Artificial intelligence has emerged as the hottest buzzword in tech‚ despite the fact that it’s been around since the dawn of computing. Countless startups are now grabbing onto AI to explain what they do and tech marketers are branding with AI to make simple things like algorithms and basic machine learning sound a lot smarter and more sophisticated.
“AI is a ridiculously broad umbrella these days,” said Michael Karasick, IBM Research’s vice president of cognitive computing (IBM’s fancy terms for AI).
Karasick gave a presentation at IBM InterConnect 2017 this week in Las Vegas where he laid out IBM Research’s roadmap for AI. The approach of Karasick’s team is ridiculously practical, since their mandate is to incubate technologies that could be useful to businesses. As you’d expect, a lot of things they’re working on boil down to automation and big data.
“The reason we use machine learning in these problems is because there’s too much data,” said Karasick, whose team at IBM Research contains a mashup of mathematicians and systems analysts. The team uses AI for three types of things:
- Develop industrial strength solutions
- Make more efficient use of people
- Improve time-to-value
SEE: Special report: How to implement AI and machine learning (free PDF)
Karasick’s IBM InterConnect session “Looking Ahead: The Future of Artificial Intelligence” offered a window into the AI projects IBM is already working on. Here’s a quick summary of 10 of them.
1. Understanding PDFs
Many of IBM Watson‘s key functions are powered by processing large swaths of knowledge in various fields, from medical research to cooking meals. A lot of industry knowledge is locked away in unstructured PDFs, so if IBM can teach AI and machine learning ways to organize, process, and assimilate that information then it will accelerate almost all areas of its work in AI.
2. Understanding obligations
Businesses, governments, organizations of all sizes, and even just project teams are constantly trying to accomplish their work within certain rules and parameters. When those parameters change, it often moves the goal line. IBM wants to use AI to quickly recognize and flag changing rules, regulations, laws, and requirements.
3. Image captioning
By using machine learning to understand, evaluate, and categorize the content images, AI can unlock a lot of value and make valuable connections between visual data sets. Today, much of this work is done by hand by very low-wage data labelers.
4. Automatically building movie trailers
Taking the concept of image content analysis even further, IBM has already shown that its AI can pair video imagery with natural language processing to create trailers for movies. While Hollywood will probably continue to rely on digital artists, companies that are strapped for design resources could use this to create previews of their video content.
5. Cognitive assistant for data scientists
Data scientists are one of the most in-demand jobs in tech and the shortage of them puts future innovation at risk. The ones we have today spend too much of their time sorting and organizing data before they can evaluate it and make important connections. AI is needed for large scale data cleansing and to enable natural language searching for data searching using Watson Conversations.
SEE: Why Bill Gates and others think the robot that steals your job should pay taxes
6. Radiologist’s assistant
IBM Watson Health is already proving itself valuable as a tool for processing journal articles and the latest research and then acting as a diagnostic assistant. Now, IBM is teaching Watson how to read medical imagery to speed up the work of radiologists and help reduce errors.
7. Operational research
Doing traditional research into operational performance can often take 3 months or more, and involves complicated analytics models to ensure minimal errors. With AI, it can be done in less than four weeks and with superior accuracy.
8. Conversing without deep instance knowledge
IBM is actively developing chatbot technology using its strengths in AI and natural language processing. It’s looking to develop systems that can function with the kind of “deep instance knowledge” that Watson can get by processing large data sets on a subject area, as well as chatbots that can function without that kind of deep understanding.
9. Cognitive software DevOps
IBM also believes AI can empower DevOps–and not just the software lifecycle that we usually associate with DevOps. IBM also sees it having an impact on “cognitive UX” and machine modeling (a cousin of machine learning).
10. Scaling deep learning
Perhaps the biggest umbrella for what IBM is doing with AI is using it to scale deep learning. IBM sees the explosion of unstructured data as the catalyst behind machine learning, which takes a subset of data, analyzes it deeply, and then uses it to help extract value from the rest of the data set. “Deep learning is a way to specify machine learning earlier,” said Karasick. And scaling our ability to do that will increase the connections that can be made and the speed at which we can innovate.
