Next Phase of AI: Advanced General Intelligence (AGI)

Next Phase of AI: Advanced General Intelligence (AGI)

As one of the main tech trends in 2018, AI has now definitely entered businesses as well as private zones. Yet, most intelligent systems around us use what’s called “narrow intelligence” or “narrow AI”. AI’s next phase is about approximating human intelligence.

At the core, advanced narrow intelligence (ANI) in systems such as Alexa, Siri, Google Home and Netflix, are powered by a finite set of possibility-outcome combinations that can retrieve seemingly unlimited information. AGI, on the other hand, requires machines to be at least as intelligent as humans with traits such as metacognition, deep comprehension, intuition, abstract reasoning, real-time learning, etc. Moreover, it requires the machine to take a human-like understanding of the implications of actions.

Autonomous driving cars are a great example of ANI – trillions of outcomes to be calculated in nanoseconds, with human life in the balance. But imagine if your car had a better understanding of you. Say she understood that you were in a financial crunch and “thought” about how she could help by offering ridesharing during the times when you weren’t using her. That’s an example of AGI: a comprehensive, intuitive solution that wasn’t coded into her at “birth.”

Across enterprises, there’s strong interest in adopting “intelligent,” customer-centric solutions to reduce cost pressure while enhancing customer service and the ever-looming threat posed by disruptors. The industry is flooded with intelligent point solutions. But how do we make the jump from narrow, niche intelligence to AGI awesomeness? How do we develop algorithms that encompass pragmatism and compassion, and avoid unconscious biases?

The possibilities of AGI
To make the business case for AGI R&D, let’s look at what AGI could potentially bring to a business or our lives as a whole. Take the case of inventory management. A simple use case would be a program to reorder items if the amount in stock drops below “X.”

  • What if we could add a layer of analytics to it, built with ANI? If the item were a seasonal item, like, say, “turkeys,” the system could look at the trend and bulk-order more turkeys in early November, in preparation for Thanksgiving.
  • In a connected world, the intelligent system could also have the farmer raise more (or fewer) turkeys based on population variation and social media trend analysis on how many people are planning to celebrate Thanksgiving. We could go even further and have a farmer adjust how much seed is planted based on the quantity of turkeys that would need to be fed.
  • On the consumer side, robots could prep and ship turkeys to automated warehouses. There, robots could load autonomous trucks that would deliver the turkeys to our houses, where an intelligent “Roomba” could receive the package, ask Alexa for a recipe, roast the turkey, and have it carved and ready to eat when you got home. This scenario is not impossible with today’s technology and some interconnected ANI point solutions.
  • Now imagine a world with AGI. Or can we? What if the AGI analyzed Thanksgiving and developed a solution to combat the impact of meat eating on climate change? What if the AGI could create meat substitutes that truly substituted meat? What if the AGI could clearly articulate the impact of overeating and transform the human race into frugivorous beings? On the other hand, what if the AGI encouraged us to eat ourselves into non-existence? The potential of what a super-intelligent being may develop is limitless.

To go from interconnected point solutions to AGI requires an extremely high level of computational creativity. And here’s the kicker: How do we define goals clearly enough so that strategic solutions devised by a super-intelligent neural network will balance both capitalistic gain and humanity? So far, there’s no good answer to those questions.

The wide gap between ANI and AGI
One popular suggestion for moving from ANI to AGI, is to imagine these point solutions as individual neurons, like in our brains, and then connecting them to form a neural network, similar to a biological nervous system. The result would be disparate point systems sharing data on customer behavior and industry trends, with an overarching algorithm that makes decisions on how to interpret the data, plan strategy and set goals – in other words, individual ANI systems working together for a greater goal. It’s a great idea, but there are very few (if any) proofs of concept.

A huge AGI stumbling block is that it isn’t as commercially viable as ANI. Institutions across domains are hard-pressed to deliver great customer service, lower cost, faster turnaround – all of which can be met through point ANI solutions. Strategists can define a dollar value to these advances: X seconds saved here translates to Y cents saved per transaction, for example, or knowledge of this trend allows prediction of a product choice, which increases click-through by Z%.

These are tangible, immediate gains. It could take decades, on the other hand, to build an AGI system that could replace the strategists and consultants that an enterprise currently employs. How many capitalists are willing to invest in a “tomorrow” that’s so far away?

Innovation, magnanimity, creativity – these are AGI dreams. For now, the way forward is continued reliance on humans to make the required connections – and encompass the requisite intelligence, empathy and reasoning – to bridge the divide that will take businesses from ANI to AGI.

Related Publication

Applied AI: Beyond Science Fiction to Business Fact

As the digital data that surrounds us grows exponentially, it will power advanced forms of artificial intelligence that, over time, will augment human capabilities to make us smarter, more productive and more effective in our personal and professional lives.