For the first time in history, the invention process is not entirely dependent on human intellect.
Throughout the ages, human innovation has been accelerating at a mind-boggling rate. Consider that 1 million years elapsed between the control of fire and the invention of the wheel, but just 5,400 more years until the creation of the Gutenberg press–and a mere 455 additional years before the development of the light bulb.
When plotted on a chart to illustrate the human innovation curve, it is clear that there is only one word that can describe the increase in the pace of progress: exponential. However, the innovation curve now is shifting into even higher gear with the proliferation of artificial intelligence (AI).
AI grew exponentially in 2017, with no signs of stopping in 2018. AI fundamentally changes the equation of innovation, adding a new variable that dramatically accelerates the rate of advancement. For the first time in history, the invention process is not entirely dependent on human intellect. Machines are now augmenting and will eventually supplant human brainpower.
Although AI is still in its early stages, the arrival of a new approach –called “conducted learning”–will expedite the rate of overcoming current limitations and is set to affect the speed of both AI and the human innovation curve.
AI algorithms now are reaching, and even exceeding, human capabilities in areas such as strategic game playing and image classification. However, these algorithms fall under the category of artificial narrow intelligence (ANI) since they are limited to excelling in narrowly defined tasks.
We can train an AI algorithm to recognize the shape of a gun, for example, and it will be able to detect the image faster and better than humans. However, due to this narrowness limitation, in a real-world application, such as in a TSA scanning, this effective scanning method will be restricted only to the specific gun models on which the algorithm was trained.
Consequently, we still have a way to go until we reach artificial general intelligence (AGI), which will be more akin to humans and present capabilities similar to what we see in sci-fi movies.
The maestro of AI
Conducted Learning presents a promising solution towards achieving AGI by leveraging the combined power of separate ANI engines. Conducted learning enables running several cognitive engines in concert, picking the best engine or engines to perform the task, similarly to an orchestral performance. This results in a more accurate outcome than what can be obtained from any single network, while cutting down on computational costs and speed.
Like other deep learning models, conducted learning initially formats data, preprocesses it, and generates input. The magic happens during the next stage, when the technology acts like the orchestra’s conductor, instructing each cognitive engine when to play its part in the complete composition.
The conducted learning model extracts the accurate parts of the output and recycles the remnant through a process of transformation and rerouting to the relevant engine or engines. By using multiple cognitive engines simultaneously, the algorithm continually learns and improves its capabilities, building a more effective topography to complete the task. This dramatically improves the accuracy and performance.
Conducted learning facilitates overcoming the limitations of narrowness through two methods: intra-class learning and interclass learning. Intra-class learning uses multiple cognitive engines in the same class (e.g. translation). Interclass learning employs several cognitive engines across different classes (e.g. translation and facial recognition).
Taking transcription as an example, with intra-class learning, conducted learning enables the transcription of a soccer match of the English League by first transcribing the words in the English language. It then fills in the gaps of running a transcription engine that is trained on sports terms. The next engine will cover words pronounced in a heavy British accent to catch unclear words that weren’t detected in high confidence, etc. All this is done in milliseconds.
With inter-class learning, if the auditory speech recognition engines cannot catch the accurate transcription of the names of the players, then visually trained engines, which use face recognition to match the player’s face or “read” their names off their t-shirts, will be activated to accurately complete the task.
Bending the curve
The implications of conducted learning go beyond achieving greater accuracy. It is a big leap for machines, but even more importantly, for AI’s ability to teach itself by one engine or class of engines informing the other about the data and completing the task.
The human innovation curve has now changed with the addition of AI to the calculation. AI will become the dominant factor, dramatically outpacing humans’ capability to invent on their own.
This article originally appeared on Inc.com