Definitions and Predictions for Machine Intelligence
The term Artificial Intelligence in 2017 feels a bit like cloud computing back in 2010 – it’s the hot buzz phrase of the moment and it’s being broadly over-applied and mis-applied in industry.
I find that I can get asked about artificial intelligence, machine learning, cognitive computing, machine intelligence and advanced analytics all in the same meeting – and folks can end up using many of these terms almost interchangeably. Sometimes I want to carry around an AI buzzword bingo card.
Over the recent holiday, several of us were brainstorming trying to provide some more definition to all of this.
I’ve seen a definition of machine intelligence from Gartner Group that I very much like.
Paraphrasing a bit, the folks at Gartner define machine intelligence as “Special-purpose smart machines or technologies that appear to perform some tasks previously thought of as the domain of only humans, that are capable of autonomous operation, that learn from experience and that appear as though they understand what they have learned. These machines operate on big data, leverage machine learning and make probabilistic predictions of future states”
In our discussion, we first wanted to start with a goal – and we think the goal of machine intelligence is to leverage big data to build applications which solve high value business problems and which augment human capabilities, automate tasks that used to be human-only and even solve problems that were formerly outside the reach of humans.
I like this idea of starting with goals, because it starts to differentiate machine intelligence from general purpose consumer interest AI – which also includes self-driving cars, winning Jeopardy and personal assistants. I feel like this adds a bit more richness to Gartner’s definition by being explicit about applications working together with humans, as well as applications surpassing human capabilities. I also like putting the focus on solving real high-value enterprise business problems.
Then we went on to craft a definition of machine intelligence, which we characterized as the ability of the system or application to:
- Discover information that is latent in big and/or high dimensional data
- Learn and Predict the future with high accuracy as well as improve over time with experience
- Act based on data, recommend actions to humans, and react to changes
- Justify actions, predictions and discoveries in a transparent and statistically proven fashion
This definition again puts more meat on the bones that Gartner started with. To me this seems like an effective filter for defining whether a system or application is demonstrating intelligence.
This was also a helpful discussion at Ayasdi for thinking about how the AI market will evolve, and how machine intelligence platforms will become part of the enterprise ecosystem.
We clearly see Ayasdi as an enterprise-grade machine intelligence platform that enables the rapid creation of intelligent business applications.
This is why Ayasdi often layers on top of existing big data infrastructure, data warehouses, enterprise systems etc – and that the focus of the platform is more on very quickly building new intelligent applications (and less on enterprise developers manually stringing together machine learning APIs, or on data scientists making new discoveries using data science toolkits). The key is the focus on solving real-world business problems with intelligent applications. Hence the need for the definition of the goals of those applications, as well as what intelligence means.
For my 2017 prediction, I think that you will increasingly see this kind of thinking – that we will see both packaged intelligent applications that solve specfic business problems using smart software and big data, and we’ll also see AI-based application development platforms that are used to build these apps, and that also become an important way to get real value out of big-data investments.
What do you think?