Last week I attended the O'Reilly conference : https://conferences.oreilly.com/artificial-intelligence/ai-ny
If you have read my tweets, than there should be little news here: Tweets about #OReillyAI from:mpjjonker
Put AI to work
Given the title of the conference, the theme of this blog (IBM and Google) and the fact that these two companies are strategic sponsors, I had to go to this AI conference.
I must say that this conference was both humbling:
there is so much I don't know and there is so much still to be figured out by the industry
as it was a mix of academic and business oriented presentations:
The top of the AI researchers were presenting and all big IT companies wanted to make clear that they are the best in this field.
To illustrate this last feeling:
Both Google, IBM and Microsoft announced MachineLearning capabilities for analyzing video content this week.
This leads to my overall conclusion
Although you can really do some cool stuff today with AI, be prepared to overcome some obstacles and don't be surprised if one of those obstacles is actually part of current state of art research.
And don't expect any portability, should you want to change supplier.
But things move quickly, 2011, the year of Watson, is not so long ago and see what we you already can do today !
Sometimes teaching is telling
Looking forward, these tools will let us explore the oldest and best idea for building human-like artificial intelligence: Build a machine that starts like a baby, and learns like a child.
It's the only known scaling path to intelligence that actually works. And only now are we in a position to deliver on it.
Make sure that the intelligent systems brought into the work place are able to communicate not just the answer but the reasoning and data that supported it.
For me the conference started on Tuesday with the 3 hour session:
Here and now: Bringing AI into the enterprise , given by Kristian Hammond (Narrative Science)
The session was a lot of fun (bashing spreadsheets) and the most important take away:
It is not so easy to find a good use case for IBM Watson, fortunately for me (and e-office) the use case for finding knowledge and expertise turned out to be the right one !
Scalable deep learning with the Microsoft Cognitive Toolkit, given by Anusua Trivedi (Microsoft) and Patrick Buehler (Microsoft).
Image recognition still is the "go-to" place for AI/Machine Learning. Same in this session, although there was some attention for text.
Too bad that the conference WIFI was too weak to support the actual hands on part. Especially the use of Jupyter Notebooks whould have made this worth while.
The following tweet
Was the result of that session.
The day starts with keynotes, the pace is high and one sponsor after the other gets some "airtime".
The keynote session got better session after session, healthcare cases (Suchi Saria (Johns Hopkins University)) are always inspiring, but David Ferrucci (Elemental Cognition) and Josh Tenenbaum (MIT) were really impressive and contributed largely to the overall conclusion stated in the begin of this blogitem.
After the keynote I attended the following sessions :
Beyond the state of the art in reading comprehension , by Jennifer Chu-Carroll (Elemental Cognition)
This was the extended version of David Ferrucci's keynote talk
The future of AI is now , by Bjorn Austraat (IBM)
Good recap of the Cognitive Cloud services of IBM (see other items in this blog) and the announcement of the video analytics service.
Again Kristian Hammond, this time attention to Natural Language Generation, let machine explain the structured (spreadsheet) data.
The session had a nice start with the comparison between steam engines and the current state of (enterprise) AI. The observation that we are in early stages with a lot of tailor made engineering and the lack of industry standards, made a lot of sense. Personally I don't share the problems (yet) that were described about performance and scaling.
AI-powered natural language understanding applications in the financial industry , byFrancisco Webber (Cortical.io)
A nice alternative way of machine learning by using "Semantic Folding", the use cases are similar to other NLP solutions. I am curious about the future of this way of working, especially since the "big guys" seem to have alternative technologies.
Already the last day, starting again with keynotes, a bit more relaxed pace this time.
The Google speakers stood out, both Doug Eck (artistic AI) and Peter Norvig (software engineering and AI).
We had to leave on time to get to the airport , so we missed the last two slots, these are the sessions we could attend:
This was (again) a more academic session that overlapped with Josh Tenenbaum's keynote session, again good to know what we don't know yet.
This session explained how Clara labs scheduling solution works in the back office, how they applied machine learning and what the part of the humans is. A good example for put AI to work.
As we speak a smaller language (Dutch) I though I would give this a try, but this session was about computer, edge equiment (robot) vision and how you can tackle the problems that arise when you don't have the real world data available. The use of vitual worlds was one way to solve this.
This session was a better match for us, dutch speaking people. Codruta told the story of the news industry in Norway and how they were able to apply AI (especially NLG or automated writing) to improve the customer experience and to actually get back some (paying !) subscribers.
It was a good thing we left early, the Fire at JFK Restaurant caused long waiting lines and a bit of chaos at the airport.
So an exciting ending to an exciting conference