Watson as a service: IBM preps artificial intelligence in the cloud, and e-discovery asks “can we have a bite?”

By:  Gregory P. Bufithis, Esq.   Founder/CEO

19 November 2013 – The biggest news last week was probably the IBM announcement that it was offering Watson’s machine-learning system in the cloud, thereby expanding artificial intelligence’s (AI) frontiers and also monetizing the work as it happens.  More importantly, IBM’s Watson announcement is significant, not necessarily because of the sophistication of the Watson technology, but because of IBM’s ability to successfully market the Watson concept.   I have just finished the new book Smart Machines: IBM’s Watson and the Era of Cognitive Systems by Steve Hamm (IBM Communications Strategist) and John Kelly (IBM Research Director) so the announcement was very well timed.

Watson is something I have closely watched through the last few years.  My first intellectual property client was IBM when I was a pup on Wall Street, back in the late 1970s, and I kept connections with many people at the company ever since.  As I have previously noted from a FutureMed conference I attended last year, Dr. Martin Kohn, chief medical scientist at IBM research, brought the crowd to tears of laughter when he said “You’ll not be surprised to learn that the executive leaders of IBM fairly quickly decided that playing Jeopardy! was not a long-term business model.”

IBM must have a lot of confidence in its newly revamped Watson machine-learning technology because by making it a cloud service by way of a service and an API set is not to be taken lightly.  It’s initially being offered to only a small, select group of IBM partners, which means IBM wants to build ties first and foremost with outfits that can program for Watson in a robust way. But even the sketchy hints that have leaked so far are revealing. And there will some more expanded programs in the coming months.  If IBM follows the models set out by the likes of everyone from Apple to HP, it could become a way for companies to monetize not just access to a specific data set, but a custom way of processing that data set. For example, machine translation: it ought to be possible for IBM to allow a company to sell apps for translating to or from specific languages, with each language pair being its own app and a discount for packs of multiple languages. I have addressed the advanced machine translation language issue before, especially as it affects the contract attorney market. And the obvious disruption to/impact on the e-discovery market will most likely soon be addressed by vendors and e-discovery pundits alike as I have noted below.

The biggest elephant in the room is Google, of course. What gives Google a possible edge is its unprecedented access to more raw information — and more kinds of it — than just about anyone else which allows it to execute innumerable data analytics.  And it has laid claim to the legal search market before IBM as I have noted in previous posts.  But I suspect that if Google has something under wraps right now to rival both Wolfram (who deserves a separate post) and IBM, it will start talking about it.

Go to any of these software developer mini-conferences on semantic technology, virtualization, media, etc. with Google, IBM, Microsoft, Thomson Reuters and others in attendance.   Search technology in the e-discovery/litigation game is a bit different than the search/semantic technology field I normally follow and am involved with in the web and media world. But the e-discovery folks are wrapped up in the same artificial intelligence, predictive coding, and non-linear vis-à-vis linear approaches as are everybody in search.  But of course the Big Dogs have $8 billion R&D budgets.

These various programs can integrate gigabytes, terabytes, or petabytes of unstructured data from web-based repositories, collect a wide range of unstructured web data stemming from user-defined seed URLs and extracts/enrich that data using unstructured information management architecture.    While its purpose is to leverage unstructured data strategies/support decisions (business intelligence) its application to text analytics for litigation is there.

Even at events like IBM’s Smarter Commerce at Monoco you stand in awe.  IBM has this analytics that identifies the “alpha” individual in its customers’ calling circles — those people who’d be most likely to take other customers with them should they switch carriers — so that it can cater to those individuals.  They also have text analytics they apply to all forms of social media to “detect sentiment” (not just chase keywords) in order to gauge customers’ reactions to brand advertising and then adjust marketing channels and messaging accordingly.  It’s industry-specific — healthcare, insurance, telecom, consumer goods, etc. – and they are in the process of applying it to the litigation world.

The future of search is to not search at all, as Stephen Arnold of Beyond Search is fond of saying. And although this may sound contrarian, if you follow his columns and those of writers in KMWorld you know we are on the threshold of search technology that will eliminate the need to explicitly ask for information.

AI is finally getting smart.  The search technology of tomorrow will be built on a truly intelligent system that can think, understand and make decisions on its own. While initial attempts at analytics included brute force techniques of feeding enormous data and rules to systems, improvements in machine learning algorithms (that learn on their own, simulating the working of the human brain) have revolutionized search significantly.

Its application to e-discovery and legal search is still to be determined but I note a few points Ralph Losey made waaaaaaay back in 2011 (seems like forever ago, yes?) about Watson and e-discovery … prescient Time Lord that he can sometimes be :

“The ideal e-discovery Watson computer must not only search and find, he must rank. Put the highest on top please. Watson may not be able to put the five you will use as the first five documents shown, but it is not too much to expect that the 7±2 will be in the top 5,000. The humans working with Watson will narrow them down, and the trial lawyers making the pitch will make the final selections.”

What Ralph was getting at is that 7±2 is the true goal of e-discovery, that you  should produce five to nine of the hot documents that the triers of fact can understand. If your search finds those magic seven, and no others, it is a great success, regardless of all of its other misses. If your search finds a million relevant documents, and attains a precision and recall rate of 99%, but misses the top seven key documents, it is a complete failure.  As he said:

“We have different needs. We must design our e-discovery to be reasonably calculated to lead to admissible evidence, which means non-cumulative.  This is what the trial lawyers need to tell their story of prosecution or defense.”

When Ralph noted “telling a story” I remembered a post by Ron Friedmann (from 2009, so another trip to the Way Back Machine) in which Ron said litigation is about story telling/context. It is to tell a story that wins the case.  When I called Ron about it he commented “perhaps Watson will help put together timelines and even alternative narratives from the mass of data it ingest. Sometimes we seem so focused on arithmetic clarity when our focus should be on narrative plenitude and maybe Watson can provide that”.

It’s the human element, really, and IBM uses the same terminology we use in e-discovery when we speak of harnessing human judgments to work with technology, and “Subject Matter Expert” which Ralph discusses in detail in his launch of Legal Search Science, an interdisciplinary field of study and practice concerning the search, review, and classification of large collections of electronic documents.

Because that is the key … the human “working with” Watson.  It is what IBM truly understands.  Over the past two years, at the Mobile World Congress events in Barcelona and at the IBM cognitive computing sessions in New York and Israel, I have had the opportunity to spend some time with members of the Watson team and several approaches standout:

1. Whatever the space, IBM does not do anything unless it can bring significant value, it can bring a transformational nature to eliminate the problem, deal with the issue better.
2. I don’t know if they are taking a page from Ray Kurzweil’s team over at Google or vis-a- versa, but IBM believes in a robust communication between arts, humanities, technology and law so there is a confluence of lawyers, doctors, technicians, BI people, laymen to consult so that their brainiac engineers understand human nature, get asked questions from left field.
3. IBM’s Watson is an evolved collection of technologies wrapped with some deep human expertise. If you’re trying to sell the message of deploying more advanced technologies within your organization, then “buying Watson” is going to be an easy sell.

Andy Wilson of Logikcull certainly understands this and sees a direct application of Watson to e-discovery for these same reasons.  (Andy serves as my Obi-Wan Kenobi in all matters tech). What he sees in the IBM announcement:

“What I think is interesting about Watson-as-a-Service (WaaS), is just that:  as-a-service. Similar to Jeopardy contestants, companies, even pure “predictive coding” companies, will not be able to compete with the level of AI sophistication that something like Watson can supply cheaply and quickly. However, predictive coding vendors can make their software stronger by integrating with WaaS. So if you think about how this powerful AI-as-a-service can be applied to eDiscovery the possibilities are unlimited. It’s kind of like crowdsourcing, isn’t it? 

The document analytic possibilities are pretty obvious. e.g. Wouldn’t it be cool to find out X about Y with just Z documents in just a few seconds? Simple question-answer features. But that requires the documents to leave your system, go to WaaS, crunch, then come back. Maybe not ideal.”

What’s more interesting to me is Logikcull’s trend analysis based on their own user generated content as an area where WaaS could help eDiscovery. As Andy said “predicting which reviewer will be the right fit for a project based on prior usage history + a hundred other data points about that user pulled in from around the web (LinkedIn, twitter, etc.) could be a very interesting way to use WaaS.”

Ralph Losey echoed some of the same thoughts on crowdsourcing (because great minds think alike) and over the weekend told me this:

“Does this represent a kind of crowd-sourcing strategy by IBM to develop new Watson applications? Put it out there and see who can come up with really good products/services. Buy up the best and don’t worry about the rest. Could be a good way to accelerate new product developments. Or could it be an abdication? There is a fine line between genius and madness. 

What intrigues me is this: will IBM make Watson the lawyer or will a team in a garage somewhere do it, with or without IBM? Or will it be one of the major existing legal software developers who make it happen? They know the law far better than IT-centric Big Blue. One way or another AI in the law is coming, fast and Big Time. The real hackers out there will be bold and fast and rush in to give Watson codes a try. Or will they? Is it really bold to just piggy back on IBM? Remember the Apple commercial? Is it really bold just to look to Watson? To let the big companies dominate and control? Will we all walk in lockstep? Or will the AI for tomorrow’s lawyers come from a more home-brew concoction?”

If you are a regular reader of Ralph’s columns you know this: his mantra is that it’s the methods that matter most, not just the code. GIGO … Garbage In, Garbage Out, an informal rule holding that the integrity of output is dependent on the integrity of input … may be mitigated by smart code, but it’s still there, still following the Second Law of Entropy. But to Ralph “maybe IBMers get that, maybe that’s why they are giving it away? Could it be they know their limitations?”

This past year has been a rush.  I was able to witness the IBM roll-out of Watson this past Spring for call center/customer service  clients … through this amazing interface titled “Ask Watson” … I saw it pull together/churn web chats, email, smartphone apps, SMS, catalogs, training manuals, product disclosures, terms and conditions, emails, customer forums, and call center logs, as well as publicly available feeds and reviews from places like Amazon, Yelp, etc. and generate answers … in seconds.

Yes, it’s the tech, stupid.  Watson pulls up stuff that a “mortal” cannot because it is looking for semantic links, not just doing text-matching based on keywords. Watson can understand natural-language questions — either spoken or written in English — and it responds with not just answers but evidence and confidence levels in those answers. Second, Watson can engage across channels, be it phones, tablets, websites, messaging or email. Third, Watson learns over time and serves customers better because it can personalize the customer experience.

And over the summer, when Watson was rolled out for human resources departments … at a special human resources technology conference … attendees saw first-hand how cognitive computing systems that are taught not programmed provide personalized, evidence-based, contextually relevant responses to questions posed in natural language.

I will have several more posts about Watson in the coming months, and will include much more in my e-book IBM: a culture of innovation and analytics, still a work-in-progress. I have left unaddressed some of the finer, nuanced issues of cognitive computing vis-a-vis artificial intelligence, a very specific – and more accurate – term. It’s not real intelligence. It’s artificial. Real intelligence is not just about being able to quickly recall some facts or identify patterns in information at blazingly fast speeds. It’s as much about understanding context, intuition and peripheral vision as it is about intelligently handling what data you already have.

And I have also left unaddressed the legal theories of categorization in artificial intelligence, information retrieval, data mining, and other computational fields critical to legal search processing.  Not to mention the expansion of laws as algorithms (converting simple laws to machine-readable code), or the potential use of Watson in actual statutory interpretation (which has advanced far beyond the Yale law review article in that link). IBM Research has put together a “Cognitive Computing” Playlist on Youtube which I think you’ll find enjoyable. Just click here.

But as the ability to capture and process large volumes of data easily and accurately increases … as Watson so ably attests to … its effect on the e-discovery ecosystem is obvious.

However, quite frankly, after my two years (and still going) in a neuroscience/informatics program I remain in awe of the  brain, its remarkable ability to comprehend language, conduct abstract reasoning, control movement — an amazing instrument of power, imagination and knowledge.  Yes, we are facing a wide open field of opportunity for the next generation of legal entrepreneurs. But it’s going to take some pretty amazing neuromorphic technology to beat the human brain.

Endnote:

Large datasets, multimodal approachs to legal review, machine learning, artificial intelligence, etc.  It all starts with a background in computer science. While much of it I learned on my own, I need to give an enormous hat tip to Professor Eric Grimson at MIT.  I found this excellent video on the MIT Open Source network (most of which is on Youtube).  It is an introduction to computer science and programming: what is computation, an introduction to data types, etc.  I hope you find it helpful: