AI is affecting every sector, and almost every business is trying to create value out of it.
However, significant restrictions are keeping down the acknowledgment of AI’s actual potential.
Second Wave of AI which is also known as “Narrow AI” Started in the early 2010s, vast
amounts of training data combined with massive computational power revived the 30-year-old
neural network algorithms that lead to ‘Deep Learning’ whose performance grew as compared
to traditional approaches in several domains — particularly in image and speech recognition, as
well as most other categorization tasks. As the name suggests, these systems have proven to
be beneficial only in narrow domains. The system fails when applied to different fields. For
Example: If you took the software from a self-driving car and put it in a golf cart, it would be
useless without retraining. Whereas, any human who had learned to drive a car could drive a
golf cart for the first time and can quickly adapt to driving conditions. This, of course, is because
humans are great at transfer learning and apply them in different contexts and situations.
Current Second wave AI systems based on machine learning are unable to do this. It is not
about just transfer learning but also how we humans also apply common sense and
subconscious thinking in everything we do. While driving when we humans merge on the
highway, shift lanes, or come to a crossing, we need to predict the actions of other cars,
negotiate with them, and consider different possibilities to make a safe decision. As a human
driver, “we’ll consciously think about those types of possible parallel futures,” “That is enabled
by our common sense engine in the brain that gives us the ability to handle some new situations
that we never faced before.” This is currently not possible in self-driving cars built on deep
learning.

The vast majority of the AI community agrees that current second AI technology is nowhere
near human intelligence even dumb than one-year-old ones in terms of general cognitive ability.
Today’s second Wave AI is very poor at learning interactively on-the-fly, adapting to new
circumstances, abstraction and transfer learning, reasoning, and language understanding.
Some even believe Deep Learning is less capable than some First Wave (GOFAI) approaches
when it comes to specific tasks and will not get us to AGI.

DL pioneer and Turing award winner, Geoffrey Hinton: “My view is thrown it all away and start
again.”
Demis Hassabis, founder of Google’s DeepMind: “contemporary AI programs are… not that
intelligent.”

To overcome these limitations and reach a true potential of AI, The Pentagon is launching a
new artificial intelligence initiative it calls ‘AI Next’ which aims to improve the relationship
between machines and humans. As part of the multi-year initiative, DARPA is set to invest
more than $2bn in the program. DARPA says AI Next will accelerate “the Third Wave” which
enables AI to adapt to changing situations.

Dr. Steven Walker, Director of DARPA, said: “Today, machines lack contextual reasoning capabilities, and their training must cover every eventuality – which is not only costly – but ultimately impossible. We want to explore how
machines can acquire human-like communication and reasoning capabilities, with the ability to
recognize new situations and adapt to them.

Third-wave AI systems will lead to dramatic improvements. This next Wave of AI will also be
capable of learning in a way that is much more similar to how we humans learn from descriptive
and contextual models. Instead of relying on mathematical formulas derived from millions of
image pixels, it will be given a model that describes the physical features of a cat, e.g., “four
legs, pointy ears, claws, has fur.” Then through probabilistic reasoning, it can identify an object
like a cat, regardless of whether that object is presented via image, text, or voice. This is more
or less how we humans identify objects.

Not only will this reduce the need for large data sets, but it will also address the problem of
adversarial attacks through unfettered training data. A famous example of this was Microsoft’s
Tay bot, a chatbot designed to learn how to converse with Twitter users by digesting their
tweets but soon went horribly wrong and racist after users intentionally taught it to post racist
and offensive tweets. The third Wave AI will understand the context and meaning of words, not
just perform a blind statistical evaluation, and therefore won’t be susceptible to adversarial
attacks.

Founder , Digibeings
A software engineer and filmmaker working on bias-free purposeful technology. Always preferring ethics over the business. Founded a fact-checking company Metafact in Ireland and currently working on the next user interface to humanize our interactions with the technology.
We are a team of researchers, developers, CGI experts and artists working on emotionally responsive digital/virtual humans with personality and character that allow machines to talk to us face-to-face both verbally and non-verbally. In simple terms just imagine an Alexa/Siri with a face, that’s the product we are building.
×
Founder , Digibeings
A software engineer and filmmaker working on bias-free purposeful technology. Always preferring ethics over the business. Founded a fact-checking company Metafact in Ireland and currently working on the next user interface to humanize our interactions with the technology.