It seems as if you can’t pull up a website on data these days without running into the term Artificial Intelligence. And with good reason: Markets and Markets has it pegged as reaching $16 billion by 2022. Wander around the Web and you can hit on stories about everything from the top AI stocks to buy to what universities are offering MBA courses on managing Artificial Intelligence applications and algorithms.
There’s more action afoot to help developers deal with the surging AI trend, too. Amazon, for example, recently announced new self-service tools to help the development community more easily build Artificial Intelligence capabilities into apps, and Google has a new AI unit that has democratizing AI as part of its mission.
Artificial Intelligence is new terrain for many developers, and getting on board in some way, shape or form must be appealing when they hear about the latest moves in the space, including how they’re targeting expert talent in the area in a significant way. That’s a big signal when it comes to considering their own careers. Google, for example, hired the Director of Stanford University’s Artificial Intelligence lab, Fei-Fei Li, to lead its new AI unit. In an article on the topic, the Wall Street Journal quotes Andrew Moore, Dean of Carnegie Mellon University’s School of Computer Science, saying that students in Artificial Intelligence are “worth somewhere between $5 million and $10 million to a company’s bottom line.”
Other companies are putting their own stamps on democratizing AI, as well. That includes startup Bonsai, whose tagline reads: AI for Everyone. “We are working to make AI and Machine Learning technology more accessible to people who don’t have that expertise,” says Mark Hammond, Bonsai founder and CEO. Which is pretty much everybody, since there’s only about a handful of people around who are well and truly versed Machine Learning and AI practitioners, he says.
That means most companies are going to need to turn to the development talent they have on hand to build sophisticated Artificial Intelligence and Machine Learning solutions, so any help in getting individuals up to speed should be a plus when it comes to making plain their value to the organization. “We aim to make it possible for [companies] to do that, by leveraging [that talent’s] subject matter and domain expertise,” Hammond says.
He adds that’s what’s going on today feels a lot like 1995, when everyone suddenly realized they needed a website. “Everyone realizes now that they need to add intelligence to their hardware and software applications,” he says. Most companies are still figuring out what that means, but the key is that when you add intelligence beyond the prescriptive guidance that traditional programming allows, a whole world of possibilities opens up.
He points to users who’ve come to Bonsai with highly unusual potential AI use cases – everything from optimizing licorice extrusion to identifying missing Lego pieces in rented-out sets (along with ones that are becoming more “mainstream,” like autonomous vehicles).
AI with Bonsai
Hammond says there are three broad approaches to building Artificial Intelligence solutions today that are trying to make it easier for developers to get a grasp on things. There are the toolkits like Google Tensorflow that collect the algorithms a developer might want to use and put them in a standardized package with a consistent interface. But they still require some expertise to optimally use them when it comes to knowing what algorithms to deploy at any time and what parameters to use to configure them. Then there are statistical analysis packages, including black boxes where users basically provide the data they have, indicate which parts of that data should be predicted by which other parts of the data, and press a button to have a predictive model built for them. It’s easy to use but hard to refine efforts, though, since developers don’t have a lot of control.
And there are the APIs, most famously those associated with IBM Watson. Developers get a specific slice of functionality via the API with all the AI details hidden beneath that. That works great if the slice of functionality aligns with what a business wants to do – perhaps conduct sentiment analysis on a body of text – “but there are no licorice extrusion APIs,” he says.
Bonsai takes a different approach to building Artificial Intelligence solutions and it revolves around the idea of increasing the level of abstraction for developers. To imagine what that means, think of life for developers before databases. Those building a sophisticated application to work with data in a complex way had to worry about a lot of low-level management details, like which tree structure to use to store data. That took time away from their thinking about the business problems they were trying to solve with the new app.
“Databases solved this problem by moving the level of abstraction up,” he says. Database creators identified that they could give users a special-purpose program to codify aspects of their data at a business level and the questions they wanted to ask, and then have them turn that code over to the database engine to generate the underlying management pieces for it.
“We do that for AI instead of for data,” Hammond says, using its special purpose programming language, Inkling, which represents Artificial Intelligence in terms of what developers want to teach instead of the low level mechanics of how it is learned. Inkling programs feed into the Bonsai AI Engine to generate and train appropriate models to add intelligence into an application or system. “By shifting the level of abstraction up, developers can think about the subject matter expertise they wish to impart to the systems they are trying to build,” he says.
For instance, in the insurance industry, a company may have great expertise and institutional knowledge around actuarial tasks for computing insurance tables, but it may prove difficult to codify that via traditional programming. “We shine there,” he says. “where you really want to take the intelligence you already have and have a way to codify what that is and how it can be taught.”
Focus Now is on Control and Optimization Problems
As an example, Bonsai works with one company, Siemens that does a lot of motion control applications for industrial robotics. Their expertise lies there, not with Machine Learning or AI.
“They engage with us to leverage their expertise with that knowledge and codify what it is they are trying to teach the robots to do, and not have to worry about the low-level details of getting a data science expert who knows a lot about deep learning to make that happen,” Hammond says.
Right now, Bonsai mostly has pilot engagements underway, its beta having just launched at the end of September. Its main focus at the moment is on helping enterprise developers leverage Artificial Intelligence for control and optimization problems that use simulation-based workloads, for industries like industrial automation, robotics, automotive, aerospace, and energy. Longer term it is building out a general purpose platform and development community around its efforts, to enable developers to build all sorts of AI-infused applications with less complexity.
“We’re really trying to help educate people and we want to make sure developers have access to tools and training materials to learn new ways of thinking about programming,” Hammond says. “It’s fun to engage with developers and see that light-bulb moment where they get the different regime for how you build these applications. Then it’s off to the races imagining what they can do.”