Re-posted from Wired.
Artificial Intelligence (AI) has been enjoying a major resurgence in recent months and for some seasoned professionals, who have been in the AI industry since the 1980s, it feels like déjà vu all over again.
AI, being a loosely defined collection of techniques inspired by natural intelligence, does have a mystic aspect to it. After all, we do culturally assign positive value to all things smart, and so we naturally expect any system imbued with AI to be good, or it is not AI. When AI works, it is only doing what it is supposed to do, no matter how complex an algorithm being used to enable it, but when it fails to work–even if what was asked of it is impractical or out of scope—it is often not considered intelligent anymore. Just think of your personal assistant.
For these reasons, AI has typically gone through cycles of promise, leading to investment, and then under-delivery, due to the expectation problem noted above, which has inevitably led to a tapering off of the funding.
This time, however, the scale and scope of this surge in attention to AI is much larger than before. During the latter half of 2014, there was an injection of nearly half a billion dollars into the AI industry.
What are the drivers behind this?
For starters, the infrastructure speed, availability, and sheer scale has enabled bolder algorithms to tackle more ambitious problems. Not only is the hardware faster, sometimes augmented by specialized arrays of processors (e.g., GPUs), it is also available in the shape of cloud services. What used to be run in specialized labs with access to super computers can now be deployed to the cloud at a fraction of the cost and much more easily. This has democratized access to the necessary hardware platforms to run AI, enabling a proliferation of start-ups.
Furthermore, new emerging open source technologies, such as Hadoop, allow speedier development of scaled AI technologies applied to large and distributed data sets.
A combination of other events has helped AI gain the critical-mass necessary for it to become the center of attention for technology investment. Larger players are investing heavily in various AI technologies. These investments go beyond simple R&D extensions of existing products, and are often quite strategic in nature. Take for example, IBM’s scale of investment in Watson, or Google’s investment in driverless cars, Deep Learning (i.e., DeepMind), and even Quantum Computing, which promises to significantly improve on efficiency of machine learning algorithms.
On top of this, there’s a more wide scale awareness of AI in the general population, thanks in no small part to the advent and relative success of natural language mobile personal assistants. Incidentally, the fact that Siri can be funny sometimes, which ironically is technically relatively simple to implement, does add to the impression that it is truly intelligent.
But there’s more substance to this resurgence than the impression of intelligence that Siri’s jocularity gives its users. The recent advances in Machine Learning are truly groundbreaking. Artificial Neural Networks (deep learning computer systems that mimic the human brain) are now scaled to several tens of hidden layer nodes, increasing their abstraction power. They can be trained on tens of thousands of cores, speeding up the process of developing generalizing learning models. Other mainstream classification approaches, such as Random Forest classification, have been scaled to run on very large numbers of compute nodes, enabling the tackling of ever more ambitious problems on larger and larger data-sets (e.g., Wise.io).
Big Data is, of course, another driver of the recent investment interest in AI companies. Cheap storage, sometimes in the cloud, paired with the intuition by all manners of industry that collecting every piece of data possible will someday come in handy, has brought about a high demand for solutions that go beyond simple statistical analysis of data, and promise new insights and intelligence. The AI industry to Big Data is as petrochemical industry was to crude oil. We have the promise of doing more with the Big-Data crude than to simply burn it.
Most recently, confidence in AI has risen to the point where hedge funds, traditionally weary of black-box approaches to trading, are also starting to explore the use of machine learning (e.g., Bridgewater).
The financial boost from the recent investment in AI has led to a rapid expansion of the AI industry. More companies are looking to provide smarter solutions for their customers, and an explosion of new AI related companies that are looking to provide these solutions are emerging. Most industries that don’t want to be left behind are looking to employ AI in some form or other. The impact on industry has been broken down many different ways, but perhaps one of the best AI market landscapes I’ve seen is from Shivon Zilis, an investor at Bloomberg Beta. This is yet more evidence supporting the fact that AI is being applied in practically every industry possible, from Finance to Medicine, from Automotive to Oil and Gas.
Broadly speaking, AI companies fall into the following categories:
- Platform companies, providing general-purpose AI APIs to practitioners (e.g., Nuance, PredictionIO, Wise.io)
- Enterprise start-ups, bringing a combination of their core technologies and professional services customization to the wider enterprise, following a model not dissimilar to SAP’s (e.g., Skymind, Predii).
- Product companies focused on specific vertical applications of AI (e.g., Euclid Analytics, HoneyComb, Judicata)
It is still early days for assessing the impact these players are having on their respective industries and hard to measure their success in the various application areas they have focused upon. In some cases, the investments have been more on the promise with little actual proof.
Are we in another AI hype-cycle? Perhaps. In many cases, the term AI is certainly being conflated and sometimes confused with techniques that have never been thought of as AI being recast to be able to ride-out the buzz-wave. But the breakthroughs have also been coming at a fast and furious pace, pointing to the fact that there is ample room for innovation yet to be explored.
What can we realistically expect from AI in the next two to three years? Some of the more promising areas in my mind are: better fraud detection, breakthroughs in medical diagnostics, more intelligent personal assistants, and superior browsing and discovery of products in online retail and commerce. Hype or not, AI is once again promising to be the next frontier for software innovation and application.
Babak Hodjat is the founder and chief scientist at Sentient Technologies