can use it to download applications to monitor the weather, trade stocks, summon…

  • Mind Melding: Its Imminence, Regulatory Hurdles & Engineering Challenges

    Mind Melding: Its Imminence, Regulatory Hurdles & Engineering Challenges

    Neuralink, Synchron, Kernel…….are all companies that want to merge mind with silicon not only to cure disease but to…

  • Machine Learning Services: To Develop Inhouse or To Outsource / Lessons from History

    Machine Learning Services: To Develop Inhouse or To Outsource / Lessons from History

    Until 10,000 BC farmers cultivated plots just large enough for sustenance for their families. Subsequently with the…

    1 条评论
  • The Ethics of Genetic Editing: People vs "The Experts"

    The Ethics of Genetic Editing: People vs "The Experts"

    The Phenomenon History is littered with examples of Public’s antagonism towards Scientific Consensus on environmental…

    2 条评论
  • AI Eats R&D

    AI Eats R&D

    “Software is eating the world” Marc Andreesen (2011) “Software is eating the world but AI is going to eat software”…

  • David Bowie and Pharmaceutical Finance

    David Bowie and Pharmaceutical Finance

    Home ownership was a pipe dream for many until Lewis Ranieri came along, started securitizing mortgages and created the…

  • EU vs Google: A case study in bullying

    EU vs Google: A case study in bullying

    The erstwhile Google created a search engine. That search engine acquired utility like features.

社区洞察

其他会员也浏览了