Ali Jafri's Blog

My thoughts on tech & futurism

You are here: Home / Archives for Machine Learning & AI

My take-aways from EmTech Digital 2016 @ SF.

by

EmTech_SF_2016Artificial intelligence, autonomous vehicles and digital agriculture, the Emtech Digital conference in San Francisco opened a window into the future. For two days some of the smartest people in the world shared their knowledge into the most promising areas of AI research.

Last month, I had the pleasure of attending EmTech Digital 2016 hosted by MIT Technology Review. It was inspiring to hear from such an extraordinary group of people about their vision for a better future enabled by machine learning and robotics. This event was particularly meaningful for me; not only do I have deep personal interest in AI stemming from my days in machine learning research, I’m also responsible for strategy and new product development at Digitate (where our core product Ignio is a machine learning-based enterprise technology platform).
Here are a few key themes from the conference:

Machine learning is 99% human work

The opening speaker, Oren Etzioni, CEO of The Allen Institute, put the focus on AI’s true value; as a facilitator & augmenter of human talent rather than a replacement. He went on to say that this will remain the status quo for the foreseeable future as intelligence doesn’t mean autonomy. The value of close human-machine collaboration was illustrated by Anthony Goldbloom, when he discussed the results of Kaggle, a platform for predictive modelling and analytics competitions. In recent months, teams had tackled issues such as heart failure, mapping dark matter and challenges in chemical informatics.

AI is the new electricity

Baidu’s Chief Scientist, Andrew Ng, likened the transformative potential of AI to that of electricity, which upended every single industry almost a century ago.

Companies are hiring chief data officers or chief AI officers, whereas there used to be VPs of Electricity. Now electricity is just assumed to be everywhere.

Major tech companies already know this. Peter Norvig, Director of Research at Google, described how his firm is re-positioning itself as an AI-first company. Alan Packer, Director of Engineering at Facebook described a recently launched Applied Machine Learning group that the social network claims helps users find content more quickly.
As witnessed from the firms presenting at the conference, AI has already started making inroads in more traditional industries. Narrative Scienceleverages natural language processing to generate clear and legible reports from data. The Climate Corporation aims to end world hunger by bringing smart machines into agriculture, an industry which is surprisingly receptive to new technology. Kensho, maps the impact of real world events to financial markets and they boast the CIA as one of their clients. GE is pioneering the concept of a “Digital Twin” focused on the industrial sector to model wear & tear. And, last but certainly not least, John Kelly, the “father of Watson”, discussed the role the IBM Watson is playing in transforming the Electronic Medical Records space.
Despite the promise, AI still has a long way to go. Things like nuance, context, causation and common sense are currently out of the scope of AI (seeWinograd schemas for an example). Machines have a hard time with ill-structured data and queries. In order to be a viable solution, machine learning must be trained with large amounts of data which would incorporate all possible known scenarios. The machine learning algorithms then make inferences based on this data. However, the results can be highly unpredictable. A recent example is Microsoft’s AI-powered twitter chatbot, Tay, which was quickly shut down after it started sending racist tweets.
Therein lies the challenge; the data isn’t exhaustive enough, is unstructured, or simply doesn’t exist. This is the next critical barrier; how to improve the efficiency of existing AI algorithms and reduce the data need.
However, even in its current limited state, AI is driving economic value in areas such as financial risk, big data analysis, language and image processing and automated report writing. Furthermore, even though AI has been around since the 1950s, after having gone through multiple boom and bust cycles, the so called “AI Winters”, it has only now started entering the mainstream. With a veritable army of researchers, scientists and corporations vested in the growth of AI, these challenges aren’t insurmountable.
All the presentations can be viewed from the EmTech Digital portal.

Cybersecurity & Machine Learning: A great mix

by

Last month, the United States Office of Personnel Management (OPM) announced that it was hit by what has been described as the largest breaches of government data in history. Hackers appropriated over 20 million records containing not only personally identifiable information but also highly classified security clearance data.

This event was far from isolated; one only need to look the data breaches perpetrated at firms such as Target, Ashley Madison, Chase, Sony and a myriad of others over the past few years.

The director of OPM (who recently resigned), was defensive of the agency’s security protocols in light of the “millions” of unsuccessful attacks EVERY MONTH!

A machine learning based cybersecurity framework would have detected the attack even with an unknown signature.

Current security solutions are most useful in situations involving identified attack vectors. However, they fail against “zero day” attacks,  especially in high distributed environments, the attack “signatures” are unknown, and the solutions aren’t scalable enough.

Thanks to self-learning, machine learning detects anomalous behavior by analyzing massive amounts of data in multiple formats, relying on behavioral analysis and improving self-knowledge and understanding of phenomena.

AI has already started making its presence known in the cyber security space. Google is already using in spam detection, and Albatros cybersecurity incorporates machine learning in the aeronautics industry.

In the years to come, I see a great machine vs. machine war where even the hackers have adopted AI (anti-AI?), the stakes have gotten bigger and the cost of failure prohibitively high.