Third Wave AI: The Coming Revolution in Artificial Intelligence
(Republished with permission from: Six Kin Development)
Artificial intelligence is everywhere, fueling a multi-billion dollar industry and radically changing how businesses operate and how people work and play. Yet as powerful as it is today, major limitations are holding back the realization of AI’s true capabilities.
Current commercial AI technology falls squarely in the “narrow AI” category, i.e. highly specialized systems that are very good at specific, well-defined tasks…and nothing else. Even autonomous vehicles, as impressive as they are, utilize a composite of narrow AI systems. If you took the software from a self-driving car and put it in a golf cart, it would be useless without reprogramming. By contrast, any human who had learned to drive a car could get in a golf cart for the first time and have no problem navigating the fairways. This of course is because humans are very good at abstraction — we can easily generalize solutions and apply them to similar but different problems. Contemporary AI systems are unable do this.
Another limitation of today’s AI is its dependence on huge training data sets. A typical machine learning algorithm for recognizing cats, for example, must be fed tens of thousands of cat pictures before it reaches an acceptable level of accuracy, something a three year old child can achieve after seeing only a few examples.
To overcome these and other limitations and reach its true potential, artificial intelligence must become more humanlike in several aspects. Enter Third Wave AI.
The “three waves of AI,” as described by DARPA’s John Launchbury, refers to the state of artificial intelligence capabilities past, present and future. In the following illustration, I’ve elaborated upon this and added a fourth stage to account for what many believe will be the ultimate phase of AI: human-level intelligence and beyond.
We are currently experiencing the Second Wave of AI, dominated by deep learning and statistical, “big data” approaches to AI. But just around the corner, perhaps sooner than many people realize, is the Third Wave.
Third Wave AI systems will feature dramatic improvements, most notably in their ability for contextual adaptation. They will understand context and meaning, and be able to adapt accordingly. Third Wave AI will not only recognize the cat, but will be able to explain why it’s a cat and how it arrived at that conclusion — a giant leap from today’s “black box” systems.
This will allow the next generation of AI to overcome the brittle nature of today’s machine learning systems, which function well on the majority of cases but can fail spectacularly when presented with a case that doesn’t fit its training model. A tragic example of this is the recent fatality of a pedestrian in Tempe, Arizona who was struck by an Uber self-driving car after the on-board AI failed to identify the pedestrian in time to take preventive action.
This next phase of AI will also be capable of learning in a way that is much more similar to how humans learn. Instead of being fed enormous sets of labeled training data, Third Wave AI systems will learn from descriptive, contextual models. Instead of deriving a mathematical formula from millions of image pixels, it will be given a model that describes the features of a cat, e.g. “has fur, four legs, whiskers, pointy ears, claws.” Then through probabilistic abduction, it can identify an object as a cat, regardless of whether that object is conveyed via image, text, or voice. This is more or less how humans recognize objects — “If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.”
Not only will this reduce the dependency on large data sets, it will also address the problem of skewed training data. A notorious example of this was Microsoft’s Tay, a Twitter based AI chatbot that went horribly astray in 2016 after members of the public intentionally taught it to post inflammatory and offensive tweets. Third Wave AI will understand the context and meaning of words, not just perform blind statistical evaluation, and therefore shouldn’t be so easily swayed by bogus information.
It does not take much imagination to envision the tremendous possibilities of Third Wave AI. Personal AI assistants that can converse in natural language and understand what you tell them, AI research scientists that learn independently and work tirelessly 24 hours a day, intelligent tutoring systems that can teach as effectively as a human tutor, domestic robots that can finally do far more than just vacuum the floor, and so forth. This all may sound like futuristic science fiction, but in fact many companies around the world are already working hard to create next generation artificial intelligence solutions. To highlight a few examples:
Pandai
My company’s own flagship project, Pandai, is an advanced artificially intelligent personal teacher with humanlike abilities. Pandai will be able to interact with students in natural language, observe and learn about the student, and tailor the instruction to each individual. Pandai aims to substantially reduce time spent with private tuition, and make quality education available in remote and developing areas.
Aigo
Developed by long time AGI researcher Peter Voss, Aigo is a Third Wave AI powered personal assistant that endeavors to greatly improve upon Second Wave AI assistants like Siri and Alexa. Aigo will understand context, be able to remember what you told it previously, learn new facts across different domains, and interact with other people in your network.
SingularityNET / OpenCog
SingularityNET is a non-profit foundation that’s creating an open source, decentralized network for consuming and exchanging AI services. It will provide a blockchain based platform and set of APIs to allow different AI components to interoperate autonomously for the first time. OpenCog is an open source cognitive framework that aspires to ultimately create AGI (Fourth Wave AI), but in an incremental approach, several building blocks of OpenCog, including algorithms for learning/reasoning and natural language processing, are being seeded on SingularityNET to enable Third Wave AI capabilities.
While the ultimate goal of true human-level thinking machines may still be a long way off, Third Wave AI is very much within reach. There is a lot of work to do, but I’m confident we’ll see the first applications of this revolutionary technology within just a few years.