Beyond Generative AI : Rise & Risk of Co-Intelligence (CI)
Generative AI : In recent past the most widely searched terminology over internet is "Generative AI" (GAI). With LLM (Large Language Models) based solutions coming to exposure, every individual got interested, as one can directly converse with an AI engine and get human like responses. Easy accessibility for general public to interact and test out an AI engine made it popular. However, today "Generative AI" is the most non-optimally used terminology (purposefully not saying "mis-used") in general.
With Generative AI (GAI) coming in, we should not undermine the capabilities of erstwhile AI as Generative AI cannot replace many of those purpose specific AI (Narrow AI) solutions. Generative AI (GAI) is specific to solutions which require content generation as conversation, speech, code generation, image description or creation of image from description and many such use cases. Since GAI is able to generate contents for any normal user without requiring to train with your own data (leaving out the LLM developers) , it has become quite popular. Content generation is exciting but how many of us validate the "Right" content being generated by the LLM models. In fact, if you keep asking the same question repeatedly, it gives different answers. This is governed by a phenomenon known as "Hallucination" where GAI generates content with incorrect information but presents as if true. It is caused due to multiple factors as training data, biases, algorithms and many more but the content generated is not only wrong but at times can be harmful. That is where a combination of Human Intelligence (HI) comes in to give birth to Co-Intelligence (CI) - (a concept explained in one of my earlier articles in 2017 Digital World and Artificial Intelligence (pchats2000.blogspot.com)
Potential of CI: With CI we can move the world from the realm of Narrow AI coupled with Generative AI to Artificial General Intelligence (AGI). AGI can handle multiple tasks of GAI and Narrow AI through one AI engine. CI should help to eliminate "Hallucination." It can be a workplace where human and machine work side by side in collaboration with each other and seamlessly communicating with each other. A use case can be an intelligent car understanding where its passenger wants to go the moment he/she steps inside the car.
Risk and Control Measures of CI: While we see there is lot of potential for CI, if not controlled properly it can lead to Super Intelligent AI, which can outperform human brain in every field. In this state a machine can outperform human and human will have no control over machines. But whether to take it to that level is dependent on us.
We will use an analogy here. We all know missiles are destructive devices but equally important for security of nations. Nations use it judiciously against enemy nations as and when required in a controlled manner. Imagine a situation, where missiles are getting manufactured, at the last station there is an auto ignition machine which can automatically ignite each missile getting produced. We can imagine the amount of destruction it can cause to our own individual nations if that simple machine of automation is put in place. But we have a control there that when required missiles are taken to respective areas and consciously ignited to use it against enemy.
Similarly with CI there has to be controls implemented to prevent such auto-ignition which can cause destruction. There needs to be guiderails set by AI consortia of nations, which should be acceptable across nations and regulatory and legal compliance guidelines should be published under legal frameworks of nations.
Development and advancement of technology within defined guidelines can help to leverage the power of Co-Intelligence (CI) and take the society to a completely different orbit of advancement.
Comments
Post a Comment