Originally published on Medium
Over the last few years AI has begun to dominate news cycles, and it has captured the attention — and the imagination — of entrepreneurs, investors, and consumers alike.
We see the potential: self-driving transportation on-demand, robotic assistants in the home, and Amazon Echo version 14.0 which will do things the human mind could never even contemplate. That future isn’t that far off—a decade or so, maybe.
But as much as we talk and read about AI, many of us still think about it in the wrong way. People overly compare artificial intelligence to human intelligence, and to see human intellect as the AI end goal. Human intelligence is familiar, and it is natural to want to use it as the bar, but here is the thing: if ten is the upper limit of intelligence, human intelligence must be closer to the bottom end of the scale.
To many, the goal of AI is to create technology that can think like humans — but it’s oversimplifying to suggest that any human or AI intelligence can be rated on such a simple scale as “better” or “worse.” Some people excel at memorization, logic, or emotional IQ, yet others excel at the visual or auditory. Similarly, an AI may have strengths and weaknesses. Furthermore, why have a goal of just matching human capability, when beating it is within reach - at least across some dimensions of intelligence!
Think about all the dimensions where AI seems to have already surpassed human intellect. Can anyone you know translate a passage into any one of 300 languages within a fraction of a second, or instantly determine the optimal driving route to avoid all traffic? On many tasks, specifically those that involve the processing of big data, machines have already learned to outperform us.
Don’t get me wrong. I am excited about the deeper AI which begins to mimic a human’s ability to learn purely by observing and interacting with the world just as a child can learn a new language simply by being immersed in the appropriate environment. This is known as Artificial General Intelligence or AGI and it requires no training data beyond direct “human” experience.
Certainly the movie industry is intrigued by AGI - machines taking human form complete with five full senses and the capability to comprehend and communicate with the world. It is surreal and fun to imagine coexisting with machines that are indistinguishable from humans in form and in intellect, but it is not a useful benchmark for understanding the current environment and trajectory of AI, and how AI will actually impact most products and industries. AI technology should not necessarily be judged by how human it is.
On the contrary, AI’s largest impacts over the next ten years are likely to be in the realm of domain-specific use cases. To achieve this, AI does need data and lots of it. These new awe-inspiring forms of intelligence are born from speedy algorithms that can process increasingly massive amounts of data.
We are on the brink of entire industries being disrupted by domain-focused AI and data-driven software. For example, at Applied Semantics and later Google, we built machine learning systems to choose the best ad out of a pool of millions, all in milliseconds. Each time an ad is served and not clicked, it is an extra data point to train the AI— a small opportunity for the system to learn, and more importantly, make new conclusions about the world. With trillions of data points, the systems become eerily effective, certainly far beyond anything within human capability.
We are witnessing an exponentially, increasing demand for data. Nearly every industry and aspect of business has moved, or is moving towards a digital reinvention, from brick-and-mortar shopping to e-commerce, from tv ads to mobile marketing, from cash to crypto, and so on. These new normals require software, AI, and data—massive amounts of data.
That’s exactly why I founded a data company in Factual — to provide the highest-quality location data to power digital innovation, including AI. We have an incredible opportunity to help businesses develop new products, acquire customers, and understand patterns of use within a real-world context. In order to build an engine for producing such data, we’ve had to build our own AI, which is in turn fed by even more data from our partners – a terrific feedback loop.
The effectiveness of our proprietary AI can’t be easily compared to a human scale because we the capabilities lie within a different dimension of intelligence such as processing trillions of data points to derive meaning. Many of the most promising applications of AI aren’t the ones that seem most like us, but rather, the ones that can do things we never even contemplated.